deepbox 0.1.0

package/dist/chunk-PR647I7R.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/optim/index.ts","../src/optim/Optimizer.ts","../src/optim/_internal.ts","../src/optim/optimizers/adadelta.ts","../src/optim/optimizers/adagrad.ts","../src/optim/optimizers/adam.ts","../src/optim/optimizers/adamw.ts","../src/optim/optimizers/nadam.ts","../src/optim/optimizers/rmsprop.ts","../src/optim/optimizers/sgd.ts","../src/optim/schedulers.ts"],"names":["isRecord"],"mappings":";;;AAAA,IAAA,aAAA,GAAA;AAAA,QAAA,CAAA,aAAA,EAAA;AAAA,EAAA,QAAA,EAAA,MAAA,QAAA;AAAA,EAAA,OAAA,EAAA,MAAA,OAAA;AAAA,EAAA,IAAA,EAAA,MAAA,IAAA;AAAA,EAAA,KAAA,EAAA,MAAA,KAAA;AAAA,EAAA,iBAAA,EAAA,MAAA,iBAAA;AAAA,EAAA,aAAA,EAAA,MAAA,aAAA;AAAA,EAAA,WAAA,EAAA,MAAA,WAAA;AAAA,EAAA,QAAA,EAAA,MAAA,QAAA;AAAA,EAAA,WAAA,EAAA,MAAA,WAAA;AAAA,EAAA,KAAA,EAAA,MAAA,KAAA;AAAA,EAAA,UAAA,EAAA,MAAA,UAAA;AAAA,EAAA,SAAA,EAAA,MAAA,SAAA;AAAA,EAAA,OAAA,EAAA,MAAA,OAAA;AAAA,EAAA,iBAAA,EAAA,MAAA,iBAAA;AAAA,EAAA,GAAA,EAAA,MAAA,GAAA;AAAA,EAAA,MAAA,EAAA,MAAA,MAAA;AAAA,EAAA,QAAA,EAAA,MAAA;AAAA,CAAA,CAAA;;;AC6DA,SAAS,SAAS,KAAA,EAAkD;AAClE,EAAA,OAAO,OAAO,KAAA,KAAU,QAAA,IAAY,KAAA,KAAU,IAAA;AAChD;AAEA,SAAS,YAAA,CAAa,OAAgB,OAAA,EAAiB;AACrD,EAAA,IAAI,CAAC,QAAA,CAAS,KAAK,CAAA,EAAG;AACpB,IAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,EAAG,OAAO,CAAA,kBAAA,CAAoB,CAAA;AAAA,EAC9D;AACA,EAAA,OAAO,KAAA;AACT;AAEA,SAAS,cAAc,KAAA,EAAkD;AACvE,EAAA,OAAO,SAAS,KAAK,CAAA;AACvB;AAEA,SAAS,kBAAA,CAAmB,OAAgB,OAAA,EAAiB;AAC3D,EAAA,IAAI,CAAC,KAAA,CAAM,OAAA,CAAQ,KAAK,CAAA,EAAG;AACzB,IAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,EAAG,OAAO,CAAA,6BAAA,CAA+B,CAAA;AAAA,EACzE;AACA,EAAA,MAAM,SAAmB,EAAC;AAC1B,EAAA,KAAA,MAAW,SAAS,KAAA,EAAO;AACzB,IAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,KAAK,CAAA,EAAG;AAC5B,MAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,EAAG,OAAO,CAAA,2BAAA,CAA6B,CAAA;AAAA,IACvE;AACA,IAAA,MAAA,CAAO,KAAK,KAAK,CAAA;AAAA,EACnB;AACA,EAAA,OAAO,MAAA;AACT;AAYA,SAAS,kBACP,MAAA,EAC8C;AAE9C,EAAA,IAAI,CAAC,KAAA,CAAM,OAAA,CAAQ,MAAM,GAAG,OAAO,KAAA;AAEnC,EAAA,IAAI,MAAA,CAAO,MAAA,KAAW,CAAA,EAAG,OAAO,IAAA;AAEhC,EAAA,MAAM,KAAA,GAAQ,OAAO,CAAC,CAAA;AACtB,EAAA,IAAI,CAAC,KAAA,IAAS,OAAO,KAAA,KAAU,UAAU,OAAO,KAAA;AAChD,EAAA,OAAO,QAAA,IAAY,KAAA;AACrB;AAQO,IAAe,YAAf,MAGL;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EA0BA,WAAA,CACE,QACmB,QAAA,EACnB;AADmB,IAAA,IAAA,CAAA,QAAA,GAAA,QAAA;AAGnB,IAAA,IAAA,CAAK,cAAc,EAAC;AAGpB,IAAA,IAAI,CAAC,iBAAA,CAA2B,MAAM,CAAA,EAAG;AAEvC,MAAA,IAAA,CAAK,YAAY,IAAA,CAAK;AAAA,QACpB,MAAA,EAAQ,KAAA,CAAM,IAAA,CAAK,MAAM,CAAA;AAAA;AAAA,QACzB,OAAA,EAAS,EAAE,GAAG,QAAA;AAAS;AAAA,OACxB,CAAA;AAAA,IACH,CAAA,MAAO;AAEL,MAAA,KAAA,MAAW,SAAS,MAAA,EAAQ;AAE1B,QAAA,MAAM,EAAE,MAAA,EAAQ,WAAA,EAAa,GAAG,cAAa,GAAI,KAAA;AACjD,QAAA,IAAA,CAAK,YAAY,IAAA,CAAK;AAAA,UACpB,MAAA,EAAQ,KAAA,CAAM,IAAA,CAAK,WAAW,CAAA;AAAA;AAAA,UAC9B,OAAA,EAAS,EAAE,GAAG,QAAA,EAAU,GAAG,YAAA;AAAa;AAAA,SACzC,CAAA;AAAA,MACH;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EA7CO,WAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EASG,KAAA,uBAAoC,GAAA,EAAI;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAyElD,QAAA,GAAiB;AAEf,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AAEpC,MAAA,KAAA,MAAW,KAAA,IAAS,MAAM,MAAA,EAAQ;AAChC,QAAA,KAAA,CAAM,QAAA,EAAS;AAAA,MACjB;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAuBA,cAAc,UAAA,EAAuC;AAEnD,IAAA,MAAM,EAAE,MAAA,EAAQ,GAAG,OAAA,EAAQ,GAAI,UAAA;AAE/B,IAAA,IAAA,CAAK,YAAY,IAAA,CAAK;AAAA,MACpB,MAAA,EAAQ,KAAA,CAAM,IAAA,CAAK,MAAM,CAAA;AAAA;AAAA,MACzB,SAAS,EAAE,GAAG,IAAA,CAAK,QAAA,EAAU,GAAG,OAAA;AAAQ;AAAA,KACzC,CAAA;AAAA,EACH;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAgCA,SAAA,GAAY;AACV,IAAA,MAAM,UAAA,uBAAiB,GAAA,EAAwB;AAC/C,IAAA,MAAM,gBAA8B,EAAC;AACrC,IAAA,MAAM,UAAA,GAAa,CAAC,KAAA,KAAsB;AACxC,MAAA,MAAM,QAAA,GAAW,UAAA,CAAW,GAAA,CAAI,KAAK,CAAA;AACrC,MAAA,IAAI,QAAA,KAAa,QAAW,OAAO,QAAA;AACnC,MAAA,MAAM,KAAK,aAAA,CAAc,MAAA;AACzB,MAAA,aAAA,CAAc,KAAK,KAAK,CAAA;AACxB,MAAA,UAAA,CAAW,GAAA,CAAI,OAAO,EAAE,CAAA;AACxB,MAAA,OAAO,EAAA;AAAA,IACT,CAAA;AAEA,IAAA,OAAO;AAAA;AAAA,MAEL,KAAA,EAAO,KAAA,CAAM,IAAA,CAAK,IAAA,CAAK,KAAA,CAAM,OAAA,EAAS,CAAA,CAAE,GAAA,CAAI,CAAC,CAAC,KAAA,EAAO,KAAK,CAAA,MAAO;AAAA,QAC/D,OAAA,EAAS,WAAW,KAAK,CAAA;AAAA,QACzB,KAAA;AAAA;AAAA,QACA;AAAA;AAAA,OACF,CAAE,CAAA;AAAA;AAAA,MAEF,WAAA,EAAa,IAAA,CAAK,WAAA,CAAY,GAAA,CAAI,CAAC,KAAA,MAAW;AAAA,QAC5C,QAAQ,KAAA,CAAM,MAAA;AAAA;AAAA,QACd,QAAA,EAAU,MAAM,MAAA,CAAO,GAAA,CAAI,CAAC,KAAA,KAAU,UAAA,CAAW,KAAK,CAAC,CAAA;AAAA,QACvD,SAAS,KAAA,CAAM;AAAA;AAAA,OACjB,CAAE;AAAA,KACJ;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAsBA,cAAc,SAAA,EAA0C;AACtD,IAAA,MAAM,gBAAgB,IAAA,CAAK,WAAA,CAAY,QAAQ,CAAC,KAAA,KAAU,MAAM,MAAM,CAAA;AACtE,IAAA,MAAM,oBAAoB,aAAA,CAAc,MAAA;AACxC,IAAA,MAAM,WAAA,uBAAkB,GAAA,EAAqB;AAC7C,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,aAAA,CAAc,QAAQ,CAAA,EAAA,EAAK;AAC7C,MAAA,WAAA,CAAY,GAAA,CAAI,aAAA,CAAc,CAAC,CAAA,EAAG,CAAC,CAAA;AAAA,IACrC;AAGA,IAAA,IAAI,MAAA,CAAO,MAAA,CAAO,SAAA,EAAW,aAAa,CAAA,EAAG;AAC3C,MAAA,MAAM,SAAA,GAAY,UAAU,aAAa,CAAA;AACzC,MAAA,IAAI,CAAC,KAAA,CAAM,OAAA,CAAQ,SAAS,CAAA,EAAG;AAC7B,QAAA,MAAM,IAAI,oBAAoB,8BAA8B,CAAA;AAAA,MAC9D;AACA,MAAA,MAAM,WAAA,GAAyB,SAAA;AAE/B,MAAA,IAAI,WAAA,CAAY,WAAW,CAAA,EAAG;AAC5B,QAAA,IAAI,IAAA,CAAK,WAAA,CAAY,MAAA,KAAW,CAAA,EAAG;AACjC,UAAA,MAAM,IAAI,oBAAoB,6BAA6B,CAAA;AAAA,QAC7D;AACA,QAAA,IAAA,CAAK,cAAc,EAAC;AAAA,MACtB,CAAA,MAAO;AACL,QAAA,IAAI,WAAA,CAAY,MAAA,KAAW,IAAA,CAAK,WAAA,CAAY,MAAA,EAAQ;AAClD,UAAA,MAAM,IAAI,oBAAoB,4BAA4B,CAAA;AAAA,QAC5D;AAEA,QAAA,MAAM,YAAA,uBAAmB,GAAA,EAAY;AACrC,QAAA,IAAI,eAAA,GAAkB,CAAA;AACtB,QAAA,IAAI,WAAA,GAAc,KAAA;AAClB,QAAA,IAAI,aAAA,GAAgB,KAAA;AACpB,QAAA,MAAM,aAID,EAAC;AAEN,QAAA,WAAA,CAAY,OAAA,CAAQ,CAAC,QAAA,EAAU,KAAA,KAAU;AACvC,UAAA,MAAM,WAAA,GAAc,YAAA,CAAa,QAAA,EAAU,CAAA,YAAA,EAAe,KAAK,CAAA,CAAA,CAAG,CAAA;AAClE,UAAA,MAAM,aAAa,YAAA,CAAa,WAAA,CAAY,SAAS,CAAA,EAAG,CAAA,YAAA,EAAe,KAAK,CAAA,SAAA,CAAW,CAAA;AACvF,UAAA,MAAM,OAAA,GAAmB,EAAE,GAAG,IAAA,CAAK,QAAA,EAAS;AAC5C,UAAA,MAAM,aAAA,GAAyC,OAAA;AAC/C,UAAA,MAAM,cAAA,GAA0C,EAAE,GAAG,IAAA,CAAK,QAAA,EAAS;AAEnE,UAAA,KAAA,MAAW,CAAC,GAAA,EAAK,KAAK,KAAK,MAAA,CAAO,OAAA,CAAQ,UAAU,CAAA,EAAG;AACrD,YAAA,IAAI,MAAA,CAAO,MAAA,CAAO,cAAA,EAAgB,GAAG,CAAA,EAAG;AACtC,cAAA,MAAM,UAAA,GAAa,eAAe,GAAG,CAAA;AACrC,cAAA,MAAM,eAAe,OAAO,UAAA;AAC5B,cAAA,MAAM,aAAa,OAAO,KAAA;AAE1B,cAAA,IAAI,eAAe,YAAA,EAAc;AAC/B,gBAAA,MAAM,IAAI,mBAAA;AAAA,kBACR,6BAA6B,GAAG,CAAA,iBAAA,EAAoB,KAAK,CAAA,YAAA,EAAe,YAAY,SAAS,UAAU,CAAA;AAAA,iBACzG;AAAA,cACF;AACA,cAAA,aAAA,CAAc,GAAG,CAAA,GAAI,KAAA;AAAA,YACvB;AAAA,UACF;AAEA,UAAA,MAAM,WAAA,GAAc,YAAY,UAAU,CAAA;AAC1C,UAAA,MAAM,SAAA,GAAY,YAAY,QAAQ,CAAA;AACtC,UAAA,IAAI,QAAA;AACJ,UAAA,IAAI,gBAAgB,MAAA,EAAW;AAC7B,YAAA,QAAA,GAAW,kBAAA,CAAmB,WAAA,EAAa,CAAA,YAAA,EAAe,KAAK,CAAA,UAAA,CAAY,CAAA;AAC3E,YAAA,WAAA,GAAc,IAAA;AAAA,UAChB,CAAA,MAAO;AACL,YAAA,aAAA,GAAgB,IAAA;AAAA,UAClB;AAEA,UAAA,IAAI,cAAA;AAEJ,UAAA,IAAI,QAAA,EAAU;AACZ,YAAA,KAAA,MAAW,MAAM,QAAA,EAAU;AACzB,cAAA,IAAI,EAAA,GAAK,CAAA,IAAK,EAAA,IAAM,iBAAA,EAAmB;AACrC,gBAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,gBAAA,EAAmB,EAAE,CAAA,eAAA,CAAiB,CAAA;AAAA,cACtE;AACA,cAAA,IAAI,YAAA,CAAa,GAAA,CAAI,EAAE,CAAA,EAAG;AACxB,gBAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,kBAAA,EAAqB,EAAE,CAAA,eAAA,CAAiB,CAAA;AAAA,cACxE;AACA,cAAA,YAAA,CAAa,IAAI,EAAE,CAAA;AAAA,YACrB;AACA,YAAA,eAAA,IAAmB,QAAA,CAAS,MAAA;AAC5B,YAAA,cAAA,GAAiB,QAAA,CAAS,GAAA,CAAI,CAAC,EAAA,KAAO;AACpC,cAAA,MAAM,KAAA,GAAQ,cAAc,EAAE,CAAA;AAC9B,cAAA,IAAI,CAAC,KAAA,EAAO;AACV,gBAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,gBAAA,EAAmB,EAAE,CAAA,eAAA,CAAiB,CAAA;AAAA,cACtE;AACA,cAAA,OAAO,KAAA;AAAA,YACT,CAAC,CAAA;AAAA,UACH;AAEA,UAAA,IAAI,cAAc,MAAA,EAAW;AAC3B,YAAA,IAAI,CAAC,KAAA,CAAM,OAAA,CAAQ,SAAS,CAAA,EAAG;AAC7B,cAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,YAAA,EAAe,KAAK,CAAA,yBAAA,CAA2B,CAAA;AAAA,YAC/E;AACA,YAAA,MAAM,qBAAmC,EAAC;AAC1C,YAAA,IAAI,UAAA,GAAa,KAAA;AACjB,YAAA,KAAA,MAAW,YAAY,SAAA,EAAW;AAChC,cAAA,MAAM,UAAA,GAAa,WAAA,CAAY,GAAA,CAAI,QAAQ,CAAA;AAC3C,cAAA,IAAI,eAAe,MAAA,EAAW;AAC5B,gBAAA,UAAA,GAAa,IAAA;AACb,gBAAA;AAAA,cACF;AACA,cAAA,MAAM,KAAA,GAAQ,cAAc,UAAU,CAAA;AACtC,cAAA,IAAI,CAAC,KAAA,EAAO;AACV,gBAAA,UAAA,GAAa,IAAA;AACb,gBAAA;AAAA,cACF;AACA,cAAA,kBAAA,CAAmB,KAAK,KAAK,CAAA;AAAA,YAC/B;AACA,YAAA,IAAI,CAAC,UAAA,EAAY;AACf,cAAA,IAAI,QAAA,IAAY,QAAA,CAAS,MAAA,KAAW,kBAAA,CAAmB,MAAA,EAAQ;AAC7D,gBAAA,MAAM,IAAI,oBAAoB,8CAA8C,CAAA;AAAA,cAC9E;AACA,cAAA,IAAI,CAAC,cAAA,EAAgB;AACnB,gBAAA,cAAA,GAAiB,kBAAA;AAAA,cACnB;AAAA,YACF;AAAA,UACF;AAEA,UAAA,IAAI,CAAC,cAAA,EAAgB;AACnB,YAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,YAAA,EAAe,KAAK,CAAA,iCAAA,CAAmC,CAAA;AAAA,UACvF;AAEA,UAAA,IAAI,aAAa,MAAA,EAAW;AAC1B,YAAA,UAAA,CAAW,IAAA,CAAK,EAAE,MAAA,EAAQ,cAAA,EAAgB,SAAS,CAAA;AAAA,UACrD,CAAA,MAAO;AACL,YAAA,UAAA,CAAW,KAAK,EAAE,MAAA,EAAQ,cAAA,EAAgB,OAAA,EAAS,UAAU,CAAA;AAAA,UAC/D;AAAA,QACF,CAAC,CAAA;AAED,QAAA,IAAI,eAAe,aAAA,EAAe;AAChC,UAAA,MAAM,IAAI,oBAAoB,oDAAoD,CAAA;AAAA,QACpF;AAEA,QAAA,IAAI,WAAA,IAAe,oBAAoB,iBAAA,EAAmB;AACxD,UAAA,MAAM,IAAI,mBAAA;AAAA,YACR,CAAA,mCAAA,EAAsC,iBAAiB,CAAA,MAAA,EAAS,eAAe,CAAA;AAAA,WACjF;AAAA,QACF;AAEA,QAAA,IAAA,CAAK,WAAA,GAAc,UAAA,CAAW,GAAA,CAAI,CAAC,KAAA,MAAW;AAAA,UAC5C,QAAQ,KAAA,CAAM,MAAA;AAAA,UACd,SAAS,KAAA,CAAM;AAAA,SACjB,CAAE,CAAA;AAAA,MACJ;AAAA,IACF;AAGA,IAAA,IAAI,MAAA,CAAO,MAAA,CAAO,SAAA,EAAW,OAAO,CAAA,EAAG;AACrC,MAAA,MAAM,QAAA,GAAW,UAAU,OAAO,CAAA;AAClC,MAAA,IAAI,CAAC,KAAA,CAAM,OAAA,CAAQ,QAAQ,CAAA,EAAG;AAC5B,QAAA,MAAM,IAAI,oBAAoB,wBAAwB,CAAA;AAAA,MACxD;AACA,MAAA,MAAM,UAAA,GAAwB,QAAA;AAE9B,MAAA,IAAA,CAAK,MAAM,KAAA,EAAM;AAEjB,MAAA,UAAA,CAAW,OAAA,CAAQ,CAAC,QAAA,EAAU,KAAA,KAAU;AACtC,QAAA,MAAM,WAAA,GAAc,YAAA,CAAa,QAAA,EAAU,CAAA,MAAA,EAAS,KAAK,CAAA,CAAA,CAAG,CAAA;AAC5D,QAAA,IAAI,CAAC,MAAA,CAAO,MAAA,CAAO,WAAA,EAAa,OAAO,CAAA,EAAG;AACxC,UAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,MAAA,EAAS,KAAK,CAAA,mBAAA,CAAqB,CAAA;AAAA,QACnE;AACA,QAAA,MAAM,kBAAkB,YAAA,CAAa,WAAA,CAAY,OAAO,CAAA,EAAG,CAAA,MAAA,EAAS,KAAK,CAAA,OAAA,CAAS,CAAA;AAClF,QAAA,IAAI,CAAC,aAAA,CAAc,eAAe,CAAA,EAAG;AACnC,UAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,MAAA,EAAS,KAAK,CAAA,yBAAA,CAA2B,CAAA;AAAA,QACzE;AAEA,QAAA,MAAM,UAAA,GAAa,YAAY,SAAS,CAAA;AACxC,QAAA,MAAM,QAAA,GAAW,YAAY,OAAO,CAAA;AAEpC,QAAA,IAAI,aAAA;AAEJ,QAAA,IAAI,eAAe,MAAA,EAAW;AAC5B,UAAA,IACE,UAAA,KAAe,QACf,OAAO,UAAA,KAAe,YACtB,CAAC,MAAA,CAAO,SAAA,CAAU,UAAU,CAAA,EAC5B;AACA,YAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,gBAAA,EAAmB,MAAA,CAAO,UAAU,CAAC,CAAA,SAAA,CAAW,CAAA;AAAA,UAChF;AACA,UAAA,IAAI,UAAA,GAAa,CAAA,IAAK,UAAA,IAAc,iBAAA,EAAmB;AACrD,YAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,gBAAA,EAAmB,UAAU,CAAA,SAAA,CAAW,CAAA;AAAA,UACxE;AACA,UAAA,MAAM,KAAA,GAAQ,cAAc,UAAU,CAAA;AACtC,UAAA,IAAI,CAAC,KAAA,EAAO;AACV,YAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,gBAAA,EAAmB,UAAU,CAAA,SAAA,CAAW,CAAA;AAAA,UACxE;AACA,UAAA,IAAI,aAAa,MAAA,EAAW;AAC1B,YAAA,MAAM,UAAA,GAAa,WAAA,CAAY,GAAA,CAAI,QAAQ,CAAA;AAC3C,YAAA,IAAI,UAAA,KAAe,MAAA,IAAa,UAAA,KAAe,UAAA,EAAY;AACzD,cAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,QAAA,EAAW,UAAU,CAAA,8BAAA,CAAgC,CAAA;AAAA,YACrF;AAAA,UACF;AACA,UAAA,aAAA,GAAgB,KAAA;AAAA,QAClB,CAAA,MAAO;AACL,UAAA,IAAI,aAAa,MAAA,EAAW;AAC1B,YAAA,MAAM,IAAI,oBAAoB,wCAAwC,CAAA;AAAA,UACxE;AACA,UAAA,MAAM,UAAA,GAAa,WAAA,CAAY,GAAA,CAAI,QAAQ,CAAA;AAC3C,UAAA,IAAI,eAAe,MAAA,EAAW;AAC5B,YAAA,MAAM,IAAI,oBAAoB,wCAAwC,CAAA;AAAA,UACxE;AACA,UAAA,MAAM,KAAA,GAAQ,cAAc,UAAU,CAAA;AACtC,UAAA,IAAI,CAAC,KAAA,EAAO;AACV,YAAA,MAAM,IAAI,oBAAoB,wCAAwC,CAAA;AAAA,UACxE;AACA,UAAA,aAAA,GAAgB,KAAA;AAAA,QAClB;AAEA,QAAA,IAAI,CAAC,aAAA,EAAe;AAClB,UAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,sCAAA,EAAyC,KAAK,CAAA,CAAA,CAAG,CAAA;AAAA,QACjF;AACA,QAAA,IAAI,CAAC,IAAA,CAAK,OAAA,CAAQ,eAAe,CAAA,EAAG;AAClC,UAAA,MAAM,IAAI,mBAAA,CAAoB,CAAA,MAAA,EAAS,KAAK,CAAA,6BAAA,CAA+B,CAAA;AAAA,QAC7E;AACA,QAAA,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,aAAA,EAAe,eAAe,CAAA;AAAA,MAC/C,CAAC,CAAA;AAAA,IACH;AAAA,EACF;AACF;;;AC9gBA,SAAS,kBAAkB,KAAA,EAA0C;AACnE,EAAA,OAAO,KAAA,YAAiB,gBAAgB,KAAA,YAAiB,YAAA;AAC3D;AAYO,SAAS,eAAA,CAAmB,KAAA,EAAqB,KAAA,EAAe,OAAA,EAAoB;AACzF,EAAA,IAAI,KAAA,GAAQ,CAAA,IAAK,KAAA,IAAS,KAAA,CAAM,MAAA,EAAQ;AACtC,IAAA,MAAM,IAAI,WAAW,CAAA,MAAA,EAAS,KAAK,sBAAsB,KAAA,CAAM,MAAM,CAAA,KAAA,EAAQ,OAAO,CAAA,CAAA,EAAI;AAAA,MACtF,KAAA;AAAA,MACA,UAAA,EAAY,CAAC,CAAA,EAAG,KAAA,CAAM,SAAS,CAAC;AAAA,KACjC,CAAA;AAAA,EACH;AACA,EAAA,MAAM,KAAA,GAAQ,MAAM,KAAK,CAAA;AACzB,EAAA,IAAI,UAAU,MAAA,EAAW;AACvB,IAAA,MAAM,IAAI,YAAA,CAAa,CAAA,8BAAA,EAAiC,KAAK,CAAA,IAAA,EAAO,OAAO,CAAA,CAAE,CAAA;AAAA,EAC/E;AACA,EAAA,OAAO,KAAA;AACT;AASO,SAAS,uBAAA,CAAwB,MAAc,KAAA,EAAqB;AACzE,EAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,KAAK,CAAA,IAAK,QAAQ,CAAA,EAAG;AACxC,IAAA,MAAM,IAAI,sBAAsB,CAAA,QAAA,EAAW,IAAI,KAAK,KAAK,CAAA,CAAA,EAAI,MAAM,KAAK,CAAA;AAAA,EAC1E;AACF;AASO,SAAS,oBAAA,CAAqB,MAAc,KAAA,EAAqB;AACtE,EAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,KAAK,CAAA,IAAK,SAAS,CAAA,EAAG;AACzC,IAAA,MAAM,IAAI,sBAAsB,CAAA,QAAA,EAAW,IAAI,KAAK,KAAK,CAAA,cAAA,CAAA,EAAkB,MAAM,KAAK,CAAA;AAAA,EACxF;AACF;AASO,SAAS,YAAA,CAAa,MAAc,KAAA,EAAqB;AAC9D,EAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,KAAK,CAAA,EAAG;AAC3B,IAAA,MAAM,IAAI,sBAAsB,CAAA,QAAA,EAAW,IAAI,KAAK,KAAK,CAAA,CAAA,EAAI,MAAM,KAAK,CAAA;AAAA,EAC1E;AACF;AAWO,SAAS,aAAA,CAAc,IAAA,EAAc,KAAA,EAAe,GAAA,EAAa,GAAA,EAAmB;AACzF,EAAA,IAAI,CAAC,OAAO,QAAA,CAAS,KAAK,KAAK,KAAA,GAAQ,GAAA,IAAO,SAAS,GAAA,EAAK;AAC1D,IAAA,MAAM,IAAI,qBAAA;AAAA,MACR,WAAW,IAAI,CAAA,EAAA,EAAK,KAAK,CAAA,oBAAA,EAAuB,GAAG,KAAK,GAAG,CAAA,EAAA,CAAA;AAAA,MAC3D,IAAA;AAAA,MACA;AAAA,KACF;AAAA,EACF;AACF;AAaO,SAAS,kBAAA,CACd,OACA,aAAA,EAMA;AACA,EAAA,IAAI,CAAC,MAAM,YAAA,EAAc;AACvB,IAAA,MAAM,IAAI,qBAAA;AAAA,MACR,qDAAA;AAAA,MACA,cAAA;AAAA,MACA;AAAA,KACF;AAAA,EACF;AAEA,EAAA,MAAM,IAAI,KAAA,CAAM,IAAA;AAChB,EAAA,IAAI,CAAC,CAAA,EAAG;AACN,IAAA,MAAM,IAAI,cAAA;AAAA,MACR;AAAA,KACF;AAAA,EACF;AAEA,EAAA,MAAM,SAAA,GAAY,MAAM,MAAA,CAAO,IAAA;AAC/B,EAAA,MAAM,WAAW,CAAA,CAAE,IAAA;AAEnB,EAAA,IAAI,CAAC,iBAAA,CAAkB,SAAS,KAAK,CAAC,iBAAA,CAAkB,QAAQ,CAAA,EAAG;AACjE,IAAA,MAAM,IAAI,UAAA;AAAA,MACR,GAAG,aAAa,CAAA,qEAAA;AAAA,KAClB;AAAA,EACF;AAEA,EAAA,IAAI,SAAA,CAAU,WAAA,KAAgB,QAAA,CAAS,WAAA,EAAa;AAClD,IAAA,MAAM,IAAI,UAAA;AAAA,MACR,GAAG,aAAa,CAAA,0DAAA;AAAA,KAClB;AAAA,EACF;AAEA,EAAA,IAAI,KAAA,CAAM,MAAA,CAAO,IAAA,KAAS,CAAA,CAAE,IAAA,EAAM;AAChC,IAAA,MAAM,IAAI,UAAA;AAAA,MACR,qDAAqD,KAAA,CAAM,MAAA,CAAO,IAAI,CAAA,QAAA,EAAW,EAAE,IAAI,CAAA,CAAA;AAAA,KACzF;AAAA,EACF;AAEA,EAAA,OAAO;AAAA,IACL,IAAA,EAAM,QAAA;AAAA,IACN,YAAY,CAAA,CAAE,MAAA;AAAA,IACd,KAAA,EAAO,SAAA;AAAA,IACP,WAAA,EAAa,MAAM,MAAA,CAAO;AAAA,GAC5B;AACF;AAUO,SAAS,gBAAA,CACd,MAAA,EACA,YAAA,EACA,UAAA,EACM;AACN,EAAA,IAAI,MAAA,CAAO,WAAW,YAAA,EAAc;AAClC,IAAA,MAAM,IAAI,YAAA;AAAA,MACR,kCAAkC,UAAU,CAAA,WAAA,EAAc,YAAY,CAAA,MAAA,EAAS,OAAO,MAAM,CAAA;AAAA,KAC9F;AAAA,EACF;AACF;;;ACzIO,IAAM,QAAA,GAAN,cAAuB,SAAA,CAA0C;AAAA,EAC9D,UAAA,GAAa,CAAA;AAAA,EAErB,IAAI,SAAA,GAAoB;AACtB,IAAA,OAAO,IAAA,CAAK,UAAA;AAAA,EACd;AAAA,EACA,WAAA,CACE,MAAA,EACA,OAAA,GAKI,EAAC,EACL;AACA,IAAA,MAAM,QAAA,GAAW;AAAA,MACf,EAAA,EAAI,QAAQ,EAAA,IAAM,CAAA;AAAA,MAClB,GAAA,EAAK,QAAQ,GAAA,IAAO,GAAA;AAAA,MACpB,GAAA,EAAK,QAAQ,GAAA,IAAO,IAAA;AAAA,MACpB,WAAA,EAAa,QAAQ,WAAA,IAAe;AAAA,KACtC;AAEA,IAAA,KAAA,CAAM,QAAQ,QAAQ,CAAA;AAGtB,IAAA,uBAAA,CAAwB,eAAA,EAAiB,SAAS,EAAE,CAAA;AACpD,IAAA,aAAA,CAAc,KAAA,EAAO,QAAA,CAAS,GAAA,EAAK,CAAA,EAAG,CAAC,CAAA;AACvC,IAAA,oBAAA,CAAqB,SAAA,EAAW,SAAS,GAAG,CAAA;AAC5C,IAAA,uBAAA,CAAwB,oBAAA,EAAsB,SAAS,WAAW,CAAA;AAAA,EACpE;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,eAAA,CAAgB,WAAW,CAAA,EAAW;AACpC,IAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,WAAA,CAAY,QAAQ,CAAA;AACvC,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,CAAA,qBAAA,EAAwB,QAAQ,CAAA,mBAAA,EAAsB,IAAA,CAAK,YAAY,MAAM,CAAA,EAAA,CAAA;AAAA,QAC7E,UAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,OAAO,MAAM,OAAA,CAAQ,EAAA;AAAA,EACvB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,gBAAgB,EAAA,EAAkB;AAChC,IAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,KAAA,CAAM,QAAQ,EAAA,GAAK,EAAA;AAAA,IACrB;AAAA,EACF;AAAA,EAEU,QAAQ,KAAA,EAAwD;AACxE,IAAA,OAAO,MAAM,WAAW,CAAA,YAAa,YAAA,IAAgB,KAAA,CAAM,UAAU,CAAA,YAAa,YAAA;AAAA,EACpF;AAAA,EAEA,KAAK,OAAA,EAA4C;AAC/C,IAAA,IAAI,IAAA;AAEJ,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,IAAA,GAAO,OAAA,EAAQ;AAAA,IACjB;AAGA,IAAA,IAAA,CAAK,UAAA,EAAA;AAEL,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,MAAM,EAAE,EAAA,EAAI,GAAA,EAAK,GAAA,EAAK,WAAA,KAAgB,KAAA,CAAM,OAAA;AAG5C,MAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,MAAA,aAAA,CAAc,KAAA,EAAO,GAAA,EAAK,CAAA,EAAG,CAAC,CAAA;AAC9B,MAAA,oBAAA,CAAqB,WAAW,GAAG,CAAA;AACnC,MAAA,uBAAA,CAAwB,sBAAsB,WAAW,CAAA;AAEzD,MAAA,KAAA,MAAW,KAAA,IAAS,MAAM,MAAA,EAAQ;AAChC,QAAA,MAAM;AAAA,UACJ,IAAA,EAAM,QAAA;AAAA,UACN,UAAA,EAAY,IAAA;AAAA,UACZ,KAAA,EAAO,KAAA;AAAA,UACP,WAAA,EAAa;AAAA,SACf,GAAI,kBAAA,CAAmB,KAAA,EAAO,UAAU,CAAA;AACxC,QAAA,MAAM,IAAA,GAAO,MAAM,MAAA,CAAO,IAAA;AAG1B,QAAA,IAAI,KAAA,GAAQ,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAK,CAAA;AAChC,QAAA,IAAI,CAAC,KAAA,EAAO;AACV,UAAA,KAAA,GAAQ;AAAA,YACN,SAAA,EAAW,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA,YAChC,QAAA,EAAU,IAAI,YAAA,CAAa,IAAI;AAAA,WACjC;AACA,UAAA,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAA,EAAO,KAAK,CAAA;AAAA,QAC7B;AAGA,QAAA,gBAAA,CAAiB,KAAA,CAAM,SAAA,EAAW,IAAA,EAAM,oBAAoB,CAAA;AAC5D,QAAA,gBAAA,CAAiB,KAAA,CAAM,QAAA,EAAU,IAAA,EAAM,mBAAmB,CAAA;AAE1D,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAC7B,UAAA,MAAM,GAAA,GAAM,eAAA,CAAgB,QAAA,EAAU,IAAA,GAAO,GAAG,mBAAmB,CAAA;AACnE,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,KAAA,EAAO,IAAA,GAAO,GAAG,oBAAoB,CAAA;AAChE,UAAA,YAAA,CAAa,YAAY,GAAG,CAAA;AAC5B,UAAA,YAAA,CAAa,aAAa,EAAE,CAAA;AAG5B,UAAA,MAAM,EAAA,GAAK,WAAA,KAAgB,CAAA,GAAI,GAAA,GAAM,cAAc,EAAA,GAAK,GAAA;AAGxD,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,KAAA,CAAM,SAAA,EAAW,GAAG,oBAAoB,CAAA;AACnE,UAAA,MAAM,KAAA,GAAQ,GAAA,GAAM,EAAA,GAAA,CAAM,CAAA,GAAI,OAAO,EAAA,GAAK,EAAA;AAC1C,UAAA,KAAA,CAAM,SAAA,CAAU,CAAC,CAAA,GAAI,KAAA;AAGrB,UAAA,MAAM,GAAA,GAAM,IAAA,CAAK,IAAA,CAAK,KAAA,GAAQ,GAAG,CAAA;AAGjC,UAAA,MAAM,IAAA,GAAO,eAAA,CAAgB,KAAA,CAAM,QAAA,EAAU,GAAG,mBAAmB,CAAA;AACnE,UAAA,MAAM,SAAA,GAAY,IAAA,CAAK,IAAA,CAAK,IAAA,GAAO,GAAG,CAAA;AAGtC,UAAA,MAAM,KAAA,GAAS,YAAY,GAAA,GAAO,EAAA;AAGlC,UAAA,KAAA,CAAM,SAAS,CAAC,CAAA,GAAI,MAAM,IAAA,GAAA,CAAQ,CAAA,GAAI,OAAO,KAAA,GAAQ,KAAA;AAGrD,UAAA,KAAA,CAAM,IAAA,GAAO,CAAC,CAAA,GAAI,EAAA,GAAK,EAAA,GAAK,KAAA;AAAA,QAC9B;AAAA,MACF;AAAA,IACF;AAEA,IAAA,OAAO,IAAA;AAAA,EACT;AACF;;;AChJO,IAAM,OAAA,GAAN,cAAsB,SAAA,CAAwC;AAAA,EAC3D,UAAA,GAAa,CAAA;AAAA,EAErB,IAAI,SAAA,GAAoB;AACtB,IAAA,OAAO,IAAA,CAAK,UAAA;AAAA,EACd;AAAA,EACA,WAAA,CACE,MAAA,EACA,OAAA,GAKI,EAAC,EACL;AACA,IAAA,MAAM,QAAA,GAAW;AAAA,MACf,EAAA,EAAI,QAAQ,EAAA,IAAM,IAAA;AAAA,MAClB,GAAA,EAAK,QAAQ,GAAA,IAAO,KAAA;AAAA,MACpB,WAAA,EAAa,QAAQ,WAAA,IAAe,CAAA;AAAA,MACpC,OAAA,EAAS,QAAQ,OAAA,IAAW;AAAA,KAC9B;AAEA,IAAA,KAAA,CAAM,QAAQ,QAAQ,CAAA;AAEtB,IAAA,uBAAA,CAAwB,eAAA,EAAiB,SAAS,EAAE,CAAA;AACpD,IAAA,oBAAA,CAAqB,SAAA,EAAW,SAAS,GAAG,CAAA;AAC5C,IAAA,uBAAA,CAAwB,oBAAA,EAAsB,SAAS,WAAW,CAAA;AAClE,IAAA,uBAAA,CAAwB,UAAA,EAAY,SAAS,OAAO,CAAA;AAAA,EACtD;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,eAAA,CAAgB,WAAW,CAAA,EAAW;AACpC,IAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,WAAA,CAAY,QAAQ,CAAA;AACvC,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,CAAA,qBAAA,EAAwB,QAAQ,CAAA,mBAAA,EAAsB,IAAA,CAAK,YAAY,MAAM,CAAA,EAAA,CAAA;AAAA,QAC7E,UAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,OAAO,MAAM,OAAA,CAAQ,EAAA;AAAA,EACvB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,gBAAgB,EAAA,EAAkB;AAChC,IAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,KAAA,CAAM,QAAQ,EAAA,GAAK,EAAA;AAAA,IACrB;AAAA,EACF;AAAA,EAEU,QAAQ,KAAA,EAAuD;AACvE,IAAA,OAAO,OAAO,KAAA,CAAM,MAAM,MAAM,QAAA,IAAY,KAAA,CAAM,KAAK,CAAA,YAAa,YAAA;AAAA,EACtE;AAAA,EAEA,KAAK,OAAA,EAA4C;AAC/C,IAAA,IAAI,IAAA;AAEJ,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,IAAA,GAAO,OAAA,EAAQ;AAAA,IACjB;AAGA,IAAA,IAAA,CAAK,UAAA,EAAA;AAEL,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,MAAM,EAAE,EAAA,EAAI,GAAA,EAAK,WAAA,EAAa,OAAA,KAAY,KAAA,CAAM,OAAA;AAEhD,MAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,MAAA,oBAAA,CAAqB,WAAW,GAAG,CAAA;AACnC,MAAA,uBAAA,CAAwB,sBAAsB,WAAW,CAAA;AACzD,MAAA,uBAAA,CAAwB,YAAY,OAAO,CAAA;AAE3C,MAAA,KAAA,MAAW,KAAA,IAAS,MAAM,MAAA,EAAQ;AAChC,QAAA,MAAM;AAAA,UACJ,IAAA,EAAM,QAAA;AAAA,UACN,UAAA,EAAY,IAAA;AAAA,UACZ,KAAA,EAAO,KAAA;AAAA,UACP,WAAA,EAAa;AAAA,SACf,GAAI,kBAAA,CAAmB,KAAA,EAAO,SAAS,CAAA;AACvC,QAAA,MAAM,IAAA,GAAO,MAAM,MAAA,CAAO,IAAA;AAE1B,QAAA,MAAM,QAAA,GAAW,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAK,CAAA;AACrC,QAAA,MAAM,KAAA,GACJ,aACC,MAAM;AACL,UAAA,MAAM,IAAA,GAAO;AAAA,YACX,IAAA,EAAM,CAAA;AAAA,YACN,GAAA,EAAK,IAAI,YAAA,CAAa,IAAI;AAAA,WAC5B;AACA,UAAA,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAA,EAAO,IAAI,CAAA;AAC1B,UAAA,OAAO,IAAA;AAAA,QACT,CAAA,GAAG;AAGL,QAAA,gBAAA,CAAiB,KAAA,CAAM,GAAA,EAAK,IAAA,EAAM,aAAa,CAAA;AAE/C,QAAA,KAAA,CAAM,IAAA,IAAQ,CAAA;AAEd,QAAA,MAAM,GAAA,GAAM,EAAA,IAAM,CAAA,GAAA,CAAK,KAAA,CAAM,OAAO,CAAA,IAAK,OAAA,CAAA;AAEzC,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAC7B,UAAA,MAAM,GAAA,GAAM,eAAA,CAAgB,QAAA,EAAU,IAAA,GAAO,GAAG,kBAAkB,CAAA;AAClE,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,KAAA,EAAO,IAAA,GAAO,GAAG,mBAAmB,CAAA;AAC/D,UAAA,YAAA,CAAa,YAAY,GAAG,CAAA;AAC5B,UAAA,YAAA,CAAa,aAAa,EAAE,CAAA;AAE5B,UAAA,MAAM,EAAA,GAAK,WAAA,KAAgB,CAAA,GAAI,GAAA,GAAM,cAAc,EAAA,GAAK,GAAA;AAExD,UAAA,MAAM,MAAA,GAAS,eAAA,CAAgB,KAAA,CAAM,GAAA,EAAK,GAAG,aAAa,CAAA;AAC1D,UAAA,MAAM,MAAA,GAAS,SAAS,EAAA,GAAK,EAAA;AAC7B,UAAA,KAAA,CAAM,GAAA,CAAI,CAAC,CAAA,GAAI,MAAA;AAEf,UAAA,MAAM,GAAA,GAAM,IAAA,CAAK,IAAA,CAAK,MAAM,CAAA,GAAI,GAAA;AAChC,UAAA,KAAA,CAAM,IAAA,GAAO,CAAC,CAAA,GAAI,EAAA,GAAK,OAAO,EAAA,GAAK,GAAA,CAAA;AAAA,QACrC;AAAA,MACF;AAAA,IACF;AAEA,IAAA,OAAO,IAAA;AAAA,EACT;AACF;;;ACrIO,IAAM,IAAA,GAAN,cAAmB,SAAA,CAAkC;AAAA,EAClD,UAAA,GAAa,CAAA;AAAA,EAErB,IAAI,SAAA,GAAoB;AACtB,IAAA,OAAO,IAAA,CAAK,UAAA;AAAA,EACd;AAAA,EAEA,WAAA,CACE,MAAA,EACA,OAAA,GAOI,EAAC,EACL;AACA,IAAA,MAAM,QAAA,GAAW;AAAA,MACf,EAAA,EAAI,QAAQ,EAAA,IAAM,IAAA;AAAA,MAClB,KAAA,EAAO,QAAQ,KAAA,IAAS,GAAA;AAAA,MACxB,KAAA,EAAO,QAAQ,KAAA,IAAS,KAAA;AAAA,MACxB,GAAA,EAAK,QAAQ,GAAA,IAAO,IAAA;AAAA,MACpB,WAAA,EAAa,QAAQ,WAAA,IAAe,CAAA;AAAA,MACpC,OAAA,EAAS,QAAQ,OAAA,IAAW;AAAA,KAC9B;AAEA,IAAA,KAAA,CAAM,QAAQ,QAAQ,CAAA;AAEtB,IAAA,uBAAA,CAAwB,eAAA,EAAiB,SAAS,EAAE,CAAA;AACpD,IAAA,aAAA,CAAc,OAAA,EAAS,QAAA,CAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAC3C,IAAA,aAAA,CAAc,OAAA,EAAS,QAAA,CAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAC3C,IAAA,oBAAA,CAAqB,SAAA,EAAW,SAAS,GAAG,CAAA;AAC5C,IAAA,uBAAA,CAAwB,oBAAA,EAAsB,SAAS,WAAW,CAAA;AAAA,EACpE;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,eAAA,CAAgB,WAAW,CAAA,EAAW;AACpC,IAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,WAAA,CAAY,QAAQ,CAAA;AACvC,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,CAAA,qBAAA,EAAwB,QAAQ,CAAA,mBAAA,EAAsB,IAAA,CAAK,YAAY,MAAM,CAAA,EAAA,CAAA;AAAA,QAC7E,UAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,OAAO,MAAM,OAAA,CAAQ,EAAA;AAAA,EACvB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,gBAAgB,EAAA,EAAkB;AAChC,IAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,KAAA,CAAM,QAAQ,EAAA,GAAK,EAAA;AAAA,IACrB;AAAA,EACF;AAAA,EAEU,QAAQ,KAAA,EAAoD;AACpE,IAAA,MAAM,WAAA,GACJ,OAAO,KAAA,CAAM,MAAM,CAAA,KAAM,QAAA,IACzB,KAAA,CAAM,QAAQ,CAAA,YAAa,YAAA,IAC3B,KAAA,CAAM,UAAU,CAAA,YAAa,YAAA;AAC/B,IAAA,IAAI,CAAC,aAAa,OAAO,KAAA;AACzB,IAAA,IAAI,KAAA,CAAM,aAAa,CAAA,KAAM,MAAA,IAAa,EAAE,KAAA,CAAM,aAAa,aAAa,YAAA,CAAA,EAAe;AACzF,MAAA,OAAO,KAAA;AAAA,IACT;AACA,IAAA,OAAO,IAAA;AAAA,EACT;AAAA,EAEA,KAAK,OAAA,EAA4C;AAC/C,IAAA,IAAI,IAAA;AAEJ,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,IAAA,GAAO,OAAA,EAAQ;AAAA,IACjB;AAEA,IAAA,IAAA,CAAK,UAAA,EAAA;AAEL,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,MAAM,EAAE,IAAI,KAAA,EAAO,KAAA,EAAO,KAAK,WAAA,EAAa,OAAA,KAAY,KAAA,CAAM,OAAA;AAE9D,MAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,MAAA,aAAA,CAAc,OAAA,EAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAClC,MAAA,aAAA,CAAc,OAAA,EAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAClC,MAAA,oBAAA,CAAqB,WAAW,GAAG,CAAA;AACnC,MAAA,uBAAA,CAAwB,sBAAsB,WAAW,CAAA;AAEzD,MAAA,KAAA,MAAW,KAAA,IAAS,MAAM,MAAA,EAAQ;AAChC,QAAA,MAAM;AAAA,UACJ,IAAA,EAAM,QAAA;AAAA,UACN,UAAA;AAAA,UACA,KAAA,EAAO,SAAA;AAAA,UACP;AAAA,SACF,GAAI,kBAAA,CAAmB,KAAA,EAAO,MAAM,CAAA;AACpC,QAAA,MAAM,IAAA,GAAO,MAAM,MAAA,CAAO,IAAA;AAE1B,QAAA,MAAM,QAAA,GAAW,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAK,CAAA;AACrC,QAAA,MAAM,KAAA,GACJ,aACC,MAAM;AACL,UAAA,MAAM,IAAA,GAAO;AAAA,YACX,IAAA,EAAM,CAAA;AAAA,YACN,MAAA,EAAQ,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA,YAC7B,QAAA,EAAU,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA,YAC/B,GAAI,UAAU,EAAE,WAAA,EAAa,IAAI,YAAA,CAAa,IAAI,CAAA,EAAE,GAAI;AAAC,WAC3D;AACA,UAAA,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAA,EAAO,IAAI,CAAA;AAC1B,UAAA,OAAO,IAAA;AAAA,QACT,CAAA,GAAG;AAGL,QAAA,gBAAA,CAAiB,KAAA,CAAM,MAAA,EAAQ,IAAA,EAAM,aAAa,CAAA;AAClD,QAAA,gBAAA,CAAiB,KAAA,CAAM,QAAA,EAAU,IAAA,EAAM,eAAe,CAAA;AACtD,QAAA,IAAI,OAAA,IAAW,MAAM,WAAA,EAAa;AAChC,UAAA,gBAAA,CAAiB,KAAA,CAAM,WAAA,EAAa,IAAA,EAAM,kBAAkB,CAAA;AAAA,QAC9D;AAEA,QAAA,KAAA,CAAM,IAAA,IAAQ,CAAA;AAGd,QAAA,MAAM,eAAA,GAAkB,CAAA,GAAI,KAAA,IAAS,KAAA,CAAM,IAAA;AAC3C,QAAA,MAAM,eAAA,GAAkB,CAAA,GAAI,KAAA,IAAS,KAAA,CAAM,IAAA;AAE3C,QAAA,MAAM,WAAW,EAAA,GAAK,eAAA;AAEtB,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAC7B,UAAA,MAAM,GAAA,GAAM,eAAA,CAAgB,QAAA,EAAU,UAAA,GAAa,GAAG,eAAe,CAAA;AACrE,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,SAAA,EAAW,WAAA,GAAc,GAAG,gBAAgB,CAAA;AACvE,UAAA,YAAA,CAAa,YAAY,GAAG,CAAA;AAC5B,UAAA,YAAA,CAAa,aAAa,EAAE,CAAA;AAG5B,UAAA,MAAM,EAAA,GAAK,WAAA,KAAgB,CAAA,GAAI,GAAA,GAAM,cAAc,EAAA,GAAK,GAAA;AAExD,UAAA,MAAM,CAAA,GAAI,eAAA,CAAgB,KAAA,CAAM,MAAA,EAAQ,GAAG,aAAa,CAAA;AACxD,UAAA,MAAM,CAAA,GAAI,eAAA,CAAgB,KAAA,CAAM,QAAA,EAAU,GAAG,eAAe,CAAA;AAE5D,UAAA,MAAM,IAAA,GAAO,KAAA,GAAQ,CAAA,GAAA,CAAK,CAAA,GAAI,KAAA,IAAS,EAAA;AACvC,UAAA,MAAM,IAAA,GAAO,KAAA,GAAQ,CAAA,GAAA,CAAK,CAAA,GAAI,SAAS,EAAA,GAAK,EAAA;AAE5C,UAAA,KAAA,CAAM,MAAA,CAAO,CAAC,CAAA,GAAI,IAAA;AAClB,UAAA,KAAA,CAAM,QAAA,CAAS,CAAC,CAAA,GAAI,IAAA;AAEpB,UAAA,IAAI,OAAA,GAAU,IAAA;AACd,UAAA,IAAI,OAAA,EAAS;AACX,YAAA,MAAM,SAAS,KAAA,CAAM,WAAA;AACrB,YAAA,IAAI,CAAC,MAAA,EAAQ;AACX,cAAA,MAAM,IAAI,aAAa,4DAA4D,CAAA;AAAA,YACrF;AACA,YAAA,MAAM,IAAA,GAAO,KAAK,GAAA,CAAI,eAAA,CAAgB,QAAQ,CAAA,EAAG,kBAAkB,GAAG,IAAI,CAAA;AAC1E,YAAA,MAAA,CAAO,CAAC,CAAA,GAAI,IAAA;AACZ,YAAA,OAAA,GAAU,IAAA;AAAA,UACZ;AAEA,UAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,IAAA,CAAK,OAAA,GAAU,eAAe,CAAA,GAAI,GAAA;AACrD,UAAA,SAAA,CAAU,WAAA,GAAc,CAAC,CAAA,GAAI,EAAA,GAAK,YAAY,IAAA,GAAO,KAAA,CAAA;AAAA,QACvD;AAAA,MACF;AAAA,IACF;AAEA,IAAA,OAAO,IAAA;AAAA,EACT;AACF;;;AChJO,IAAM,KAAA,GAAN,cAAoB,SAAA,CAAoC;AAAA;AAAA,EAErD,UAAA,GAAa,CAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOrB,IAAI,SAAA,GAAoB;AACtB,IAAA,OAAO,IAAA,CAAK,UAAA;AAAA,EACd;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAeA,WAAA,CACE,MAAA,EACA,OAAA,GAOI,EAAC,EACL;AAEA,IAAA,MAAM,QAAA,GAAW;AAAA,MACf,EAAA,EAAI,QAAQ,EAAA,IAAM,IAAA;AAAA,MAClB,KAAA,EAAO,QAAQ,KAAA,IAAS,GAAA;AAAA,MACxB,KAAA,EAAO,QAAQ,KAAA,IAAS,KAAA;AAAA,MACxB,GAAA,EAAK,QAAQ,GAAA,IAAO,IAAA;AAAA,MACpB,WAAA,EAAa,QAAQ,WAAA,IAAe,IAAA;AAAA;AAAA,MACpC,OAAA,EAAS,QAAQ,OAAA,IAAW;AAAA,KAC9B;AAEA,IAAA,KAAA,CAAM,QAAQ,QAAQ,CAAA;AAGtB,IAAA,uBAAA,CAAwB,eAAA,EAAiB,SAAS,EAAE,CAAA;AACpD,IAAA,aAAA,CAAc,OAAA,EAAS,QAAA,CAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAC3C,IAAA,aAAA,CAAc,OAAA,EAAS,QAAA,CAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAC3C,IAAA,oBAAA,CAAqB,SAAA,EAAW,SAAS,GAAG,CAAA;AAC5C,IAAA,uBAAA,CAAwB,oBAAA,EAAsB,SAAS,WAAW,CAAA;AAAA,EACpE;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,eAAA,CAAgB,WAAW,CAAA,EAAW;AACpC,IAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,WAAA,CAAY,QAAQ,CAAA;AACvC,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,CAAA,qBAAA,EAAwB,QAAQ,CAAA,mBAAA,EAAsB,IAAA,CAAK,YAAY,MAAM,CAAA,EAAA,CAAA;AAAA,QAC7E,UAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,OAAO,MAAM,OAAA,CAAQ,EAAA;AAAA,EACvB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,gBAAgB,EAAA,EAAkB;AAChC,IAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,KAAA,CAAM,QAAQ,EAAA,GAAK,EAAA;AAAA,IACrB;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUU,QAAQ,KAAA,EAAqD;AACrE,IAAA,MAAM,WAAA,GACJ,OAAO,KAAA,CAAM,MAAM,CAAA,KAAM,QAAA,IACzB,KAAA,CAAM,QAAQ,CAAA,YAAa,YAAA,IAC3B,KAAA,CAAM,UAAU,CAAA,YAAa,YAAA;AAC/B,IAAA,IAAI,CAAC,aAAa,OAAO,KAAA;AACzB,IAAA,IAAI,KAAA,CAAM,aAAa,CAAA,KAAM,MAAA,IAAa,EAAE,KAAA,CAAM,aAAa,aAAa,YAAA,CAAA,EAAe;AACzF,MAAA,OAAO,KAAA;AAAA,IACT;AACA,IAAA,OAAO,IAAA;AAAA,EACT;AAAA,EAEA,KAAK,OAAA,EAA4C;AAC/C,IAAA,IAAI,IAAA;AAGJ,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,IAAA,GAAO,OAAA,EAAQ;AAAA,IACjB;AAGA,IAAA,IAAA,CAAK,UAAA,EAAA;AAGL,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,MAAM,EAAE,IAAI,KAAA,EAAO,KAAA,EAAO,KAAK,WAAA,EAAa,OAAA,KAAY,KAAA,CAAM,OAAA;AAG9D,MAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,MAAA,aAAA,CAAc,OAAA,EAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAClC,MAAA,aAAA,CAAc,OAAA,EAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAClC,MAAA,oBAAA,CAAqB,WAAW,GAAG,CAAA;AACnC,MAAA,uBAAA,CAAwB,sBAAsB,WAAW,CAAA;AAGzD,MAAA,KAAA,MAAW,KAAA,IAAS,MAAM,MAAA,EAAQ;AAEhC,QAAA,MAAM;AAAA,UACJ,IAAA;AAAA,UACA,UAAA;AAAA,UACA,KAAA,EAAO,KAAA;AAAA,UACP,WAAA,EAAa;AAAA,SACf,GAAI,kBAAA,CAAmB,KAAA,EAAO,OAAO,CAAA;AACrC,QAAA,MAAM,IAAA,GAAO,MAAM,MAAA,CAAO,IAAA;AAG1B,QAAA,MAAM,QAAA,GAAW,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAK,CAAA;AACrC,QAAA,MAAM,KAAA,GACJ,aACC,MAAM;AAEL,UAAA,MAAM,IAAA,GAAO;AAAA,YACX,IAAA,EAAM,CAAA;AAAA,YACN,MAAA,EAAQ,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA;AAAA,YAC7B,QAAA,EAAU,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA;AAAA,YAC/B,GAAI,UAAU,EAAE,WAAA,EAAa,IAAI,YAAA,CAAa,IAAI,CAAA,EAAE,GAAI;AAAC;AAAA,WAC3D;AACA,UAAA,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAA,EAAO,IAAI,CAAA;AAC1B,UAAA,OAAO,IAAA;AAAA,QACT,CAAA,GAAG;AAGL,QAAA,gBAAA,CAAiB,KAAA,CAAM,MAAA,EAAQ,IAAA,EAAM,cAAc,CAAA;AACnD,QAAA,gBAAA,CAAiB,KAAA,CAAM,QAAA,EAAU,IAAA,EAAM,gBAAgB,CAAA;AACvD,QAAA,IAAI,OAAA,IAAW,MAAM,WAAA,EAAa;AAChC,UAAA,gBAAA,CAAiB,KAAA,CAAM,WAAA,EAAa,IAAA,EAAM,mBAAmB,CAAA;AAAA,QAC/D;AAGA,QAAA,KAAA,CAAM,IAAA,IAAQ,CAAA;AAGd,QAAA,MAAM,eAAA,GAAkB,CAAA,GAAI,KAAA,IAAS,KAAA,CAAM,IAAA;AAC3C,QAAA,MAAM,eAAA,GAAkB,CAAA,GAAI,KAAA,IAAS,KAAA,CAAM,IAAA;AAG3C,QAAA,MAAM,WAAW,EAAA,GAAK,eAAA;AAGtB,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAE7B,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,IAAA,EAAM,UAAA,GAAa,GAAG,gBAAgB,CAAA;AACjE,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,KAAA,EAAO,IAAA,GAAO,GAAG,iBAAiB,CAAA;AAG7D,UAAA,YAAA,CAAa,YAAY,EAAE,CAAA;AAC3B,UAAA,YAAA,CAAa,aAAa,EAAE,CAAA;AAG5B,UAAA,MAAM,CAAA,GAAI,eAAA,CAAgB,KAAA,CAAM,MAAA,EAAQ,GAAG,cAAc,CAAA;AACzD,UAAA,MAAM,CAAA,GAAI,eAAA,CAAgB,KAAA,CAAM,QAAA,EAAU,GAAG,gBAAgB,CAAA;AAG7D,UAAA,MAAM,IAAA,GAAO,KAAA,GAAQ,CAAA,GAAA,CAAK,CAAA,GAAI,KAAA,IAAS,EAAA;AAGvC,UAAA,MAAM,IAAA,GAAO,KAAA,GAAQ,CAAA,GAAA,CAAK,CAAA,GAAI,SAAS,EAAA,GAAK,EAAA;AAG5C,UAAA,KAAA,CAAM,MAAA,CAAO,CAAC,CAAA,GAAI,IAAA;AAClB,UAAA,KAAA,CAAM,QAAA,CAAS,CAAC,CAAA,GAAI,IAAA;AAGpB,UAAA,IAAI,OAAA,GAAU,IAAA;AACd,UAAA,IAAI,OAAA,EAAS;AACX,YAAA,MAAM,SAAS,KAAA,CAAM,WAAA;AACrB,YAAA,IAAI,CAAC,MAAA,EAAQ;AACX,cAAA,MAAM,IAAI,aAAa,4DAA4D,CAAA;AAAA,YACrF;AAEA,YAAA,MAAM,IAAA,GAAO,KAAK,GAAA,CAAI,eAAA,CAAgB,QAAQ,CAAA,EAAG,mBAAmB,GAAG,IAAI,CAAA;AAC3E,YAAA,MAAA,CAAO,CAAC,CAAA,GAAI,IAAA;AACZ,YAAA,OAAA,GAAU,IAAA;AAAA,UACZ;AAGA,UAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,IAAA,CAAK,OAAA,GAAU,eAAe,CAAA,GAAI,GAAA;AAIrD,UAAA,KAAA,CAAM,IAAA,GAAO,CAAC,CAAA,GAAI,EAAA,GAAK,YAAY,IAAA,GAAO,KAAA,CAAA,GAAS,KAAK,WAAA,GAAc,EAAA;AAAA,QACxE;AAAA,MACF;AAAA,IACF;AAEA,IAAA,OAAO,IAAA;AAAA,EACT;AACF;;;ACjPO,IAAM,KAAA,GAAN,cAAoB,SAAA,CAAoC;AAAA,EACrD,UAAA,GAAa,CAAA;AAAA,EAErB,IAAI,SAAA,GAAoB;AACtB,IAAA,OAAO,IAAA,CAAK,UAAA;AAAA,EACd;AAAA,EACA,WAAA,CACE,MAAA,EACA,OAAA,GAOI,EAAC,EACL;AACA,IAAA,MAAM,QAAA,GAAW;AAAA,MACf,EAAA,EAAI,QAAQ,EAAA,IAAM,IAAA;AAAA,MAClB,KAAA,EAAO,QAAQ,KAAA,IAAS,GAAA;AAAA,MACxB,KAAA,EAAO,QAAQ,KAAA,IAAS,KAAA;AAAA,MACxB,GAAA,EAAK,QAAQ,GAAA,IAAO,IAAA;AAAA,MACpB,WAAA,EAAa,QAAQ,WAAA,IAAe,CAAA;AAAA,MACpC,aAAA,EAAe,QAAQ,aAAA,IAAiB;AAAA,KAC1C;AAEA,IAAA,KAAA,CAAM,QAAQ,QAAQ,CAAA;AAGtB,IAAA,uBAAA,CAAwB,eAAA,EAAiB,SAAS,EAAE,CAAA;AACpD,IAAA,aAAA,CAAc,OAAA,EAAS,QAAA,CAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAC3C,IAAA,aAAA,CAAc,OAAA,EAAS,QAAA,CAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAC3C,IAAA,oBAAA,CAAqB,SAAA,EAAW,SAAS,GAAG,CAAA;AAC5C,IAAA,uBAAA,CAAwB,oBAAA,EAAsB,SAAS,WAAW,CAAA;AAClE,IAAA,uBAAA,CAAwB,gBAAA,EAAkB,SAAS,aAAa,CAAA;AAAA,EAClE;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,eAAA,CAAgB,WAAW,CAAA,EAAW;AACpC,IAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,WAAA,CAAY,QAAQ,CAAA;AACvC,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,CAAA,qBAAA,EAAwB,QAAQ,CAAA,mBAAA,EAAsB,IAAA,CAAK,YAAY,MAAM,CAAA,EAAA,CAAA;AAAA,QAC7E,UAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,OAAO,MAAM,OAAA,CAAQ,EAAA;AAAA,EACvB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,gBAAgB,EAAA,EAAkB;AAChC,IAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,KAAA,CAAM,QAAQ,EAAA,GAAK,EAAA;AAAA,IACrB;AAAA,EACF;AAAA,EAEU,QAAQ,KAAA,EAAqD;AACrE,IAAA,OACE,OAAO,KAAA,CAAM,MAAM,CAAA,KAAM,QAAA,IACzB,MAAM,QAAQ,CAAA,YAAa,YAAA,IAC3B,KAAA,CAAM,UAAU,CAAA,YAAa,YAAA,IAC7B,OAAO,KAAA,CAAM,WAAW,CAAA,KAAM,QAAA;AAAA,EAElC;AAAA,EAEA,KAAK,OAAA,EAA4C;AAC/C,IAAA,IAAI,IAAA;AAEJ,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,IAAA,GAAO,OAAA,EAAQ;AAAA,IACjB;AAGA,IAAA,IAAA,CAAK,UAAA,EAAA;AAEL,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,MAAM,EAAE,IAAI,KAAA,EAAO,KAAA,EAAO,KAAK,WAAA,EAAa,aAAA,KAAkB,KAAA,CAAM,OAAA;AAGpE,MAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,MAAA,aAAA,CAAc,OAAA,EAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAClC,MAAA,aAAA,CAAc,OAAA,EAAS,KAAA,EAAO,CAAA,EAAG,CAAC,CAAA;AAClC,MAAA,oBAAA,CAAqB,WAAW,GAAG,CAAA;AACnC,MAAA,uBAAA,CAAwB,sBAAsB,WAAW,CAAA;AACzD,MAAA,uBAAA,CAAwB,kBAAkB,aAAa,CAAA;AAEvD,MAAA,KAAA,MAAW,KAAA,IAAS,MAAM,MAAA,EAAQ;AAChC,QAAA,MAAM;AAAA,UACJ,IAAA,EAAM,QAAA;AAAA,UACN,UAAA,EAAY,IAAA;AAAA,UACZ,KAAA,EAAO,KAAA;AAAA,UACP,WAAA,EAAa;AAAA,SACf,GAAI,kBAAA,CAAmB,KAAA,EAAO,OAAO,CAAA;AACrC,QAAA,MAAM,IAAA,GAAO,MAAM,MAAA,CAAO,IAAA;AAG1B,QAAA,IAAI,KAAA,GAAQ,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAK,CAAA;AAChC,QAAA,IAAI,CAAC,KAAA,EAAO;AACV,UAAA,KAAA,GAAQ;AAAA,YACN,IAAA,EAAM,CAAA;AAAA,YACN,MAAA,EAAQ,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA,YAC7B,QAAA,EAAU,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA,YAC/B,SAAA,EAAW;AAAA,WACb;AACA,UAAA,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAA,EAAO,KAAK,CAAA;AAAA,QAC7B;AAGA,QAAA,gBAAA,CAAiB,KAAA,CAAM,MAAA,EAAQ,IAAA,EAAM,cAAc,CAAA;AACnD,QAAA,gBAAA,CAAiB,KAAA,CAAM,QAAA,EAAU,IAAA,EAAM,gBAAgB,CAAA;AAEvD,QAAA,KAAA,CAAM,IAAA,EAAA;AACN,QAAA,MAAM,IAAI,KAAA,CAAM,IAAA;AAEhB,QAAA,MAAM,eAAA,GAAkB,IAAI,KAAA,IAAS,CAAA;AACrC,QAAA,MAAM,EAAA,GAAK,KAAA,IAAS,CAAA,GAAI,GAAA,GAAM,SAAS,CAAA,GAAI,aAAA,CAAA,CAAA;AAC3C,QAAA,MAAM,SAAS,KAAA,IAAS,CAAA,GAAI,GAAA,GAAM,IAAA,KAAA,CAAU,IAAI,CAAA,IAAK,aAAA,CAAA,CAAA;AACrD,QAAA,MAAM,SAAA,GAAY,MAAM,SAAA,GAAY,EAAA;AACpC,QAAA,MAAM,gBAAgB,SAAA,GAAY,MAAA;AAClC,QAAA,KAAA,CAAM,SAAA,GAAY,SAAA;AAElB,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAC7B,UAAA,MAAM,GAAA,GAAM,eAAA,CAAgB,QAAA,EAAU,IAAA,GAAO,GAAG,gBAAgB,CAAA;AAChE,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,KAAA,EAAO,IAAA,GAAO,GAAG,iBAAiB,CAAA;AAC7D,UAAA,YAAA,CAAa,YAAY,GAAG,CAAA;AAC5B,UAAA,YAAA,CAAa,aAAa,EAAE,CAAA;AAG5B,UAAA,MAAM,EAAA,GAAK,WAAA,KAAgB,CAAA,GAAI,GAAA,GAAM,cAAc,EAAA,GAAK,GAAA;AAGxD,UAAA,MAAM,CAAA,GAAI,eAAA,CAAgB,KAAA,CAAM,MAAA,EAAQ,GAAG,cAAc,CAAA;AACzD,UAAA,MAAM,IAAA,GAAO,KAAA,GAAQ,CAAA,GAAA,CAAK,CAAA,GAAI,KAAA,IAAS,EAAA;AACvC,UAAA,KAAA,CAAM,MAAA,CAAO,CAAC,CAAA,GAAI,IAAA;AAGlB,UAAA,MAAM,CAAA,GAAI,eAAA,CAAgB,KAAA,CAAM,QAAA,EAAU,GAAG,gBAAgB,CAAA;AAC7D,UAAA,MAAM,IAAA,GAAO,KAAA,GAAQ,CAAA,GAAA,CAAK,CAAA,GAAI,SAAS,EAAA,GAAK,EAAA;AAC5C,UAAA,KAAA,CAAM,QAAA,CAAS,CAAC,CAAA,GAAI,IAAA;AAEpB,UAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,IAAA,CAAK,IAAA,GAAO,eAAe,CAAA,GAAI,GAAA;AAClD,UAAA,MAAM,QAAA,GAAW,QAAQ,CAAA,GAAI,aAAA,CAAA;AAC7B,UAAA,MAAM,IAAA,GAAO,MAAM,CAAA,GAAI,SAAA,CAAA;AACvB,UAAA,MAAM,SAAA,GAAY,MAAA,GAAS,QAAA,GAAA,CAAY,CAAA,GAAI,EAAA,IAAM,IAAA;AACjD,UAAA,KAAA,CAAM,IAAA,GAAO,CAAC,CAAA,GAAI,EAAA,GAAM,KAAK,SAAA,GAAa,KAAA;AAAA,QAC5C;AAAA,MACF;AAAA,IACF;AAEA,IAAA,OAAO,IAAA;AAAA,EACT;AACF;;;AClJO,IAAM,OAAA,GAAN,cAAsB,SAAA,CAAwC;AAAA;AAAA,EAE3D,UAAA,GAAa,CAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOrB,IAAI,SAAA,GAAoB;AACtB,IAAA,OAAO,IAAA,CAAK,UAAA;AAAA,EACd;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAcA,WAAA,CACE,MAAA,EACA,OAAA,GAOI,EAAC,EACL;AAEA,IAAA,MAAM,QAAA,GAAW;AAAA,MACf,EAAA,EAAI,QAAQ,EAAA,IAAM,IAAA;AAAA,MAClB,KAAA,EAAO,QAAQ,KAAA,IAAS,IAAA;AAAA,MACxB,GAAA,EAAK,QAAQ,GAAA,IAAO,IAAA;AAAA,MACpB,WAAA,EAAa,QAAQ,WAAA,IAAe,CAAA;AAAA,MACpC,QAAA,EAAU,QAAQ,QAAA,IAAY,CAAA;AAAA,MAC9B,QAAA,EAAU,QAAQ,QAAA,IAAY;AAAA,KAChC;AAEA,IAAA,KAAA,CAAM,QAAQ,QAAQ,CAAA;AAGtB,IAAA,uBAAA,CAAwB,eAAA,EAAiB,SAAS,EAAE,CAAA;AACpD,IAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,QAAA,CAAS,KAAK,CAAA,IAAK,QAAA,CAAS,KAAA,GAAQ,CAAA,IAAK,QAAA,CAAS,KAAA,GAAQ,CAAA,EAAG;AAChF,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,CAAA,eAAA,EAAkB,SAAS,KAAK,CAAA,0BAAA,CAAA;AAAA,QAChC,OAAA;AAAA,QACA,QAAA,CAAS;AAAA,OACX;AAAA,IACF;AACA,IAAA,oBAAA,CAAqB,SAAA,EAAW,SAAS,GAAG,CAAA;AAC5C,IAAA,uBAAA,CAAwB,oBAAA,EAAsB,SAAS,WAAW,CAAA;AAClE,IAAA,uBAAA,CAAwB,gBAAA,EAAkB,SAAS,QAAQ,CAAA;AAAA,EAC7D;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,eAAA,CAAgB,WAAW,CAAA,EAAW;AACpC,IAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,WAAA,CAAY,QAAQ,CAAA;AACvC,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,CAAA,qBAAA,EAAwB,QAAQ,CAAA,mBAAA,EAAsB,IAAA,CAAK,YAAY,MAAM,CAAA,EAAA,CAAA;AAAA,QAC7E,UAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,OAAO,MAAM,OAAA,CAAQ,EAAA;AAAA,EACvB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,gBAAgB,EAAA,EAAkB;AAChC,IAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,KAAA,CAAM,QAAQ,EAAA,GAAK,EAAA;AAAA,IACrB;AAAA,EACF;AAAA,EAEU,QAAQ,KAAA,EAAuD;AACvE,IAAA,IAAI,EAAE,KAAA,CAAM,WAAW,CAAA,YAAa,eAAe,OAAO,KAAA;AAC1D,IAAA,IACE,KAAA,CAAM,gBAAgB,CAAA,KAAM,MAAA,IAC5B,EAAE,KAAA,CAAM,gBAAgB,aAAa,YAAA,CAAA,EACrC;AACA,MAAA,OAAO,KAAA;AAAA,IACT;AACA,IAAA,IAAI,KAAA,CAAM,SAAS,CAAA,KAAM,MAAA,IAAa,EAAE,KAAA,CAAM,SAAS,aAAa,YAAA,CAAA,EAAe;AACjF,MAAA,OAAO,KAAA;AAAA,IACT;AACA,IAAA,OAAO,IAAA;AAAA,EACT;AAAA,EAEA,KAAK,OAAA,EAA4C;AAC/C,IAAA,IAAI,IAAA;AAGJ,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,IAAA,GAAO,OAAA,EAAQ;AAAA,IACjB;AAGA,IAAA,IAAA,CAAK,UAAA,EAAA;AAGL,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,MAAM,EAAE,IAAI,KAAA,EAAO,GAAA,EAAK,aAAa,QAAA,EAAU,QAAA,KAAa,KAAA,CAAM,OAAA;AAGlE,MAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,MAAA,IAAI,CAAC,OAAO,QAAA,CAAS,KAAK,KAAK,KAAA,GAAQ,CAAA,IAAK,QAAQ,CAAA,EAAG;AACrD,QAAA,MAAM,IAAI,qBAAA;AAAA,UACR,kBAAkB,KAAK,CAAA,0BAAA,CAAA;AAAA,UACvB,OAAA;AAAA,UACA;AAAA,SACF;AAAA,MACF;AACA,MAAA,oBAAA,CAAqB,WAAW,GAAG,CAAA;AACnC,MAAA,uBAAA,CAAwB,sBAAsB,WAAW,CAAA;AACzD,MAAA,uBAAA,CAAwB,kBAAkB,QAAQ,CAAA;AAGlD,MAAA,KAAA,MAAW,KAAA,IAAS,MAAM,MAAA,EAAQ;AAEhC,QAAA,MAAM;AAAA,UACJ,IAAA,EAAM,QAAA;AAAA,UACN,UAAA,EAAY,IAAA;AAAA,UACZ,KAAA,EAAO,KAAA;AAAA,UACP,WAAA,EAAa;AAAA,SACf,GAAI,kBAAA,CAAmB,KAAA,EAAO,SAAS,CAAA;AACvC,QAAA,MAAM,IAAA,GAAO,MAAM,MAAA,CAAO,IAAA;AAG1B,QAAA,IAAI,KAAA,GAAQ,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAK,CAAA;AAChC,QAAA,IAAI,CAAC,KAAA,EAAO;AACV,UAAA,KAAA,GAAQ;AAAA,YACN,SAAA,EAAW,IAAI,YAAA,CAAa,IAAI;AAAA,WAClC;AACA,UAAA,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAA,EAAO,KAAK,CAAA;AAAA,QAC7B;AAGA,QAAA,IAAI,QAAA,GAAW,CAAA,IAAK,CAAC,KAAA,CAAM,cAAA,EAAgB;AACzC,UAAA,KAAA,CAAM,cAAA,GAAiB,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA,QAC9C;AAGA,QAAA,IAAI,QAAA,IAAY,CAAC,KAAA,CAAM,OAAA,EAAS;AAC9B,UAAA,KAAA,CAAM,OAAA,GAAU,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA,QACvC;AAGA,QAAA,gBAAA,CAAiB,KAAA,CAAM,SAAA,EAAW,IAAA,EAAM,mBAAmB,CAAA;AAC3D,QAAA,IAAI,QAAA,GAAW,CAAA,IAAK,KAAA,CAAM,cAAA,EAAgB;AACxC,UAAA,gBAAA,CAAiB,KAAA,CAAM,cAAA,EAAgB,IAAA,EAAM,wBAAwB,CAAA;AAAA,QACvE;AACA,QAAA,IAAI,QAAA,IAAY,MAAM,OAAA,EAAS;AAC7B,UAAA,gBAAA,CAAiB,KAAA,CAAM,OAAA,EAAS,IAAA,EAAM,iBAAiB,CAAA;AAAA,QACzD;AAGA,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAE7B,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,QAAA,EAAU,IAAA,GAAO,GAAG,kBAAkB,CAAA;AACjE,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,KAAA,EAAO,IAAA,GAAO,GAAG,mBAAmB,CAAA;AAG/D,UAAA,YAAA,CAAa,YAAY,EAAE,CAAA;AAC3B,UAAA,YAAA,CAAa,aAAa,EAAE,CAAA;AAG5B,UAAA,IAAI,IAAA,GAAO,EAAA;AACX,UAAA,IAAI,gBAAgB,CAAA,EAAG;AACrB,YAAA,IAAA,GAAO,OAAO,WAAA,GAAc,EAAA;AAAA,UAC9B;AAIA,UAAA,MAAM,KAAA,GAAQ,eAAA,CAAgB,KAAA,CAAM,SAAA,EAAW,GAAG,mBAAmB,CAAA;AACrE,UAAA,MAAM,QAAA,GAAW,KAAA,GAAQ,KAAA,GAAA,CAAS,CAAA,GAAI,SAAS,IAAA,GAAO,IAAA;AACtD,UAAA,KAAA,CAAM,SAAA,CAAU,CAAC,CAAA,GAAI,QAAA;AAGrB,UAAA,IAAI,GAAA,GAAM,QAAA;AAGV,UAAA,IAAI,QAAA,EAAU;AACZ,YAAA,MAAM,IAAA,GAAO,MAAM,OAAA,GAAU,eAAA,CAAgB,MAAM,OAAA,EAAS,CAAA,EAAG,iBAAiB,CAAA,GAAI,CAAA;AAEpF,YAAA,MAAM,OAAA,GAAU,KAAA,GAAQ,IAAA,GAAA,CAAQ,CAAA,GAAI,KAAA,IAAS,IAAA;AAC7C,YAAA,IAAI,KAAA,CAAM,OAAA,EAAS,KAAA,CAAM,OAAA,CAAQ,CAAC,CAAA,GAAI,OAAA;AAEtC,YAAA,GAAA,GAAM,WAAW,OAAA,GAAU,OAAA;AAAA,UAC7B;AAEA,UAAA,MAAM,KAAA,GAAQ,QAAA,GAAW,IAAA,CAAK,IAAA,CAAK,KAAK,GAAA,CAAI,GAAA,EAAK,CAAC,CAAA,GAAI,GAAG,CAAA,GAAI,IAAA,CAAK,IAAA,CAAK,GAAG,CAAA,GAAI,GAAA;AAC9E,UAAA,MAAM,iBAAiB,IAAA,GAAO,KAAA;AAG9B,UAAA,IAAI,WAAW,CAAA,EAAG;AAChB,YAAA,MAAM,GAAA,GAAM,MAAM,cAAA,GACd,eAAA,CAAgB,MAAM,cAAA,EAAgB,CAAA,EAAG,wBAAwB,CAAA,GACjE,CAAA;AACJ,YAAA,MAAM,MAAA,GAAS,WAAW,GAAA,GAAM,cAAA;AAChC,YAAA,IAAI,KAAA,CAAM,cAAA,EAAgB,KAAA,CAAM,cAAA,CAAe,CAAC,CAAA,GAAI,MAAA;AAEpD,YAAA,KAAA,CAAM,IAAA,GAAO,CAAC,CAAA,GAAI,EAAA,GAAK,EAAA,GAAK,MAAA;AAAA,UAC9B,CAAA,MAAO;AAEL,YAAA,KAAA,CAAM,IAAA,GAAO,CAAC,CAAA,GAAI,EAAA,GAAK,EAAA,GAAK,cAAA;AAAA,UAC9B;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAEA,IAAA,OAAO,IAAA;AAAA,EACT;AACF;;;ACxPO,IAAM,GAAA,GAAN,cAAkB,SAAA,CAAgC;AAAA;AAAA,EAE/C,UAAA,GAAa,CAAA;AAAA,EAErB,IAAI,SAAA,GAAoB;AACtB,IAAA,OAAO,IAAA,CAAK,UAAA;AAAA,EACd;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAYA,WAAA,CACE,MAAA,EACA,OAAA,GAMI,EAAC,EACL;AACA,IAAA,MAAM,QAAA,GAAW;AAAA,MACf,EAAA,EAAI,QAAQ,EAAA,IAAM,IAAA;AAAA,MAClB,QAAA,EAAU,QAAQ,QAAA,IAAY,CAAA;AAAA,MAC9B,SAAA,EAAW,QAAQ,SAAA,IAAa,CAAA;AAAA,MAChC,WAAA,EAAa,QAAQ,WAAA,IAAe,CAAA;AAAA,MACpC,QAAA,EAAU,QAAQ,QAAA,IAAY;AAAA,KAChC;AAEA,IAAA,KAAA,CAAM,QAAQ,QAAQ,CAAA;AAGtB,IAAA,uBAAA,CAAwB,eAAA,EAAiB,SAAS,EAAE,CAAA;AACpD,IAAA,uBAAA,CAAwB,gBAAA,EAAkB,SAAS,QAAQ,CAAA;AAC3D,IAAA,uBAAA,CAAwB,WAAA,EAAa,SAAS,SAAS,CAAA;AACvD,IAAA,uBAAA,CAAwB,oBAAA,EAAsB,SAAS,WAAW,CAAA;AAClE,IAAA,IAAI,SAAS,QAAA,KAAa,QAAA,CAAS,YAAY,CAAA,IAAK,QAAA,CAAS,cAAc,CAAA,CAAA,EAAI;AAC7E,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,0DAAA;AAAA,QACA,UAAA;AAAA,QACA;AAAA,UACE,UAAU,QAAA,CAAS,QAAA;AAAA,UACnB,WAAW,QAAA,CAAS,SAAA;AAAA,UACpB,UAAU,QAAA,CAAS;AAAA;AACrB,OACF;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUU,QAAQ,KAAA,EAAmD;AACnE,IAAA,IACE,KAAA,CAAM,gBAAgB,CAAA,KAAM,MAAA,IAC5B,EAAE,KAAA,CAAM,gBAAgB,aAAa,YAAA,CAAA,EACrC;AACA,MAAA,OAAO,KAAA;AAAA,IACT;AACA,IAAA,OAAO,IAAA;AAAA,EACT;AAAA,EAEA,KAAK,OAAA,EAA4C;AAC/C,IAAA,IAAI,IAAA;AAGJ,IAAA,IAAI,OAAA,EAAS;AACX,MAAA,IAAA,GAAO,OAAA,EAAQ;AAAA,IACjB;AAGA,IAAA,IAAA,CAAK,UAAA,EAAA;AAGL,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,MAAM,EAAE,EAAA,EAAI,QAAA,EAAU,WAAW,WAAA,EAAa,QAAA,KAAa,KAAA,CAAM,OAAA;AAEjE,MAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,MAAA,uBAAA,CAAwB,kBAAkB,QAAQ,CAAA;AAClD,MAAA,uBAAA,CAAwB,aAAa,SAAS,CAAA;AAC9C,MAAA,uBAAA,CAAwB,sBAAsB,WAAW,CAAA;AAEzD,MAAA,IAAI,QAAA,KAAa,QAAA,IAAY,CAAA,IAAK,SAAA,KAAc,CAAA,CAAA,EAAI;AAClD,QAAA,MAAM,IAAI,qBAAA;AAAA,UACR,0DAAA;AAAA,UACA,UAAA;AAAA,UACA,EAAE,QAAA,EAAU,SAAA,EAAW,QAAA;AAAS,SAClC;AAAA,MACF;AAEA,MAAA,KAAA,MAAW,KAAA,IAAS,MAAM,MAAA,EAAQ;AAChC,QAAA,MAAM;AAAA,UACJ,IAAA,EAAM,QAAA;AAAA,UACN,UAAA;AAAA,UACA,KAAA,EAAO,SAAA;AAAA,UACP;AAAA,SACF,GAAI,kBAAA,CAAmB,KAAA,EAAO,KAAK,CAAA;AACnC,QAAA,MAAM,IAAA,GAAO,MAAM,MAAA,CAAO,IAAA;AAE1B,QAAA,IAAI,KAAA,GAAQ,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAK,CAAA;AAChC,QAAA,IAAI,CAAC,KAAA,EAAO;AACV,UAAA,KAAA,GAAQ,EAAC;AACT,UAAA,IAAA,CAAK,KAAA,CAAM,GAAA,CAAI,KAAA,EAAO,KAAK,CAAA;AAAA,QAC7B;AAGA,QAAA,IAAI,cAAA;AACJ,QAAA,IAAI,aAAa,CAAA,EAAG;AAClB,UAAA,IAAI,CAAC,MAAM,cAAA,EAAgB;AACzB,YAAA,KAAA,CAAM,cAAA,GAAiB,IAAI,YAAA,CAAa,IAAI,CAAA;AAAA,UAC9C;AACA,UAAA,cAAA,GAAiB,KAAA,CAAM,cAAA;AAAA,QACzB;AAEA,QAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,IAAA,EAAM,CAAA,EAAA,EAAK;AAC7B,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,QAAA,EAAU,UAAA,GAAa,GAAG,cAAc,CAAA;AACnE,UAAA,MAAM,EAAA,GAAK,eAAA,CAAgB,SAAA,EAAW,WAAA,GAAc,GAAG,eAAe,CAAA;AACtE,UAAA,YAAA,CAAa,YAAY,EAAE,CAAA;AAC3B,UAAA,YAAA,CAAa,aAAa,EAAE,CAAA;AAG5B,UAAA,IAAI,CAAA,GAAI,EAAA;AACR,UAAA,IAAI,gBAAgB,CAAA,EAAG;AACrB,YAAA,CAAA,GAAI,IAAI,WAAA,GAAc,EAAA;AAAA,UACxB;AAEA,UAAA,IAAI,cAAA,EAAgB;AAClB,YAAA,MAAM,KAAA,GAAQ,eAAA,CAAgB,cAAA,EAAgB,CAAA,EAAG,qBAAqB,CAAA;AACtE,YAAA,MAAM,IAAA,GAAO,QAAA,GAAW,KAAA,GAAA,CAAS,CAAA,GAAI,SAAA,IAAa,CAAA;AAClD,YAAA,cAAA,CAAe,CAAC,CAAA,GAAI,IAAA;AACpB,YAAA,CAAA,GAAI,QAAA,GAAW,CAAA,GAAI,QAAA,GAAW,IAAA,GAAO,IAAA;AAAA,UACvC;AAGA,UAAA,SAAA,CAAU,WAAA,GAAc,CAAC,CAAA,GAAI,EAAA,GAAK,EAAA,GAAK,CAAA;AAAA,QACzC;AAAA,MACF;AAAA,IACF;AAEA,IAAA,OAAO,IAAA;AAAA,EACT;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,eAAA,CAAgB,WAAW,CAAA,EAAW;AACpC,IAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,WAAA,CAAY,QAAQ,CAAA;AACvC,IAAA,IAAI,CAAC,KAAA,EAAO;AACV,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,CAAA,qBAAA,EAAwB,QAAQ,CAAA,mBAAA,EAAsB,IAAA,CAAK,YAAY,MAAM,CAAA,EAAA,CAAA;AAAA,QAC7E,UAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,OAAO,MAAM,OAAA,CAAQ,EAAA;AAAA,EACvB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,gBAAgB,EAAA,EAAkB;AAChC,IAAA,uBAAA,CAAwB,iBAAiB,EAAE,CAAA;AAC3C,IAAA,KAAA,MAAW,KAAA,IAAS,KAAK,WAAA,EAAa;AACpC,MAAA,KAAA,CAAM,QAAQ,EAAA,GAAK,EAAA;AAAA,IACrB;AAAA,EACF;AACF;;;AC3NA,SAASA,UAAS,KAAA,EAAkD;AAClE,EAAA,OAAO,OAAO,KAAA,KAAU,QAAA,IAAY,KAAA,KAAU,IAAA;AAChD;AAEA,SAAS,cAAA,CAAe,OAA4B,KAAA,EAAe;AACjE,EAAA,MAAM,UAAUA,SAAAA,CAAS,KAAA,CAAM,OAAO,CAAA,GAAI,MAAM,OAAA,GAAU,MAAA;AAC1D,EAAA,MAAM,OAAA,GAAU,KAAA,CAAM,EAAA,IAAM,OAAA,GAAU,IAAI,CAAA;AAC1C,EAAA,IAAI,OAAO,YAAY,QAAA,IAAY,CAAC,OAAO,QAAA,CAAS,OAAO,CAAA,IAAK,OAAA,GAAU,CAAA,EAAG;AAC3E,IAAA,MAAM,IAAI,qBAAA;AAAA,MACR,yBAAyB,KAAK,CAAA,4BAAA,CAAA;AAAA,MAC9B,yBAAyB,KAAK,CAAA,IAAA,CAAA;AAAA,MAC9B;AAAA,KACF;AAAA,EACF;AACA,EAAA,OAAO,OAAA;AACT;AAEA,SAAS,UAAA,CAAW,OAA4B,EAAA,EAAY;AAC1D,EAAA,IAAIA,SAAAA,CAAS,KAAA,CAAM,OAAO,CAAA,EAAG;AAC3B,IAAA,KAAA,CAAM,OAAA,CAAQ,IAAI,CAAA,GAAI,EAAA;AAAA,EACxB;AACA,EAAA,IAAI,QAAQ,KAAA,EAAO;AACjB,IAAA,KAAA,CAAM,EAAA,GAAK,EAAA;AAAA,EACb;AACA,EAAA,IAAI,EAAE,IAAA,IAAQ,KAAA,CAAA,IAAU,CAACA,SAAAA,CAAS,KAAA,CAAM,OAAO,CAAA,EAAG;AAChD,IAAA,KAAA,CAAM,EAAA,GAAK,EAAA;AAAA,EACb;AACF;AAEA,SAAS,kBAAkB,KAAA,EAAe;AACxC,EAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,KAAK,CAAA,IAAK,QAAQ,EAAA,EAAI;AAC1C,IAAA,MAAM,IAAI,qBAAA,CAAsB,oCAAA,EAAsC,WAAA,EAAa,KAAK,CAAA;AAAA,EAC1F;AACA,EAAA,OAAO,KAAA;AACT;AAEA,SAAS,oBAAA,CAAqB,OAAe,IAAA,EAAc;AACzD,EAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,KAAK,CAAA,EAAG;AAC3B,IAAA,MAAM,IAAI,qBAAA,CAAsB,CAAA,EAAG,IAAI,CAAA,eAAA,CAAA,EAAmB,MAAM,KAAK,CAAA;AAAA,EACvE;AACA,EAAA,OAAO,KAAA;AACT;AAEA,SAAS,sBAAA,CAAuB,OAAe,IAAA,EAAc;AAC3D,EAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,KAAK,CAAA,IAAK,SAAS,CAAA,EAAG;AACzC,IAAA,MAAM,IAAI,qBAAA,CAAsB,CAAA,EAAG,IAAI,CAAA,YAAA,CAAA,EAAgB,MAAM,KAAK,CAAA;AAAA,EACpE;AACA,EAAA,OAAO,KAAA;AACT;AAEA,SAAS,uBAAA,CAAwB,OAAe,IAAA,EAAc;AAC5D,EAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,KAAK,CAAA,IAAK,SAAS,CAAA,EAAG;AAC1C,IAAA,MAAM,IAAI,qBAAA,CAAsB,CAAA,EAAG,IAAI,CAAA,2BAAA,CAAA,EAA+B,MAAM,KAAK,CAAA;AAAA,EACnF;AACA,EAAA,OAAO,KAAA;AACT;AAEA,SAAS,yBAAA,CAA0B,OAAe,IAAA,EAAc;AAC9D,EAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,KAAK,CAAA,IAAK,QAAQ,CAAA,EAAG;AACxC,IAAA,MAAM,IAAI,qBAAA,CAAsB,CAAA,EAAG,IAAI,CAAA,aAAA,CAAA,EAAiB,MAAM,KAAK,CAAA;AAAA,EACrE;AACA,EAAA,OAAO,KAAA;AACT;AAEA,SAAS,0BAAA,CAA2B,OAAe,IAAA,EAAc;AAC/D,EAAA,IAAI,CAAC,MAAA,CAAO,SAAA,CAAU,KAAK,CAAA,IAAK,QAAQ,CAAA,EAAG;AACzC,IAAA,MAAM,IAAI,qBAAA,CAAsB,CAAA,EAAG,IAAI,CAAA,+BAAA,CAAA,EAAmC,MAAM,KAAK,CAAA;AAAA,EACvF;AACA,EAAA,OAAO,KAAA;AACT;AAEA,SAAS,kBAAkB,SAAA,EAA+B;AACxD,EAAA,IAAI,CAAC,SAAA,IAAa,OAAO,SAAA,KAAc,QAAA,IAAY,CAAC,KAAA,CAAM,OAAA,CAAQ,SAAA,CAAU,WAAW,CAAA,EAAG;AACxF,IAAA,MAAM,IAAI,qBAAA;AAAA,MACR,yCAAA;AAAA,MACA,WAAA;AAAA,MACA;AAAA,KACF;AAAA,EACF;AACA,EAAA,IAAI,SAAA,CAAU,WAAA,CAAY,MAAA,KAAW,CAAA,EAAG;AACtC,IAAA,MAAM,IAAI,qBAAA;AAAA,MACR,uDAAA;AAAA,MACA,uBAAA;AAAA,MACA,SAAA,CAAU;AAAA,KACZ;AAAA,EACF;AACA,EAAA,KAAA,IAAS,IAAI,CAAA,EAAG,CAAA,GAAI,SAAA,CAAU,WAAA,CAAY,QAAQ,CAAA,EAAA,EAAK;AACrD,IAAA,MAAM,KAAA,GAAQ,SAAA,CAAU,WAAA,CAAY,CAAC,CAAA;AACrC,IAAA,IAAI,CAAC,KAAA,IAAS,OAAO,KAAA,KAAU,QAAA,EAAU;AACvC,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,yBAAyB,CAAC,CAAA,mBAAA,CAAA;AAAA,QAC1B,uBAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,IAAI,CAAC,KAAA,CAAM,OAAA,CAAQ,KAAA,CAAM,MAAM,CAAA,EAAG;AAChC,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,yBAAyB,CAAC,CAAA,yBAAA,CAAA;AAAA,QAC1B,yBAAyB,CAAC,CAAA,QAAA,CAAA;AAAA,QAC1B,KAAA,CAAM;AAAA,OACR;AAAA,IACF;AACA,IAAA,cAAA,CAAe,OAAO,CAAC,CAAA;AAAA,EACzB;AACF;AAEA,SAAS,mBAAmB,UAAA,EAAsB;AAChD,EAAA,IAAI,CAAC,KAAA,CAAM,OAAA,CAAQ,UAAU,CAAA,IAAK,UAAA,CAAW,WAAW,CAAA,EAAG;AACzD,IAAA,MAAM,IAAI,qBAAA;AAAA,MACR,+DAAA;AAAA,MACA,YAAA;AAAA,MACA;AAAA,KACF;AAAA,EACF;AAEA,EAAA,MAAM,MAAA,GAAS,CAAC,GAAG,UAAU,CAAA,CAAE,KAAK,CAAC,CAAA,EAAG,CAAA,KAAM,CAAA,GAAI,CAAC,CAAA;AACnD,EAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,GAAI,MAAA,CAAO,QAAQ,CAAA,EAAA,EAAK;AACtC,IAAA,MAAM,KAAA,GAAQ,OAAO,CAAC,CAAA;AACtB,IAAA,IAAI,KAAA,KAAU,UAAa,CAAC,MAAA,CAAO,UAAU,KAAK,CAAA,IAAK,QAAQ,CAAA,EAAG;AAChE,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,oDAAA;AAAA,QACA,YAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,IAAI,IAAI,CAAA,EAAG;AACT,MAAA,MAAM,IAAA,GAAO,MAAA,CAAO,CAAA,GAAI,CAAC,CAAA;AACzB,MAAA,IAAI,IAAA,KAAS,MAAA,IAAa,KAAA,IAAS,IAAA,EAAM;AACvC,QAAA,MAAM,IAAI,qBAAA;AAAA,UACR,wCAAA;AAAA,UACA,YAAA;AAAA,UACA;AAAA,SACF;AAAA,MACF;AAAA,IACF;AAAA,EACF;AACA,EAAA,OAAO,MAAA;AACT;AAwBO,IAAe,cAAf,MAA2B;AAAA,EACtB,SAAA;AAAA,EACA,SAAA;AAAA,EACA,OAAA;AAAA,EAEV,WAAA,CAAY,SAAA,EAA+B,SAAA,GAAoB,EAAA,EAAI;AACjE,IAAA,iBAAA,CAAkB,SAAS,CAAA;AAC3B,IAAA,IAAA,CAAK,SAAA,GAAY,kBAAkB,SAAS,CAAA;AAC5C,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AAGjB,IAAA,IAAA,CAAK,OAAA,GAAU,SAAA,CAAU,WAAA,CAAY,GAAA,CAAI,CAAC,OAAO,KAAA,KAAU,cAAA,CAAe,KAAA,EAAO,KAAK,CAAC,CAAA;AAAA,EACzF;AAAA,EAEU,wBAAwB,SAAA,EAAyB;AACzD,IAAA,MAAM,SAAA,GAAY,kBAAkB,SAAS,CAAA;AAC7C,IAAA,IAAI,YAAY,CAAA,EAAG;AACjB,MAAA;AAAA,IACF;AACA,IAAA,IAAA,CAAK,SAAA,GAAY,EAAA;AACjB,IAAA,KAAA,IAAS,CAAA,GAAI,CAAA,EAAG,CAAA,IAAK,SAAA,EAAW,CAAA,EAAA,EAAK;AACnC,MAAA,IAAA,CAAK,IAAA,EAAK;AAAA,IACZ;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAeA,IAAA,GAAa;AACX,IAAA,IAAA,CAAK,SAAA,EAAA;AACL,IAAA,MAAM,MAAA,GAAS,KAAK,KAAA,EAAM;AAE1B,IAAA,KAAA,IAAS,IAAI,CAAA,EAAG,CAAA,GAAI,KAAK,SAAA,CAAU,WAAA,CAAY,QAAQ,CAAA,EAAA,EAAK;AAC1D,MAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,SAAA,CAAU,WAAA,CAAY,CAAC,CAAA;AAC1C,MAAA,IAAI,KAAA,EAAO;AACT,QAAA,MAAM,IAAA,GAAO,OAAO,CAAC,CAAA;AACrB,QAAA,IAAI,SAAS,MAAA,EAAW;AACtB,UAAA,UAAA,CAAW,OAAO,IAAI,CAAA;AAAA,QACxB;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,SAAA,GAAsB;AACpB,IAAA,OAAO,IAAA,CAAK,SAAA,CAAU,WAAA,CAAY,GAAA,CAAI,CAAC,OAAO,KAAA,KAAU,cAAA,CAAe,KAAA,EAAO,KAAK,CAAC,CAAA;AAAA,EACtF;AAAA;AAAA;AAAA;AAAA,EAKA,IAAI,KAAA,GAAQ;AACV,IAAA,OAAO,IAAA,CAAK,SAAA;AAAA,EACd;AACF;AAkBO,IAAM,MAAA,GAAN,cAAqB,WAAA,CAAY;AAAA,EAC9B,QAAA;AAAA,EACA,KAAA;AAAA,EAER,WAAA,CACE,WACA,OAAA,EACA;AACA,IAAA,MAAM,QAAA,GAAW,uBAAA,CAAwB,OAAA,CAAQ,QAAA,EAAU,UAAU,CAAA;AACrE,IAAA,MAAM,KAAA,GAAQ,sBAAA,CAAuB,OAAA,CAAQ,KAAA,IAAS,KAAK,OAAO,CAAA;AAClE,IAAA,MAAM,SAAA,GAAY,iBAAA,CAAkB,OAAA,CAAQ,SAAA,IAAa,EAAE,CAAA;AAC3D,IAAA,KAAA,CAAM,WAAW,EAAE,CAAA;AACnB,IAAA,IAAA,CAAK,QAAA,GAAW,QAAA;AAChB,IAAA,IAAA,CAAK,KAAA,GAAQ,KAAA;AACb,IAAA,IAAA,CAAK,wBAAwB,SAAS,CAAA;AAAA,EACxC;AAAA,EAEA,KAAA,GAAkB;AAChB,IAAA,MAAM,MAAA,GAAS,KAAK,KAAA,IAAS,IAAA,CAAK,MAAM,IAAA,CAAK,SAAA,GAAY,KAAK,QAAQ,CAAA;AACtE,IAAA,OAAO,KAAK,OAAA,CAAQ,GAAA,CAAI,CAAC,EAAA,KAAO,KAAK,MAAM,CAAA;AAAA,EAC7C;AACF;AAgBO,IAAM,aAAA,GAAN,cAA4B,WAAA,CAAY;AAAA,EACrC,KAAA;AAAA,EAER,WAAA,CAAY,WAA+B,OAAA,EAAgD;AACzF,IAAA,MAAM,KAAA,GAAQ,sBAAA,CAAuB,OAAA,CAAQ,KAAA,EAAO,OAAO,CAAA;AAC3D,IAAA,MAAM,SAAA,GAAY,iBAAA,CAAkB,OAAA,CAAQ,SAAA,IAAa,EAAE,CAAA;AAC3D,IAAA,KAAA,CAAM,WAAW,EAAE,CAAA;AACnB,IAAA,IAAA,CAAK,KAAA,GAAQ,KAAA;AACb,IAAA,IAAA,CAAK,wBAAwB,SAAS,CAAA;AAAA,EACxC;AAAA,EAEA,KAAA,GAAkB;AAChB,IAAA,OAAO,IAAA,CAAK,QAAQ,GAAA,CAAI,CAAC,OAAO,EAAA,GAAK,IAAA,CAAK,KAAA,IAAS,IAAA,CAAK,SAAS,CAAA;AAAA,EACnE;AACF;AAeO,IAAM,iBAAA,GAAN,cAAgC,WAAA,CAAY;AAAA,EACzC,KAAA;AAAA,EACA,MAAA;AAAA,EAER,WAAA,CACE,WACA,OAAA,EACA;AACA,IAAA,MAAM,IAAA,GAAO,uBAAA,CAAwB,OAAA,CAAQ,KAAA,EAAO,OAAO,CAAA;AAC3D,IAAA,MAAM,MAAA,GAAS,yBAAA,CAA0B,OAAA,CAAQ,MAAA,IAAU,GAAG,QAAQ,CAAA;AACtE,IAAA,MAAM,SAAA,GAAY,iBAAA,CAAkB,OAAA,CAAQ,SAAA,IAAa,EAAE,CAAA;AAC3D,IAAA,KAAA,CAAM,WAAW,EAAE,CAAA;AACnB,IAAA,IAAA,CAAK,KAAA,GAAQ,IAAA;AACb,IAAA,IAAA,CAAK,MAAA,GAAS,MAAA;AACd,IAAA,IAAA,CAAK,wBAAwB,SAAS,CAAA;AAAA,EACxC;AAAA,EAEA,KAAA,GAAkB;AAChB,IAAA,OAAO,IAAA,CAAK,OAAA,CAAQ,GAAA,CAAI,CAAC,MAAA,KAAW;AAClC,MAAA,OACE,IAAA,CAAK,MAAA,GAAA,CACH,MAAA,GAAS,IAAA,CAAK,WAAW,CAAA,GAAI,IAAA,CAAK,GAAA,CAAK,IAAA,CAAK,EAAA,GAAK,IAAA,CAAK,SAAA,GAAa,IAAA,CAAK,KAAK,CAAA,CAAA,GAAM,CAAA;AAAA,IAEzF,CAAC,CAAA;AAAA,EACH;AACF;AAiBO,IAAM,WAAA,GAAN,cAA0B,WAAA,CAAY;AAAA,EACnC,gBAAA;AAAA,EACA,KAAA;AAAA,EAER,WAAA,CACE,WACA,OAAA,EACA;AACA,IAAA,MAAM,UAAA,GAAa,kBAAA,CAAmB,OAAA,CAAQ,UAAU,CAAA;AACxD,IAAA,MAAM,KAAA,GAAQ,sBAAA,CAAuB,OAAA,CAAQ,KAAA,IAAS,KAAK,OAAO,CAAA;AAClE,IAAA,MAAM,SAAA,GAAY,iBAAA,CAAkB,OAAA,CAAQ,SAAA,IAAa,EAAE,CAAA;AAC3D,IAAA,KAAA,CAAM,WAAW,EAAE,CAAA;AACnB,IAAA,IAAA,CAAK,gBAAA,GAAmB,UAAA;AACxB,IAAA,IAAA,CAAK,KAAA,GAAQ,KAAA;AACb,IAAA,IAAA,CAAK,wBAAwB,SAAS,CAAA;AAAA,EACxC;AAAA,EAEA,KAAA,GAAkB;AAEhB,IAAA,IAAI,SAAA,GAAY,CAAA;AAChB,IAAA,KAAA,MAAW,SAAA,IAAa,KAAK,gBAAA,EAAkB;AAC7C,MAAA,IAAI,IAAA,CAAK,aAAa,SAAA,EAAW;AAC/B,QAAA,SAAA,EAAA;AAAA,MACF;AAAA,IACF;AACA,IAAA,MAAM,MAAA,GAAS,KAAK,KAAA,IAAS,SAAA;AAC7B,IAAA,OAAO,KAAK,OAAA,CAAQ,GAAA,CAAI,CAAC,EAAA,KAAO,KAAK,MAAM,CAAA;AAAA,EAC7C;AACF;AAqBO,IAAM,QAAA,GAAN,cAAuB,WAAA,CAAY;AAAA,EAChC,WAAA;AAAA,EACA,SAAA;AAAA,EACA,UAAA;AAAA,EAER,WAAA,CACE,WACA,OAAA,EAMA;AACA,IAAA,MAAM,cAAc,sBAAA,CAAuB,OAAA,CAAQ,WAAA,IAAe,CAAA,GAAI,GAAG,aAAa,CAAA;AACtF,IAAA,MAAM,SAAA,GAAY,sBAAA,CAAuB,OAAA,CAAQ,SAAA,IAAa,GAAK,WAAW,CAAA;AAC9E,IAAA,MAAM,UAAA,GAAa,uBAAA,CAAwB,OAAA,CAAQ,UAAA,EAAY,YAAY,CAAA;AAC3E,IAAA,MAAM,SAAA,GAAY,iBAAA,CAAkB,OAAA,CAAQ,SAAA,IAAa,EAAE,CAAA;AAC3D,IAAA,KAAA,CAAM,WAAW,EAAE,CAAA;AACnB,IAAA,IAAA,CAAK,WAAA,GAAc,WAAA;AACnB,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AACjB,IAAA,IAAA,CAAK,UAAA,GAAa,UAAA;AAClB,IAAA,IAAA,CAAK,wBAAwB,SAAS,CAAA;AAAA,EACxC;AAAA,EAEA,KAAA,GAAkB;AAChB,IAAA,IAAI,IAAA,CAAK,SAAA,IAAa,IAAA,CAAK,UAAA,EAAY;AACrC,MAAA,OAAO,KAAK,OAAA,CAAQ,GAAA,CAAI,CAAC,EAAA,KAAO,EAAA,GAAK,KAAK,SAAS,CAAA;AAAA,IACrD;AAEA,IAAA,MAAM,MAAA,GACJ,KAAK,WAAA,GAAA,CAAe,IAAA,CAAK,YAAY,IAAA,CAAK,WAAA,KAAgB,IAAA,CAAK,SAAA,GAAY,IAAA,CAAK,UAAA,CAAA;AAClF,IAAA,OAAO,KAAK,OAAA,CAAQ,GAAA,CAAI,CAAC,EAAA,KAAO,KAAK,MAAM,CAAA;AAAA,EAC7C;AACF;AAyBO,IAAM,oBAAN,MAAwB;AAAA,EACrB,SAAA;AAAA,EACA,IAAA;AAAA,EACA,MAAA;AAAA,EACA,QAAA;AAAA,EACA,SAAA;AAAA,EACA,QAAA;AAAA,EACA,KAAA;AAAA,EACA,IAAA;AAAA,EACA,YAAA;AAAA,EACA,eAAA;AAAA,EAER,WAAA,CACE,SAAA,EACA,OAAA,GAOI,EAAC,EACL;AACA,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AACjB,IAAA,iBAAA,CAAkB,SAAS,CAAA;AAC3B,IAAA,IAAA,CAAK,IAAA,GAAO,QAAQ,IAAA,IAAQ,KAAA;AAC5B,IAAA,IAAI,IAAA,CAAK,IAAA,KAAS,KAAA,IAAS,IAAA,CAAK,SAAS,KAAA,EAAO;AAC9C,MAAA,MAAM,IAAI,qBAAA,CAAsB,6BAAA,EAA+B,MAAA,EAAQ,QAAQ,IAAI,CAAA;AAAA,IACrF;AACA,IAAA,IAAA,CAAK,MAAA,GAAS,oBAAA,CAAqB,OAAA,CAAQ,MAAA,IAAU,KAAK,QAAQ,CAAA;AAClE,IAAA,IAAI,IAAA,CAAK,MAAA,IAAU,CAAA,IAAK,IAAA,CAAK,UAAU,CAAA,EAAG;AACxC,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,uCAAA;AAAA,QACA,QAAA;AAAA,QACA,IAAA,CAAK;AAAA,OACP;AAAA,IACF;AACA,IAAA,IAAA,CAAK,QAAA,GAAW,0BAAA,CAA2B,OAAA,CAAQ,QAAA,IAAY,IAAI,UAAU,CAAA;AAC7E,IAAA,IAAA,CAAK,SAAA,GAAY,yBAAA,CAA0B,OAAA,CAAQ,SAAA,IAAa,MAAM,WAAW,CAAA;AACjF,IAAA,IAAA,CAAK,QAAA,GAAW,0BAAA,CAA2B,OAAA,CAAQ,QAAA,IAAY,GAAG,UAAU,CAAA;AAC5E,IAAA,IAAA,CAAK,KAAA,GAAQ,yBAAA,CAA0B,OAAA,CAAQ,KAAA,IAAS,GAAG,OAAO,CAAA;AAClE,IAAA,IAAA,CAAK,IAAA,GAAO,IAAA,CAAK,IAAA,KAAS,KAAA,GAAQ,QAAA,GAAW,CAAA,QAAA;AAC7C,IAAA,IAAA,CAAK,YAAA,GAAe,CAAA;AACpB,IAAA,IAAA,CAAK,eAAA,GAAkB,CAAA;AAAA,EACzB;AAAA;AAAA;AAAA;AAAA,EAKQ,SAAS,OAAA,EAA0B;AACzC,IAAA,IAAI,IAAA,CAAK,SAAS,KAAA,EAAO;AACvB,MAAA,OAAO,OAAA,GAAU,IAAA,CAAK,IAAA,GAAO,IAAA,CAAK,SAAA;AAAA,IACpC;AACA,IAAA,OAAO,OAAA,GAAU,IAAA,CAAK,IAAA,GAAO,IAAA,CAAK,SAAA;AAAA,EACpC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,KAAK,MAAA,EAAsB;AACzB,IAAA,IAAI,CAAC,MAAA,CAAO,QAAA,CAAS,MAAM,CAAA,EAAG;AAC5B,MAAA,MAAM,IAAI,qBAAA,CAAsB,uBAAA,EAAyB,QAAA,EAAU,MAAM,CAAA;AAAA,IAC3E;AACA,IAAA,IAAI,IAAA,CAAK,kBAAkB,CAAA,EAAG;AAC5B,MAAA,IAAA,CAAK,eAAA,EAAA;AACL,MAAA,IAAA,CAAK,YAAA,GAAe,CAAA;AAAA,IACtB;AAEA,IAAA,IAAI,IAAA,CAAK,QAAA,CAAS,MAAM,CAAA,EAAG;AACzB,MAAA,IAAA,CAAK,IAAA,GAAO,MAAA;AACZ,MAAA,IAAA,CAAK,YAAA,GAAe,CAAA;AAAA,IACtB,CAAA,MAAA,IAAW,IAAA,CAAK,eAAA,KAAoB,CAAA,EAAG;AACrC,MAAA,IAAA,CAAK,YAAA,EAAA;AAAA,IACP;AAEA,IAAA,IAAI,IAAA,CAAK,YAAA,GAAe,IAAA,CAAK,QAAA,EAAU;AACrC,MAAA,IAAA,CAAK,QAAA,EAAS;AACd,MAAA,IAAA,CAAK,kBAAkB,IAAA,CAAK,QAAA;AAC5B,MAAA,IAAA,CAAK,YAAA,GAAe,CAAA;AAAA,IACtB;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKQ,QAAA,GAAiB;AACvB,IAAA,KAAA,IAAS,IAAI,CAAA,EAAG,CAAA,GAAI,KAAK,SAAA,CAAU,WAAA,CAAY,QAAQ,CAAA,EAAA,EAAK;AAC1D,MAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,SAAA,CAAU,WAAA,CAAY,CAAC,CAAA;AAC1C,MAAA,IAAI,CAAC,KAAA,EAAO;AACV,QAAA,MAAM,IAAI,qBAAA;AAAA,UACR,yBAAyB,CAAC,CAAA,YAAA,CAAA;AAAA,UAC1B,uBAAA;AAAA,UACA;AAAA,SACF;AAAA,MACF;AACA,MAAA,MAAM,SAAA,GAAY,cAAA,CAAe,KAAA,EAAO,CAAC,CAAA;AACzC,MAAA,MAAM,QAAQ,IAAA,CAAK,GAAA,CAAI,YAAY,IAAA,CAAK,MAAA,EAAQ,KAAK,KAAK,CAAA;AAC1D,MAAA,UAAA,CAAW,OAAO,KAAK,CAAA;AAAA,IACzB;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,SAAA,GAAsB;AACpB,IAAA,OAAO,IAAA,CAAK,SAAA,CAAU,WAAA,CAAY,GAAA,CAAI,CAAC,OAAO,KAAA,KAAU,cAAA,CAAe,KAAA,EAAO,KAAK,CAAC,CAAA;AAAA,EACtF;AACF;AAcO,IAAM,QAAA,GAAN,cAAuB,WAAA,CAAY;AAAA,EAChC,YAAA;AAAA,EACA,cAAA;AAAA,EAER,WAAA,CACE,SAAA,EACA,cAAA,EACA,OAAA,EACA;AACA,IAAA,MAAM,YAAA,GAAe,uBAAA,CAAwB,OAAA,CAAQ,YAAA,EAAc,cAAc,CAAA;AACjF,IAAA,MAAM,SAAA,GAAY,iBAAA,CAAkB,OAAA,CAAQ,SAAA,IAAa,EAAE,CAAA;AAC3D,IAAA,KAAA,CAAM,WAAW,EAAE,CAAA;AACnB,IAAA,IAAA,CAAK,YAAA,GAAe,YAAA;AACpB,IAAA,IAAA,CAAK,cAAA,GAAiB,cAAA;AACtB,IAAA,IAAA,CAAK,wBAAwB,SAAS,CAAA;AAAA,EACxC;AAAA,EAEA,KAAA,GAAkB;AAChB,IAAA,IAAI,IAAA,CAAK,SAAA,GAAY,IAAA,CAAK,YAAA,EAAc;AAEtC,MAAA,MAAM,MAAA,GAAA,CAAU,IAAA,CAAK,SAAA,GAAY,CAAA,IAAK,IAAA,CAAK,YAAA;AAC3C,MAAA,OAAO,KAAK,OAAA,CAAQ,GAAA,CAAI,CAAC,EAAA,KAAO,KAAK,MAAM,CAAA;AAAA,IAC7C;AAEA,IAAA,IAAI,KAAK,cAAA,EAAgB;AAEvB,MAAA,OAAO,IAAA,CAAK,eAAe,KAAA,EAAM;AAAA,IACnC;AAEA,IAAA,OAAO,IAAA,CAAK,OAAA;AAAA,EACd;AAAA,EAES,IAAA,GAAa;AACpB,IAAA,KAAA,CAAM,IAAA,EAAK;AAGX,IAAA,IAAI,IAAA,CAAK,SAAA,IAAa,IAAA,CAAK,YAAA,IAAgB,KAAK,cAAA,EAAgB;AAC9D,MAAA,IAAA,CAAK,eAAe,IAAA,EAAK;AAAA,IAC3B;AAAA,EACF;AACF;AAmBO,IAAM,UAAA,GAAN,cAAyB,WAAA,CAAY;AAAA,EAClC,KAAA;AAAA,EACA,UAAA;AAAA,EACA,QAAA;AAAA,EACA,SAAA;AAAA,EACA,cAAA;AAAA,EACA,cAAA;AAAA,EAER,WAAA,CACE,WACA,OAAA,EASA;AACA,IAAA,MAAM,KAAA,GAAQ,sBAAA,CAAuB,OAAA,CAAQ,KAAA,EAAO,OAAO,CAAA;AAC3D,IAAA,MAAM,UAAA,GAAa,uBAAA,CAAwB,OAAA,CAAQ,UAAA,EAAY,YAAY,CAAA;AAC3E,IAAA,MAAM,QAAA,GAAW,oBAAA,CAAqB,OAAA,CAAQ,QAAA,IAAY,KAAK,UAAU,CAAA;AACzE,IAAA,IAAI,QAAA,IAAY,CAAA,IAAK,QAAA,IAAY,CAAA,EAAG;AAClC,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,yCAAA;AAAA,QACA,UAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,MAAM,SAAA,GAAY,sBAAA,CAAuB,OAAA,CAAQ,SAAA,IAAa,IAAI,WAAW,CAAA;AAC7E,IAAA,MAAM,cAAA,GAAiB,sBAAA,CAAuB,OAAA,CAAQ,cAAA,IAAkB,KAAK,gBAAgB,CAAA;AAC7F,IAAA,MAAM,cAAA,GAAiB,QAAQ,cAAA,IAAkB,KAAA;AACjD,IAAA,IAAI,cAAA,KAAmB,KAAA,IAAS,cAAA,KAAmB,QAAA,EAAU;AAC3D,MAAA,MAAM,IAAI,qBAAA;AAAA,QACR,0CAAA;AAAA,QACA,gBAAA;AAAA,QACA;AAAA,OACF;AAAA,IACF;AACA,IAAA,MAAM,SAAA,GAAY,iBAAA,CAAkB,OAAA,CAAQ,SAAA,IAAa,EAAE,CAAA;AAC3D,IAAA,KAAA,CAAM,WAAW,EAAE,CAAA;AACnB,IAAA,IAAA,CAAK,KAAA,GAAQ,KAAA;AACb,IAAA,IAAA,CAAK,UAAA,GAAa,UAAA;AAClB,IAAA,IAAA,CAAK,QAAA,GAAW,QAAA;AAChB,IAAA,IAAA,CAAK,SAAA,GAAY,SAAA;AACjB,IAAA,IAAA,CAAK,cAAA,GAAiB,cAAA;AACtB,IAAA,IAAA,CAAK,cAAA,GAAiB,cAAA;AACtB,IAAA,IAAA,CAAK,wBAAwB,SAAS,CAAA;AAAA,EACxC;AAAA,EAEA,KAAA,GAAkB;AAChB,IAAA,MAAM,UAAU,IAAA,CAAK,SAAA;AACrB,IAAA,MAAM,OAAA,GAAU,IAAA,CAAK,GAAA,CAAI,CAAA,EAAG,IAAA,CAAK,MAAM,IAAA,CAAK,UAAA,GAAa,IAAA,CAAK,QAAQ,CAAC,CAAA;AACvE,IAAA,MAAM,YAAY,IAAA,CAAK,GAAA,CAAI,CAAA,EAAG,IAAA,CAAK,aAAa,OAAO,CAAA;AAEvD,IAAA,MAAM,SAAA,GAAY,IAAA,CAAK,KAAA,GAAQ,IAAA,CAAK,SAAA;AACpC,IAAA,MAAM,KAAA,GAAQ,IAAA,CAAK,KAAA,GAAQ,IAAA,CAAK,cAAA;AAEhC,IAAA,IAAI,EAAA;AAEJ,IAAA,IAAI,OAAA,IAAW,KAAK,UAAA,EAAY;AAC9B,MAAA,EAAA,GAAK,KAAA;AAAA,IACP,CAAA,MAAA,IAAW,UAAU,OAAA,EAAS;AAE5B,MAAA,MAAM,MAAM,OAAA,GAAU,OAAA;AACtB,MAAA,EAAA,GAAK,SAAA,GAAA,CAAa,IAAA,CAAK,KAAA,GAAQ,SAAA,IAAa,GAAA;AAAA,IAC9C,CAAA,MAAO;AAEL,MAAA,MAAM,GAAA,GAAA,CAAO,UAAU,OAAA,IAAW,SAAA;AAClC,MAAA,IAAI,IAAA,CAAK,mBAAmB,KAAA,EAAO;AACjC,QAAA,EAAA,GAAK,KAAA,GAAA,CAAU,IAAA,CAAK,KAAA,GAAQ,KAAA,KAAU,CAAA,GAAI,KAAK,GAAA,CAAI,IAAA,CAAK,EAAA,GAAK,GAAG,CAAA,CAAA,GAAM,CAAA;AAAA,MACxE,CAAA,MAAO;AACL,QAAA,EAAA,GAAK,IAAA,CAAK,KAAA,GAAA,CAAS,IAAA,CAAK,KAAA,GAAQ,KAAA,IAAS,GAAA;AAAA,MAC3C;AAAA,IACF;AAGA,IAAA,MAAM,OAAA,GAAU,IAAA,CAAK,OAAA,CAAQ,CAAC,CAAA,IAAK,CAAA;AACnC,IAAA,OAAO,IAAA,CAAK,OAAA,CAAQ,GAAA,CAAI,CAAC,MAAA,KAAW;AAClC,MAAA,IAAI,YAAY,CAAA,EAAG;AACjB,QAAA,OAAO,MAAA,KAAW,IAAI,CAAA,GAAI,EAAA;AAAA,MAC5B;AACA,MAAA,OAAO,MAAM,MAAA,GAAS,OAAA,CAAA;AAAA,IACxB,CAAC,CAAA;AAAA,EACH;AACF","file":"chunk-PR647I7R.js","sourcesContent":["// Base optimizer class\n\nexport type { ParamGroup } from \"./Optimizer\";\nexport { Optimizer } from \"./Optimizer\";\nexport { AdaDelta } from \"./optimizers/adadelta\";\nexport { Adagrad } from \"./optimizers/adagrad\";\n// Optimizers - gradient descent variants\nexport { Adam } from \"./optimizers/adam\";\nexport { AdamW } from \"./optimizers/adamw\";\nexport { Nadam } from \"./optimizers/nadam\";\nexport { RMSprop } from \"./optimizers/rmsprop\";\nexport { SGD } from \"./optimizers/sgd\";\n\n// Learning rate schedulers\nexport {\n  CosineAnnealingLR,\n  ExponentialLR,\n  LinearLR,\n  LRScheduler,\n  MultiStepLR,\n  OneCycleLR,\n  ReduceLROnPlateau,\n  StepLR,\n  WarmupLR,\n} from \"./schedulers\";\n","import { DataValidationError } from \"../core\";\nimport type { GradTensor } from \"../ndarray\";\n\n/**\n * Base class for all optimizers.\n *\n * This abstract class provides the foundation for implementing optimization algorithms\n * used in training machine learning models. All concrete optimizers (SGD, Adam, etc.)\n * must extend this class and implement the abstract `step()` method.\n *\n * **Key Features:**\n * - Parameter groups with per-group hyperparameters\n * - State management for stateful optimizers (momentum, adaptive learning rates)\n * - Gradient zeroing utilities\n * - State serialization for checkpointing\n *\n * **Design Pattern:**\n * The optimizer maintains a list of parameter groups, where each group can have\n * different hyperparameters (e.g., different learning rates for different layers).\n * This enables fine-grained control over the optimization process.\n *\n * @example\n * ```ts\n * import { SGD } from 'deepbox/optim';\n *\n * const optimizer = new SGD(model.parameters(), { lr: 0.01 });\n *\n * // Training loop\n * for (let epoch = 0; epoch < 100; epoch++) {\n *   optimizer.zeroGrad();\n *   const loss = computeLoss();\n *   loss.backward();\n *   optimizer.step();\n * }\n * ```\n *\n * @example\n * ```ts\n * // Using parameter groups with different learning rates\n * const optimizer = new SGD([\n *   { params: model.layer1.parameters(), lr: 0.01 },\n *   { params: model.layer2.parameters(), lr: 0.001 }\n * ], { lr: 0.01 });\n * ```\n *\n * References:\n * - PyTorch Optimizer: https://pytorch.org/docs/stable/optim.html\n *\n * @category Optimization\n */\n\n/**\n * Represents a group of parameters with optional per-group hyperparameters.\n *\n * @template Options - Type of optimizer-specific options\n * @property params - Iterable of parameters to optimize in this group\n */\nexport type ParamGroup<Options extends Record<string, unknown>> = {\n  readonly params: Iterable<GradTensor>;\n} & Partial<Options>;\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n  return typeof value === \"object\" && value !== null;\n}\n\nfunction ensureRecord(value: unknown, context: string) {\n  if (!isRecord(value)) {\n    throw new DataValidationError(`${context} must be an object`);\n  }\n  return value;\n}\n\nfunction isStateRecord(value: unknown): value is Record<string, unknown> {\n  return isRecord(value);\n}\n\nfunction ensureIntegerArray(value: unknown, context: string) {\n  if (!Array.isArray(value)) {\n    throw new DataValidationError(`${context} must be an array of integers`);\n  }\n  const output: number[] = [];\n  for (const entry of value) {\n    if (!Number.isInteger(entry)) {\n      throw new DataValidationError(`${context} must contain integers only`);\n    }\n    output.push(entry);\n  }\n  return output;\n}\n\n/**\n * Type guard to determine if params is an array of parameter groups.\n *\n * This function checks whether the provided params argument is a simple iterable\n * of parameters or an array of parameter groups with per-group options.\n *\n * @template Options - Type of optimizer-specific options\n * @param params - Either an iterable of parameters or array of parameter groups\n * @returns True if params is an array of parameter groups, false otherwise\n */\nfunction isParamGroupArray<Options extends Record<string, unknown>>(\n  params: Iterable<GradTensor> | ReadonlyArray<ParamGroup<Options>>\n): params is ReadonlyArray<ParamGroup<Options>> {\n  // Check if params is an array (parameter groups must be arrays)\n  if (!Array.isArray(params)) return false;\n  // Empty array is considered a valid parameter group array\n  if (params.length === 0) return true;\n  // Check if first element has a 'params' property (indicating it's a group)\n  const first = params[0];\n  if (!first || typeof first !== \"object\") return false;\n  return \"params\" in first;\n}\n\n/**\n * Abstract base class for all optimization algorithms.\n *\n * @template Options - Type defining optimizer-specific hyperparameters\n * @template State - Type defining per-parameter state (e.g., momentum buffers)\n */\nexport abstract class Optimizer<\n  Options extends Record<string, unknown>,\n  State extends Record<string, unknown>,\n> {\n  /**\n   * Groups of parameters with their associated hyperparameters.\n   * Each group can have different options (e.g., learning rates).\n   * Exposed publicly to enable scheduler integrations.\n   */\n  public paramGroups: Array<{\n    params: GradTensor[];\n    options: Options;\n  }>;\n\n  /**\n   * Per-parameter state storage.\n   * Maps each parameter to its optimizer-specific state (momentum, adaptive rates, etc.).\n   */\n  protected state: Map<GradTensor, State> = new Map();\n\n  /**\n   * Create a new optimizer.\n   *\n   * Initializes the optimizer with either a simple list of parameters or\n   * multiple parameter groups with per-group hyperparameters.\n   *\n   * @param params - Either an iterable of parameters or array of parameter groups\n   * @param defaults - Default hyperparameters applied to all groups\n   */\n  constructor(\n    params: Iterable<GradTensor> | ReadonlyArray<ParamGroup<Options>>,\n    protected readonly defaults: Readonly<Options>\n  ) {\n    // Initialize empty parameter groups array\n    this.paramGroups = [];\n\n    // Handle both simple param list and param groups\n    if (!isParamGroupArray<Options>(params)) {\n      // Simple iterable of parameters - create single group with default options\n      this.paramGroups.push({\n        params: Array.from(params), // Convert iterable to array for efficient access\n        options: { ...defaults }, // Clone defaults to avoid mutation\n      });\n    } else {\n      // Array of parameter groups - create group for each with merged options\n      for (const group of params) {\n        // Destructure to separate params from group-specific options\n        const { params: groupParams, ...groupOptions } = group;\n        this.paramGroups.push({\n          params: Array.from(groupParams), // Convert to array\n          options: { ...defaults, ...groupOptions }, // Merge defaults with group options\n        });\n      }\n    }\n  }\n\n  /**\n   * Perform a single optimization step (parameter update).\n   *\n   * This abstract method must be implemented by all optimizer subclasses.\n   * It applies the optimization algorithm to update all parameters based on\n   * their gradients.\n   *\n   * @param closure - Optional closure that reevaluates the model and returns the loss.\n   *                  Used by some optimizers (e.g., LBFGS) that require multiple\n   *                  function evaluations per step.\n   * @returns Loss value if closure is provided, undefined otherwise\n   */\n  abstract step(closure?: () => number): number | undefined;\n\n  /**\n   * Zero out the gradients of all optimized parameters.\n   *\n   * This method should be called at the beginning of each training iteration,\n   * before computing new gradients. Without this call, gradients would accumulate\n   * across iterations, leading to incorrect updates.\n   *\n   * **Implementation Note:**\n   * For parameters wrapped in GradTensor, this calls zeroGrad() on each parameter,\n   * which either sets the gradient to zero or initializes it if not yet created.\n   *\n   * @example\n   * ```ts\n   * // Typical training loop\n   * optimizer.zeroGrad();              // Clear previous gradients\n   * const output = model.forward(input);\n   * const loss = criterion(output, target);\n   * loss.backward();                   // Compute new gradients\n   * optimizer.step();                  // Update parameters\n   * ```\n   */\n  zeroGrad(): void {\n    // Iterate through all parameter groups\n    for (const group of this.paramGroups) {\n      // Zero gradient for each parameter in the group\n      for (const param of group.params) {\n        param.zeroGrad(); // Delegate to GradTensor's zeroGrad method\n      }\n    }\n  }\n\n  /**\n   * Add a parameter group to the optimizer.\n   *\n   * This method allows adding new parameters to optimize after the optimizer\n   * has been created. This is particularly useful for:\n   * - Fine-tuning: adding pre-trained layers with different learning rates\n   * - Progressive training: gradually unfreezing layers\n   * - Dynamic architectures: adding parameters while the model grows\n   *\n   * @param paramGroup - Parameter group to add with optional per-group options\n   *\n   * @example\n   * ```ts\n   * const optimizer = new SGD(model.backbone.parameters(), { lr: 0.001 });\n   * // Later, add classifier with higher learning rate\n   * optimizer.addParamGroup({\n   *   params: model.classifier.parameters(),\n   *   lr: 0.01\n   * });\n   * ```\n   */\n  addParamGroup(paramGroup: ParamGroup<Options>): void {\n    // Destructure to separate params from group-specific options\n    const { params, ...options } = paramGroup;\n    // Add new group with merged options (defaults + group-specific)\n    this.paramGroups.push({\n      params: Array.from(params), // Convert iterable to array\n      options: { ...this.defaults, ...options }, // Merge with defaults\n    });\n  }\n\n  /**\n   * Validate that a given state object matches the optimizer's state type.\n   *\n   * @param state - The state object to validate\n   * @returns True if the state object is valid, false otherwise\n   */\n  protected abstract isState(state: Record<string, unknown>): state is State;\n\n  /**\n   * Get the current state of the optimizer.\n   *\n   * Returns a dictionary containing all optimizer state that needs to be\n   * saved for checkpointing. This includes per-parameter state (momentum buffers,\n   * adaptive learning rates, etc.) and parameter group configurations.\n   *\n   * **Note:** In a production implementation, parameters would be identified by\n   * unique IDs rather than object references for proper serialization.\n   *\n   * @returns Optimizer state dictionary containing state and parameter groups\n   *\n   * @example\n   * ```ts\n   * // Save checkpoint\n   * const checkpoint = {\n   *   model: model.stateDict(),\n   *   optimizer: optimizer.stateDict(),\n   *   epoch: currentEpoch\n   * };\n   * ```\n   */\n  stateDict() {\n    const paramIdMap = new Map<GradTensor, number>();\n    const orderedParams: GradTensor[] = [];\n    const getParamId = (param: GradTensor) => {\n      const existing = paramIdMap.get(param);\n      if (existing !== undefined) return existing;\n      const id = orderedParams.length;\n      orderedParams.push(param);\n      paramIdMap.set(param, id);\n      return id;\n    };\n\n    return {\n      // Serialize per-parameter state\n      state: Array.from(this.state.entries()).map(([param, state]) => ({\n        paramId: getParamId(param),\n        param: param, // Backward-compatible references\n        state, // Optimizer-specific state (momentum, etc.)\n      })),\n      // Serialize parameter groups and their options\n      paramGroups: this.paramGroups.map((group) => ({\n        params: group.params, // Backward-compatible references\n        paramIds: group.params.map((param) => getParamId(param)),\n        options: group.options, // Hyperparameters for this group\n      })),\n    };\n  }\n\n  /**\n   * Load optimizer state from a state dictionary.\n   *\n   * Restores the optimizer to a previously saved state, including all\n   * per-parameter state and parameter group configurations. This is essential\n   * for resuming training from checkpoints.\n   *\n   * **Important:** The loaded state must be compatible with the current\n   * optimizer configuration (same parameters, same optimizer type).\n   *\n   * @param stateDict - State dictionary previously returned by stateDict()\n   *\n   * @example\n   * ```ts\n   * // Resume from checkpoint\n   * const checkpoint = loadCheckpoint('checkpoint.json');\n   * model.loadStateDict(checkpoint.model);\n   * optimizer.loadStateDict(checkpoint.optimizer);\n   * ```\n   */\n  loadStateDict(stateDict: Record<string, unknown>): void {\n    const currentParams = this.paramGroups.flatMap((group) => group.params);\n    const currentParamCount = currentParams.length;\n    const paramLookup = new Map<unknown, number>();\n    for (let i = 0; i < currentParams.length; i++) {\n      paramLookup.set(currentParams[i], i);\n    }\n\n    // Validate paramGroups if present\n    if (Object.hasOwn(stateDict, \"paramGroups\")) {\n      const rawGroups = stateDict[\"paramGroups\"];\n      if (!Array.isArray(rawGroups)) {\n        throw new DataValidationError(\"paramGroups must be an array\");\n      }\n      const groupsArray: unknown[] = rawGroups;\n\n      if (groupsArray.length === 0) {\n        if (this.paramGroups.length !== 0) {\n          throw new DataValidationError(\"paramGroups cannot be empty\");\n        }\n        this.paramGroups = [];\n      } else {\n        if (groupsArray.length !== this.paramGroups.length) {\n          throw new DataValidationError(\"paramGroups count mismatch\");\n        }\n\n        const seenParamIds = new Set<number>();\n        let totalParamCount = 0;\n        let sawParamIds = false;\n        let sawNoParamIds = false;\n        const nextGroups: Array<{\n          params: GradTensor[];\n          options: Options;\n          paramIds?: number[];\n        }> = [];\n\n        groupsArray.forEach((rawGroup, index) => {\n          const groupRecord = ensureRecord(rawGroup, `paramGroups[${index}]`);\n          const optionsRaw = ensureRecord(groupRecord[\"options\"], `paramGroups[${index}].options`);\n          const options: Options = { ...this.defaults };\n          const optionsRecord: Record<string, unknown> = options;\n          const defaultsRecord: Record<string, unknown> = { ...this.defaults };\n\n          for (const [key, value] of Object.entries(optionsRaw)) {\n            if (Object.hasOwn(defaultsRecord, key)) {\n              const defaultVal = defaultsRecord[key];\n              const expectedType = typeof defaultVal;\n              const actualType = typeof value;\n\n              if (actualType !== expectedType) {\n                throw new DataValidationError(\n                  `Type mismatch for option '${key}' in paramGroups[${index}]: expected ${expectedType}, got ${actualType}`\n                );\n              }\n              optionsRecord[key] = value;\n            }\n          }\n\n          const paramIdsRaw = groupRecord[\"paramIds\"];\n          const paramsRaw = groupRecord[\"params\"];\n          let paramIds: number[] | undefined;\n          if (paramIdsRaw !== undefined) {\n            paramIds = ensureIntegerArray(paramIdsRaw, `paramGroups[${index}].paramIds`);\n            sawParamIds = true;\n          } else {\n            sawNoParamIds = true;\n          }\n\n          let resolvedParams: GradTensor[] | undefined;\n\n          if (paramIds) {\n            for (const id of paramIds) {\n              if (id < 0 || id >= currentParamCount) {\n                throw new DataValidationError(`Invalid paramId ${id} in paramGroups`);\n              }\n              if (seenParamIds.has(id)) {\n                throw new DataValidationError(`Duplicate paramId ${id} in paramGroups`);\n              }\n              seenParamIds.add(id);\n            }\n            totalParamCount += paramIds.length;\n            resolvedParams = paramIds.map((id) => {\n              const param = currentParams[id];\n              if (!param) {\n                throw new DataValidationError(`Invalid paramId ${id} in paramGroups`);\n              }\n              return param;\n            });\n          }\n\n          if (paramsRaw !== undefined) {\n            if (!Array.isArray(paramsRaw)) {\n              throw new DataValidationError(`paramGroups[${index}].params must be an array`);\n            }\n            const resolvedFromParams: GradTensor[] = [];\n            let hasUnknown = false;\n            for (const paramRef of paramsRaw) {\n              const paramIndex = paramLookup.get(paramRef);\n              if (paramIndex === undefined) {\n                hasUnknown = true;\n                continue;\n              }\n              const param = currentParams[paramIndex];\n              if (!param) {\n                hasUnknown = true;\n                continue;\n              }\n              resolvedFromParams.push(param);\n            }\n            if (!hasUnknown) {\n              if (paramIds && paramIds.length !== resolvedFromParams.length) {\n                throw new DataValidationError(\"paramIds length does not match params length\");\n              }\n              if (!resolvedParams) {\n                resolvedParams = resolvedFromParams;\n              }\n            }\n          }\n\n          if (!resolvedParams) {\n            throw new DataValidationError(`paramGroups[${index}] must include params or paramIds`);\n          }\n\n          if (paramIds === undefined) {\n            nextGroups.push({ params: resolvedParams, options });\n          } else {\n            nextGroups.push({ params: resolvedParams, options, paramIds });\n          }\n        });\n\n        if (sawParamIds && sawNoParamIds) {\n          throw new DataValidationError(\"paramIds must be provided for all parameter groups\");\n        }\n\n        if (sawParamIds && totalParamCount !== currentParamCount) {\n          throw new DataValidationError(\n            `Parameter count mismatch: expected ${currentParamCount}, got ${totalParamCount}`\n          );\n        }\n\n        this.paramGroups = nextGroups.map((group) => ({\n          params: group.params,\n          options: group.options,\n        }));\n      }\n    }\n\n    // Load per-parameter state if present in state dict\n    if (Object.hasOwn(stateDict, \"state\")) {\n      const rawState = stateDict[\"state\"];\n      if (!Array.isArray(rawState)) {\n        throw new DataValidationError(\"state must be an array\");\n      }\n      const stateArray: unknown[] = rawState;\n      // Clear existing state before loading\n      this.state.clear();\n      // Restore each parameter's state\n      stateArray.forEach((rawEntry, index) => {\n        const entryRecord = ensureRecord(rawEntry, `state[${index}]`);\n        if (!Object.hasOwn(entryRecord, \"state\")) {\n          throw new DataValidationError(`state[${index}].state is required`);\n        }\n        const entryStateValue = ensureRecord(entryRecord[\"state\"], `state[${index}].state`);\n        if (!isStateRecord(entryStateValue)) {\n          throw new DataValidationError(`state[${index}].state must be an object`);\n        }\n\n        const paramIdRaw = entryRecord[\"paramId\"];\n        const paramRaw = entryRecord[\"param\"];\n\n        let resolvedParam: GradTensor | undefined;\n\n        if (paramIdRaw !== undefined) {\n          if (\n            paramIdRaw === null ||\n            typeof paramIdRaw !== \"number\" ||\n            !Number.isInteger(paramIdRaw)\n          ) {\n            throw new DataValidationError(`Invalid paramId ${String(paramIdRaw)} in state`);\n          }\n          if (paramIdRaw < 0 || paramIdRaw >= currentParamCount) {\n            throw new DataValidationError(`Invalid paramId ${paramIdRaw} in state`);\n          }\n          const param = currentParams[paramIdRaw];\n          if (!param) {\n            throw new DataValidationError(`Invalid paramId ${paramIdRaw} in state`);\n          }\n          if (paramRaw !== undefined) {\n            const paramIndex = paramLookup.get(paramRaw);\n            if (paramIndex === undefined || paramIndex !== paramIdRaw) {\n              throw new DataValidationError(`paramId ${paramIdRaw} does not match provided param`);\n            }\n          }\n          resolvedParam = param;\n        } else {\n          if (paramRaw === undefined) {\n            throw new DataValidationError(\"Missing param reference in state entry\");\n          }\n          const paramIndex = paramLookup.get(paramRaw);\n          if (paramIndex === undefined) {\n            throw new DataValidationError(\"Unknown param reference in state entry\");\n          }\n          const param = currentParams[paramIndex];\n          if (!param) {\n            throw new DataValidationError(\"Unknown param reference in state entry\");\n          }\n          resolvedParam = param;\n        }\n\n        if (!resolvedParam) {\n          throw new DataValidationError(`Unable to resolve parameter for state[${index}]`);\n        }\n        if (!this.isState(entryStateValue)) {\n          throw new DataValidationError(`state[${index}].state has invalid structure`);\n        }\n        this.state.set(resolvedParam, entryStateValue);\n      });\n    }\n  }\n}\n","/**\n * Internal utilities for optimizer implementations.\n * This module is not part of the public API.\n *\n * @internal\n */\n\nimport {\n  DeepboxError,\n  DTypeError,\n  IndexError,\n  InvalidParameterError,\n  NotFittedError,\n  ShapeError,\n} from \"../core\";\nimport type { GradTensor } from \"../ndarray\";\n\n/**\n * Supported floating-point typed array types for optimizer parameters.\n */\nexport type FloatTypedArray = Float32Array | Float64Array;\n\nfunction isFloatTypedArray(value: unknown): value is FloatTypedArray {\n  return value instanceof Float32Array || value instanceof Float64Array;\n}\n\n/**\n * Safely access an array element with bounds checking.\n *\n * @param array - Array to access\n * @param index - Index to access\n * @param context - Context string for error messages\n * @returns The value at the index\n * @throws {IndexError} If index is out of bounds\n * @throws {DeepboxError} If value is unexpectedly undefined\n */\nexport function safeArrayAccess<T>(array: ArrayLike<T>, index: number, context: string): T {\n  if (index < 0 || index >= array.length) {\n    throw new IndexError(`Index ${index} out of bounds [0, ${array.length}) in ${context}`, {\n      index,\n      validRange: [0, array.length - 1],\n    });\n  }\n  const value = array[index];\n  if (value === undefined) {\n    throw new DeepboxError(`Unexpected undefined at index ${index} in ${context}`);\n  }\n  return value;\n}\n\n/**\n * Validates that a numeric value is finite and non-negative.\n *\n * @param name - Name of the parameter being validated\n * @param value - Value to validate\n * @throws {InvalidParameterError} If value is not finite or is negative\n */\nexport function assertFiniteNonNegative(name: string, value: number): void {\n  if (!Number.isFinite(value) || value < 0) {\n    throw new InvalidParameterError(`Invalid ${name}: ${value}`, name, value);\n  }\n}\n\n/**\n * Validates that a numeric value is finite and positive (> 0).\n *\n * @param name - Name of the parameter being validated\n * @param value - Value to validate\n * @throws {InvalidParameterError} If value is not finite or is not positive\n */\nexport function assertFinitePositive(name: string, value: number): void {\n  if (!Number.isFinite(value) || value <= 0) {\n    throw new InvalidParameterError(`Invalid ${name}: ${value} (must be > 0)`, name, value);\n  }\n}\n\n/**\n * Validates that a numeric value is finite.\n *\n * @param name - Name of the parameter being validated\n * @param value - Value to validate\n * @throws {InvalidParameterError} If value is not finite\n */\nexport function assertFinite(name: string, value: number): void {\n  if (!Number.isFinite(value)) {\n    throw new InvalidParameterError(`Invalid ${name}: ${value}`, name, value);\n  }\n}\n\n/**\n * Validates that a value is in the range [min, max).\n *\n * @param name - Name of the parameter being validated\n * @param value - Value to validate\n * @param min - Minimum value (inclusive)\n * @param max - Maximum value (exclusive)\n * @throws {InvalidParameterError} If value is out of range\n */\nexport function assertInRange(name: string, value: number, min: number, max: number): void {\n  if (!Number.isFinite(value) || value < min || value >= max) {\n    throw new InvalidParameterError(\n      `Invalid ${name}: ${value} (must be in range [${min}, ${max}))`,\n      name,\n      value\n    );\n  }\n}\n\n/**\n * Validates that a parameter has a gradient and returns gradient information.\n *\n * @param param - Parameter to validate\n * @param optimizerName - Name of the optimizer for error messages\n * @returns Object containing gradient data, offset, parameter data, and offset\n * @throws {InvalidParameterError} If parameter doesn't require gradients\n * @throws {NotFittedError} If parameter has no gradient\n * @throws {DTypeError} If parameter or gradient has unsupported dtype\n * @throws {ShapeError} If gradient shape doesn't match parameter shape\n */\nexport function assertHasGradFloat(\n  param: GradTensor,\n  optimizerName: string\n): {\n  grad: FloatTypedArray;\n  gradOffset: number;\n  param: FloatTypedArray;\n  paramOffset: number;\n} {\n  if (!param.requiresGrad) {\n    throw new InvalidParameterError(\n      \"Cannot optimize a parameter with requiresGrad=false\",\n      \"requiresGrad\",\n      false\n    );\n  }\n\n  const g = param.grad;\n  if (!g) {\n    throw new NotFittedError(\n      \"Cannot optimize a parameter without a gradient. Did you forget backward()?\"\n    );\n  }\n\n  const paramData = param.tensor.data;\n  const gradData = g.data;\n\n  if (!isFloatTypedArray(paramData) || !isFloatTypedArray(gradData)) {\n    throw new DTypeError(\n      `${optimizerName} optimizer supports float32 and float64 parameters and gradients only`\n    );\n  }\n\n  if (paramData.constructor !== gradData.constructor) {\n    throw new DTypeError(\n      `${optimizerName} optimizer requires parameter and gradient dtypes to match`\n    );\n  }\n\n  if (param.tensor.size !== g.size) {\n    throw new ShapeError(\n      `Gradient shape must match parameter shape (param: ${param.tensor.size}, grad: ${g.size})`\n    );\n  }\n\n  return {\n    grad: gradData,\n    gradOffset: g.offset,\n    param: paramData,\n    paramOffset: param.tensor.offset,\n  };\n}\n\n/**\n * Validates that a state buffer has the correct size.\n *\n * @param buffer - State buffer to validate\n * @param expectedSize - Expected size\n * @param bufferName - Name of the buffer for error messages\n * @throws {DeepboxError} If buffer size doesn't match expected size\n */\nexport function assertBufferSize(\n  buffer: ArrayLike<number>,\n  expectedSize: number,\n  bufferName: string\n): void {\n  if (buffer.length !== expectedSize) {\n    throw new DeepboxError(\n      `State buffer size mismatch for ${bufferName}: expected ${expectedSize}, got ${buffer.length}`\n    );\n  }\n}\n","import { InvalidParameterError } from \"../../core\";\nimport type { GradTensor } from \"../../ndarray\";\nimport {\n  assertBufferSize,\n  assertFinite,\n  assertFiniteNonNegative,\n  assertFinitePositive,\n  assertHasGradFloat,\n  assertInRange,\n  safeArrayAccess,\n} from \"../_internal\";\nimport { Optimizer, type ParamGroup } from \"../Optimizer\";\n\ntype AdaDeltaOptions = {\n  lr: number;\n  readonly rho: number;\n  readonly eps: number;\n  readonly weightDecay: number;\n};\n\ntype AdaDeltaState = {\n  squareAvg: Float64Array;\n  accDelta: Float64Array;\n};\n\n/**\n * AdaDelta optimizer.\n *\n * Implements AdaDelta algorithm - an extension of Adagrad that seeks to reduce\n * its aggressive, monotonically decreasing learning rate. AdaDelta adapts learning\n * rates based on a moving window of gradient updates, rather than accumulating all\n * past gradients.\n *\n * @example\n * ```ts\n * import { AdaDelta } from 'deepbox/optim';\n *\n * const optimizer = new AdaDelta(model.parameters(), {\n *   lr: 1.0,\n *   rho: 0.9,\n *   eps: 1e-6\n * });\n *\n * // Training loop\n * for (let epoch = 0; epoch < numEpochs; epoch++) {\n *   optimizer.zeroGrad();\n *   // ...\n *   optimizer.step();\n * }\n * ```\n *\n * @category Optimizers\n */\nexport class AdaDelta extends Optimizer<AdaDeltaOptions, AdaDeltaState> {\n  private _stepCount = 0;\n\n  get stepCount(): number {\n    return this._stepCount;\n  }\n  constructor(\n    params: Iterable<GradTensor> | ReadonlyArray<ParamGroup<AdaDeltaOptions>>,\n    options: {\n      readonly lr?: number;\n      readonly rho?: number;\n      readonly eps?: number;\n      readonly weightDecay?: number;\n    } = {}\n  ) {\n    const defaults = {\n      lr: options.lr ?? 1.0,\n      rho: options.rho ?? 0.9,\n      eps: options.eps ?? 1e-6,\n      weightDecay: options.weightDecay ?? 0,\n    };\n\n    super(params, defaults);\n\n    // Validate hyperparameters\n    assertFiniteNonNegative(\"learning rate\", defaults.lr);\n    assertInRange(\"rho\", defaults.rho, 0, 1);\n    assertFinitePositive(\"epsilon\", defaults.eps);\n    assertFiniteNonNegative(\"weight_decay value\", defaults.weightDecay);\n  }\n\n  /**\n   * Get the current learning rate.\n   *\n   * @param groupIdx - Parameter group index (default: 0)\n   * @returns Current learning rate\n   */\n  getLearningRate(groupIdx = 0): number {\n    const group = this.paramGroups[groupIdx];\n    if (!group) {\n      throw new InvalidParameterError(\n        `Invalid group index: ${groupIdx} (valid range: [0, ${this.paramGroups.length}))`,\n        \"groupIdx\",\n        groupIdx\n      );\n    }\n    return group.options.lr;\n  }\n\n  /**\n   * Set the learning rate for all parameter groups.\n   *\n   * @param lr - New learning rate\n   */\n  setLearningRate(lr: number): void {\n    assertFiniteNonNegative(\"learning rate\", lr);\n    for (const group of this.paramGroups) {\n      group.options.lr = lr;\n    }\n  }\n\n  protected isState(state: Record<string, unknown>): state is AdaDeltaState {\n    return state[\"squareAvg\"] instanceof Float64Array && state[\"accDelta\"] instanceof Float64Array;\n  }\n\n  step(closure?: () => number): number | undefined {\n    let loss: number | undefined;\n\n    if (closure) {\n      loss = closure();\n    }\n\n    // Increment global step counter\n    this._stepCount++;\n\n    for (const group of this.paramGroups) {\n      const { lr, rho, eps, weightDecay } = group.options;\n\n      // Re-validate hyperparameters\n      assertFiniteNonNegative(\"learning rate\", lr);\n      assertInRange(\"rho\", rho, 0, 1);\n      assertFinitePositive(\"epsilon\", eps);\n      assertFiniteNonNegative(\"weight_decay value\", weightDecay);\n\n      for (const param of group.params) {\n        const {\n          grad: gradData,\n          gradOffset: gOff,\n          param: pData,\n          paramOffset: pOff,\n        } = assertHasGradFloat(param, \"AdaDelta\");\n        const size = param.tensor.size;\n\n        // Initialize state if needed\n        let state = this.state.get(param);\n        if (!state) {\n          state = {\n            squareAvg: new Float64Array(size),\n            accDelta: new Float64Array(size),\n          };\n          this.state.set(param, state);\n        }\n\n        // Validate state buffer sizes\n        assertBufferSize(state.squareAvg, size, \"AdaDelta squareAvg\");\n        assertBufferSize(state.accDelta, size, \"AdaDelta accDelta\");\n\n        for (let i = 0; i < size; i++) {\n          const gi0 = safeArrayAccess(gradData, gOff + i, \"AdaDelta gradient\");\n          const pi = safeArrayAccess(pData, pOff + i, \"AdaDelta parameter\");\n          assertFinite(\"gradient\", gi0);\n          assertFinite(\"parameter\", pi);\n\n          // Apply weight decay\n          const gi = weightDecay !== 0 ? gi0 + weightDecay * pi : gi0;\n\n          // Update square average: E[g²](t) = ρ * E[g²](t-1) + (1 - ρ) * g(t)²\n          const sq = safeArrayAccess(state.squareAvg, i, \"AdaDelta squareAvg\");\n          const sqNew = rho * sq + (1 - rho) * gi * gi;\n          state.squareAvg[i] = sqNew;\n\n          // Compute RMS[g](t) = √(E[g²](t) + ε)\n          const std = Math.sqrt(sqNew + eps);\n\n          // Compute RMS[Δθ](t-1) = √(E[Δθ²](t-1) + ε)\n          const accD = safeArrayAccess(state.accDelta, i, \"AdaDelta accDelta\");\n          const rmsUpdate = Math.sqrt(accD + eps);\n\n          // Compute parameter update: Δθ(t) = -RMS[Δθ](t-1) / RMS[g](t) * g(t)\n          const delta = (rmsUpdate / std) * gi;\n\n          // Update accumulated delta: E[Δθ²](t) = ρ * E[Δθ²](t-1) + (1 - ρ) * Δθ(t)²\n          state.accDelta[i] = rho * accD + (1 - rho) * delta * delta;\n\n          // Update parameter: θ(t+1) = θ(t) - lr * Δθ(t)\n          pData[pOff + i] = pi - lr * delta;\n        }\n      }\n    }\n\n    return loss;\n  }\n}\n","import { InvalidParameterError } from \"../../core\";\nimport type { GradTensor } from \"../../ndarray\";\nimport {\n  assertBufferSize,\n  assertFinite,\n  assertFiniteNonNegative,\n  assertFinitePositive,\n  assertHasGradFloat,\n  safeArrayAccess,\n} from \"../_internal\";\nimport { Optimizer, type ParamGroup } from \"../Optimizer\";\n\ntype AdagradOptions = {\n  lr: number;\n  eps: number;\n  weightDecay: number;\n  lrDecay: number;\n};\n\ntype AdagradState = {\n  step: number;\n  sum: Float64Array;\n};\n\n/**\n * Adagrad (Adaptive Gradient Algorithm) optimizer.\n *\n * Adagrad adapts the learning rate for each parameter based on the historical\n * sum of squared gradients. Parameters with larger gradients receive smaller\n * effective learning rates, while parameters with smaller gradients receive\n * larger effective learning rates.\n *\n * @example\n * ```ts\n * import { Adagrad } from 'deepbox/optim';\n *\n * const optimizer = new Adagrad(model.parameters(), {\n *   lr: 0.01,\n *   eps: 1e-10\n * });\n *\n * // Training loop\n * for (let epoch = 0; epoch < numEpochs; epoch++) {\n *   optimizer.zeroGrad();\n *   // ...\n *   optimizer.step();\n * }\n * ```\n *\n * @category Optimizers\n */\nexport class Adagrad extends Optimizer<AdagradOptions, AdagradState> {\n  private _stepCount = 0;\n\n  get stepCount(): number {\n    return this._stepCount;\n  }\n  constructor(\n    params: Iterable<GradTensor> | ReadonlyArray<ParamGroup<AdagradOptions>>,\n    options: {\n      readonly lr?: number;\n      readonly eps?: number;\n      readonly weightDecay?: number;\n      readonly lrDecay?: number;\n    } = {}\n  ) {\n    const defaults = {\n      lr: options.lr ?? 0.01,\n      eps: options.eps ?? 1e-10,\n      weightDecay: options.weightDecay ?? 0,\n      lrDecay: options.lrDecay ?? 0,\n    };\n\n    super(params, defaults);\n\n    assertFiniteNonNegative(\"learning rate\", defaults.lr);\n    assertFinitePositive(\"epsilon\", defaults.eps);\n    assertFiniteNonNegative(\"weight_decay value\", defaults.weightDecay);\n    assertFiniteNonNegative(\"lr_decay\", defaults.lrDecay);\n  }\n\n  /**\n   * Get the current learning rate.\n   *\n   * @param groupIdx - Parameter group index (default: 0)\n   * @returns Current learning rate\n   */\n  getLearningRate(groupIdx = 0): number {\n    const group = this.paramGroups[groupIdx];\n    if (!group) {\n      throw new InvalidParameterError(\n        `Invalid group index: ${groupIdx} (valid range: [0, ${this.paramGroups.length}))`,\n        \"groupIdx\",\n        groupIdx\n      );\n    }\n    return group.options.lr;\n  }\n\n  /**\n   * Set the learning rate for all parameter groups.\n   *\n   * @param lr - New learning rate\n   */\n  setLearningRate(lr: number): void {\n    assertFiniteNonNegative(\"learning rate\", lr);\n    for (const group of this.paramGroups) {\n      group.options.lr = lr;\n    }\n  }\n\n  protected isState(state: Record<string, unknown>): state is AdagradState {\n    return typeof state[\"step\"] === \"number\" && state[\"sum\"] instanceof Float64Array;\n  }\n\n  step(closure?: () => number): number | undefined {\n    let loss: number | undefined;\n\n    if (closure) {\n      loss = closure();\n    }\n\n    // Increment global step counter\n    this._stepCount++;\n\n    for (const group of this.paramGroups) {\n      const { lr, eps, weightDecay, lrDecay } = group.options;\n\n      assertFiniteNonNegative(\"learning rate\", lr);\n      assertFinitePositive(\"epsilon\", eps);\n      assertFiniteNonNegative(\"weight_decay value\", weightDecay);\n      assertFiniteNonNegative(\"lr_decay\", lrDecay);\n\n      for (const param of group.params) {\n        const {\n          grad: gradData,\n          gradOffset: gOff,\n          param: pData,\n          paramOffset: pOff,\n        } = assertHasGradFloat(param, \"Adagrad\");\n        const size = param.tensor.size;\n\n        const existing = this.state.get(param);\n        const state =\n          existing ??\n          (() => {\n            const next = {\n              step: 0,\n              sum: new Float64Array(size),\n            };\n            this.state.set(param, next);\n            return next;\n          })();\n\n        // Validate state buffer size\n        assertBufferSize(state.sum, size, \"Adagrad sum\");\n\n        state.step += 1;\n\n        const clr = lr / (1 + (state.step - 1) * lrDecay);\n\n        for (let i = 0; i < size; i++) {\n          const gi0 = safeArrayAccess(gradData, gOff + i, \"Adagrad gradient\");\n          const pi = safeArrayAccess(pData, pOff + i, \"Adagrad parameter\");\n          assertFinite(\"gradient\", gi0);\n          assertFinite(\"parameter\", pi);\n\n          const gi = weightDecay !== 0 ? gi0 + weightDecay * pi : gi0;\n\n          const sumVal = safeArrayAccess(state.sum, i, \"Adagrad sum\");\n          const sumNew = sumVal + gi * gi;\n          state.sum[i] = sumNew;\n\n          const std = Math.sqrt(sumNew) + eps;\n          pData[pOff + i] = pi - clr * (gi / std);\n        }\n      }\n    }\n\n    return loss;\n  }\n}\n","import { DeepboxError, InvalidParameterError } from \"../../core\";\nimport type { GradTensor } from \"../../ndarray\";\nimport {\n  assertBufferSize,\n  assertFinite,\n  assertFiniteNonNegative,\n  assertFinitePositive,\n  assertHasGradFloat,\n  assertInRange,\n  safeArrayAccess,\n} from \"../_internal\";\nimport { Optimizer, type ParamGroup } from \"../Optimizer\";\n\ntype AdamOptions = {\n  lr: number;\n  beta1: number;\n  beta2: number;\n  eps: number;\n  weightDecay: number;\n  amsgrad: boolean;\n};\n\ntype AdamState = {\n  step: number;\n  expAvg: Float64Array;\n  expAvgSq: Float64Array;\n  maxExpAvgSq?: Float64Array;\n};\n\n/**\n * Adam (Adaptive Moment Estimation) optimizer.\n *\n * Computes adaptive learning rates for each parameter by maintaining\n * running averages of both the gradients and their squared values.\n *\n * @example\n * ```ts\n * import { Adam } from 'deepbox/optim';\n *\n * const optimizer = new Adam(model.parameters(), {\n *   lr: 0.001,\n *   beta1: 0.9,\n *   beta2: 0.999\n * });\n * ```\n *\n * @category Optimizers\n */\nexport class Adam extends Optimizer<AdamOptions, AdamState> {\n  private _stepCount = 0;\n\n  get stepCount(): number {\n    return this._stepCount;\n  }\n\n  constructor(\n    params: Iterable<GradTensor> | ReadonlyArray<ParamGroup<AdamOptions>>,\n    options: {\n      readonly lr?: number;\n      readonly beta1?: number;\n      readonly beta2?: number;\n      readonly eps?: number;\n      readonly weightDecay?: number;\n      readonly amsgrad?: boolean;\n    } = {}\n  ) {\n    const defaults = {\n      lr: options.lr ?? 0.001,\n      beta1: options.beta1 ?? 0.9,\n      beta2: options.beta2 ?? 0.999,\n      eps: options.eps ?? 1e-8,\n      weightDecay: options.weightDecay ?? 0,\n      amsgrad: options.amsgrad ?? false,\n    };\n\n    super(params, defaults);\n\n    assertFiniteNonNegative(\"learning rate\", defaults.lr);\n    assertInRange(\"beta1\", defaults.beta1, 0, 1);\n    assertInRange(\"beta2\", defaults.beta2, 0, 1);\n    assertFinitePositive(\"epsilon\", defaults.eps);\n    assertFiniteNonNegative(\"weight_decay value\", defaults.weightDecay);\n  }\n\n  /**\n   * Get the current learning rate.\n   *\n   * @param groupIdx - Parameter group index (default: 0)\n   * @returns Current learning rate\n   */\n  getLearningRate(groupIdx = 0): number {\n    const group = this.paramGroups[groupIdx];\n    if (!group) {\n      throw new InvalidParameterError(\n        `Invalid group index: ${groupIdx} (valid range: [0, ${this.paramGroups.length}))`,\n        \"groupIdx\",\n        groupIdx\n      );\n    }\n    return group.options.lr;\n  }\n\n  /**\n   * Set the learning rate for all parameter groups.\n   *\n   * @param lr - New learning rate\n   */\n  setLearningRate(lr: number): void {\n    assertFiniteNonNegative(\"learning rate\", lr);\n    for (const group of this.paramGroups) {\n      group.options.lr = lr;\n    }\n  }\n\n  protected isState(state: Record<string, unknown>): state is AdamState {\n    const hasRequired =\n      typeof state[\"step\"] === \"number\" &&\n      state[\"expAvg\"] instanceof Float64Array &&\n      state[\"expAvgSq\"] instanceof Float64Array;\n    if (!hasRequired) return false;\n    if (state[\"maxExpAvgSq\"] !== undefined && !(state[\"maxExpAvgSq\"] instanceof Float64Array)) {\n      return false;\n    }\n    return true;\n  }\n\n  step(closure?: () => number): number | undefined {\n    let loss: number | undefined;\n\n    if (closure) {\n      loss = closure();\n    }\n\n    this._stepCount++;\n\n    for (const group of this.paramGroups) {\n      const { lr, beta1, beta2, eps, weightDecay, amsgrad } = group.options;\n\n      assertFiniteNonNegative(\"learning rate\", lr);\n      assertInRange(\"beta1\", beta1, 0, 1);\n      assertInRange(\"beta2\", beta2, 0, 1);\n      assertFinitePositive(\"epsilon\", eps);\n      assertFiniteNonNegative(\"weight_decay value\", weightDecay);\n\n      for (const param of group.params) {\n        const {\n          grad: gradData,\n          gradOffset,\n          param: paramData,\n          paramOffset,\n        } = assertHasGradFloat(param, \"Adam\");\n        const size = param.tensor.size;\n\n        const existing = this.state.get(param);\n        const state =\n          existing ??\n          (() => {\n            const next = {\n              step: 0,\n              expAvg: new Float64Array(size),\n              expAvgSq: new Float64Array(size),\n              ...(amsgrad ? { maxExpAvgSq: new Float64Array(size) } : {}),\n            };\n            this.state.set(param, next);\n            return next;\n          })();\n\n        // Validate state buffer sizes\n        assertBufferSize(state.expAvg, size, \"Adam expAvg\");\n        assertBufferSize(state.expAvgSq, size, \"Adam expAvgSq\");\n        if (amsgrad && state.maxExpAvgSq) {\n          assertBufferSize(state.maxExpAvgSq, size, \"Adam maxExpAvgSq\");\n        }\n\n        state.step += 1;\n\n        // Bias correction\n        const biasCorrection1 = 1 - beta1 ** state.step;\n        const biasCorrection2 = 1 - beta2 ** state.step;\n\n        const stepSize = lr / biasCorrection1;\n\n        for (let i = 0; i < size; i++) {\n          const gi0 = safeArrayAccess(gradData, gradOffset + i, \"Adam gradient\");\n          const pi = safeArrayAccess(paramData, paramOffset + i, \"Adam parameter\");\n          assertFinite(\"gradient\", gi0);\n          assertFinite(\"parameter\", pi);\n\n          // Optional L2 weight decay (classic Adam style)\n          const gi = weightDecay !== 0 ? gi0 + weightDecay * pi : gi0;\n\n          const m = safeArrayAccess(state.expAvg, i, \"Adam expAvg\");\n          const v = safeArrayAccess(state.expAvgSq, i, \"Adam expAvgSq\");\n\n          const mNew = beta1 * m + (1 - beta1) * gi;\n          const vNew = beta2 * v + (1 - beta2) * gi * gi;\n\n          state.expAvg[i] = mNew;\n          state.expAvgSq[i] = vNew;\n\n          let denomSq = vNew;\n          if (amsgrad) {\n            const maxBuf = state.maxExpAvgSq;\n            if (!maxBuf) {\n              throw new DeepboxError(\"Internal error: AMSGrad enabled but maxExpAvgSq is missing\");\n            }\n            const maxV = Math.max(safeArrayAccess(maxBuf, i, \"Adam maxExpAvgSq\"), vNew);\n            maxBuf[i] = maxV;\n            denomSq = maxV;\n          }\n\n          const denom = Math.sqrt(denomSq / biasCorrection2) + eps;\n          paramData[paramOffset + i] = pi - stepSize * (mNew / denom);\n        }\n      }\n    }\n\n    return loss;\n  }\n}\n","import { DeepboxError, InvalidParameterError } from \"../../core\";\nimport type { GradTensor } from \"../../ndarray\";\nimport {\n  assertBufferSize,\n  assertFinite,\n  assertFiniteNonNegative,\n  assertFinitePositive,\n  assertHasGradFloat,\n  assertInRange,\n  safeArrayAccess,\n} from \"../_internal\";\nimport { Optimizer, type ParamGroup } from \"../Optimizer\";\n\n/**\n * Options for the AdamW optimizer.\n *\n * @property lr - Learning rate (step size)\n * @property beta1 - Exponential decay rate for first moment estimates\n * @property beta2 - Exponential decay rate for second moment estimates\n * @property eps - Small constant for numerical stability\n * @property weightDecay - Weight decay coefficient (L2 penalty)\n * @property amsgrad - Whether to use the AMSGrad variant\n */\ntype AdamWOptions = {\n  lr: number;\n  beta1: number;\n  beta2: number;\n  eps: number;\n  weightDecay: number;\n  amsgrad: boolean;\n};\n\n/**\n * State maintained per parameter by AdamW.\n *\n * @property step - Number of optimization steps taken\n * @property expAvg - Exponentially weighted average of gradients (first moment)\n * @property expAvgSq - Exponentially weighted average of squared gradients (second moment)\n * @property maxExpAvgSq - Maximum of exponentially weighted average of squared gradients (AMSGrad only)\n */\ntype AdamWState = {\n  step: number;\n  expAvg: Float64Array;\n  expAvgSq: Float64Array;\n  maxExpAvgSq?: Float64Array;\n};\n\n/**\n * AdamW (Adam with decoupled Weight decay) optimizer.\n *\n * AdamW fixes the weight decay implementation in Adam by decoupling it from the\n * gradient-based update. This leads to better generalization and is the recommended\n * variant for most applications.\n *\n * @example\n * ```ts\n * import { AdamW } from 'deepbox/optim';\n *\n * const optimizer = new AdamW(model.parameters(), {\n *   lr: 0.001,\n *   weightDecay: 0.01,  // Typical value for AdamW\n *   beta1: 0.9,\n *   beta2: 0.999\n * });\n *\n * // Training loop\n * for (let epoch = 0; epoch < numEpochs; epoch++) {\n *   optimizer.zeroGrad();\n *   // ...\n *   optimizer.step();\n * }\n * ```\n *\n * @category Optimizers\n */\nexport class AdamW extends Optimizer<AdamWOptions, AdamWState> {\n  /** Internal counter tracking total number of optimization steps */\n  private _stepCount = 0;\n\n  /**\n   * Get the total number of optimization steps performed.\n   *\n   * @returns Number of steps taken\n   */\n  get stepCount(): number {\n    return this._stepCount;\n  }\n\n  /**\n   * Create a new AdamW optimizer.\n   *\n   * @param params - Iterable of parameters or parameter groups to optimize\n   * @param options - Optimization options\n   * @param options.lr - Learning rate (default: 0.001)\n   * @param options.beta1 - First moment decay rate (default: 0.9)\n   * @param options.beta2 - Second moment decay rate (default: 0.999)\n   * @param options.eps - Numerical stability constant (default: 1e-8)\n   * @param options.weightDecay - Weight decay coefficient (default: 0.01)\n   * @param options.amsgrad - Enable AMSGrad variant (default: false)\n   * @throws {InvalidParameterError} If a parameter is invalid\n   */\n  constructor(\n    params: Iterable<GradTensor> | ReadonlyArray<ParamGroup<AdamWOptions>>,\n    options: {\n      readonly lr?: number;\n      readonly beta1?: number;\n      readonly beta2?: number;\n      readonly eps?: number;\n      readonly weightDecay?: number;\n      readonly amsgrad?: boolean;\n    } = {}\n  ) {\n    // Set default values for all options\n    const defaults = {\n      lr: options.lr ?? 0.001,\n      beta1: options.beta1 ?? 0.9,\n      beta2: options.beta2 ?? 0.999,\n      eps: options.eps ?? 1e-8,\n      weightDecay: options.weightDecay ?? 0.01, // Higher default than Adam\n      amsgrad: options.amsgrad ?? false,\n    };\n\n    super(params, defaults);\n\n    // Validate all hyperparameters\n    assertFiniteNonNegative(\"learning rate\", defaults.lr);\n    assertInRange(\"beta1\", defaults.beta1, 0, 1);\n    assertInRange(\"beta2\", defaults.beta2, 0, 1);\n    assertFinitePositive(\"epsilon\", defaults.eps);\n    assertFiniteNonNegative(\"weight_decay value\", defaults.weightDecay);\n  }\n\n  /**\n   * Get the current learning rate.\n   *\n   * @param groupIdx - Parameter group index (default: 0)\n   * @returns Current learning rate\n   */\n  getLearningRate(groupIdx = 0): number {\n    const group = this.paramGroups[groupIdx];\n    if (!group) {\n      throw new InvalidParameterError(\n        `Invalid group index: ${groupIdx} (valid range: [0, ${this.paramGroups.length}))`,\n        \"groupIdx\",\n        groupIdx\n      );\n    }\n    return group.options.lr;\n  }\n\n  /**\n   * Set the learning rate for all parameter groups.\n   *\n   * @param lr - New learning rate\n   */\n  setLearningRate(lr: number): void {\n    assertFiniteNonNegative(\"learning rate\", lr);\n    for (const group of this.paramGroups) {\n      group.options.lr = lr;\n    }\n  }\n\n  /**\n   * Perform a single optimization step (parameter update).\n   *\n   * Implements the AdamW update rule with decoupled weight decay.\n   *\n   * @param closure - Optional closure that reevaluates the model and returns the loss\n   * @returns Loss value if closure is provided, undefined otherwise\n   */\n  protected isState(state: Record<string, unknown>): state is AdamWState {\n    const hasRequired =\n      typeof state[\"step\"] === \"number\" &&\n      state[\"expAvg\"] instanceof Float64Array &&\n      state[\"expAvgSq\"] instanceof Float64Array;\n    if (!hasRequired) return false;\n    if (state[\"maxExpAvgSq\"] !== undefined && !(state[\"maxExpAvgSq\"] instanceof Float64Array)) {\n      return false;\n    }\n    return true;\n  }\n\n  step(closure?: () => number): number | undefined {\n    let loss: number | undefined;\n\n    // Evaluate closure if provided (for algorithms like LBFGS)\n    if (closure) {\n      loss = closure();\n    }\n\n    // Increment global step counter\n    this._stepCount++;\n\n    // Update each parameter group\n    for (const group of this.paramGroups) {\n      const { lr, beta1, beta2, eps, weightDecay, amsgrad } = group.options;\n\n      // Re-validate hyperparameters (they might have been changed)\n      assertFiniteNonNegative(\"learning rate\", lr);\n      assertInRange(\"beta1\", beta1, 0, 1);\n      assertInRange(\"beta2\", beta2, 0, 1);\n      assertFinitePositive(\"epsilon\", eps);\n      assertFiniteNonNegative(\"weight_decay value\", weightDecay);\n\n      // Update each parameter in the group\n      for (const param of group.params) {\n        // Get gradient and validate\n        const {\n          grad,\n          gradOffset,\n          param: pData,\n          paramOffset: pOff,\n        } = assertHasGradFloat(param, \"AdamW\");\n        const size = param.tensor.size;\n\n        // Get or initialize optimizer state for this parameter\n        const existing = this.state.get(param);\n        const state =\n          existing ??\n          (() => {\n            // Initialize state on first use\n            const next = {\n              step: 0,\n              expAvg: new Float64Array(size), // First moment\n              expAvgSq: new Float64Array(size), // Second moment\n              ...(amsgrad ? { maxExpAvgSq: new Float64Array(size) } : {}), // AMSGrad buffer\n            };\n            this.state.set(param, next);\n            return next;\n          })();\n\n        // Validate state buffer sizes\n        assertBufferSize(state.expAvg, size, \"AdamW expAvg\");\n        assertBufferSize(state.expAvgSq, size, \"AdamW expAvgSq\");\n        if (amsgrad && state.maxExpAvgSq) {\n          assertBufferSize(state.maxExpAvgSq, size, \"AdamW maxExpAvgSq\");\n        }\n\n        // Increment per-parameter step counter\n        state.step += 1;\n\n        // Compute bias correction terms\n        const biasCorrection1 = 1 - beta1 ** state.step;\n        const biasCorrection2 = 1 - beta2 ** state.step;\n\n        // Compute step size with bias correction\n        const stepSize = lr / biasCorrection1;\n\n        // Update each element of the parameter\n        for (let i = 0; i < size; i++) {\n          // Get current gradient and parameter values\n          const gi = safeArrayAccess(grad, gradOffset + i, \"AdamW gradient\");\n          const pi = safeArrayAccess(pData, pOff + i, \"AdamW parameter\");\n\n          // Validate values are finite\n          assertFinite(\"gradient\", gi);\n          assertFinite(\"parameter\", pi);\n\n          // Get current moment estimates\n          const m = safeArrayAccess(state.expAvg, i, \"AdamW expAvg\");\n          const v = safeArrayAccess(state.expAvgSq, i, \"AdamW expAvgSq\");\n\n          // Update biased first moment estimate: m(t) = β1 * m(t-1) + (1 - β1) * g(t)\n          const mNew = beta1 * m + (1 - beta1) * gi;\n\n          // Update biased second raw moment estimate: v(t) = β2 * v(t-1) + (1 - β2) * g(t)^2\n          const vNew = beta2 * v + (1 - beta2) * gi * gi;\n\n          // Store updated moments\n          state.expAvg[i] = mNew;\n          state.expAvgSq[i] = vNew;\n\n          // Determine which second moment to use (AMSGrad or standard)\n          let denomSq = vNew;\n          if (amsgrad) {\n            const maxBuf = state.maxExpAvgSq;\n            if (!maxBuf) {\n              throw new DeepboxError(\"Internal error: AMSGrad enabled but maxExpAvgSq is missing\");\n            }\n            // AMSGrad: use maximum of all past second moments\n            const maxV = Math.max(safeArrayAccess(maxBuf, i, \"AdamW maxExpAvgSq\"), vNew);\n            maxBuf[i] = maxV;\n            denomSq = maxV;\n          }\n\n          // Compute denominator with bias correction: √(v̂(t)) + ε\n          const denom = Math.sqrt(denomSq / biasCorrection2) + eps;\n\n          // AdamW update: θ(t+1) = θ(t) - lr * (m̂(t) / denom + λ * θ(t))\n          // Note: weight decay is applied directly to parameters (decoupled)\n          pData[pOff + i] = pi - stepSize * (mNew / denom) - lr * weightDecay * pi;\n        }\n      }\n    }\n\n    return loss;\n  }\n}\n","import { InvalidParameterError } from \"../../core\";\nimport type { GradTensor } from \"../../ndarray\";\nimport {\n  assertBufferSize,\n  assertFinite,\n  assertFiniteNonNegative,\n  assertFinitePositive,\n  assertHasGradFloat,\n  assertInRange,\n  safeArrayAccess,\n} from \"../_internal\";\nimport { Optimizer, type ParamGroup } from \"../Optimizer\";\n\ntype NadamOptions = {\n  lr: number;\n  readonly beta1: number;\n  readonly beta2: number;\n  readonly eps: number;\n  readonly weightDecay: number;\n  readonly momentumDecay: number;\n};\n\ntype NadamState = {\n  step: number;\n  expAvg: Float64Array;\n  expAvgSq: Float64Array;\n  muProduct: number;\n};\n\n/**\n * Nadam (Nesterov-accelerated Adam) optimizer.\n *\n * Implements Nadam algorithm - combines Adam's adaptive learning rates with\n * Nesterov momentum for potentially faster convergence. Nadam applies Nesterov\n * acceleration to the momentum term, providing a \"look-ahead\" gradient.\n *\n * @example\n * ```ts\n * import { Nadam } from 'deepbox/optim';\n *\n * const optimizer = new Nadam(model.parameters(), {\n *   lr: 0.002,\n *   beta1: 0.9,\n *   beta2: 0.999\n * });\n *\n * // Training loop\n * for (let epoch = 0; epoch < numEpochs; epoch++) {\n *   optimizer.zeroGrad();\n *   // ...\n *   optimizer.step();\n * }\n * ```\n *\n * @category Optimizers\n */\nexport class Nadam extends Optimizer<NadamOptions, NadamState> {\n  private _stepCount = 0;\n\n  get stepCount(): number {\n    return this._stepCount;\n  }\n  constructor(\n    params: Iterable<GradTensor> | ReadonlyArray<ParamGroup<NadamOptions>>,\n    options: {\n      readonly lr?: number;\n      readonly beta1?: number;\n      readonly beta2?: number;\n      readonly eps?: number;\n      readonly weightDecay?: number;\n      readonly momentumDecay?: number;\n    } = {}\n  ) {\n    const defaults = {\n      lr: options.lr ?? 0.002,\n      beta1: options.beta1 ?? 0.9,\n      beta2: options.beta2 ?? 0.999,\n      eps: options.eps ?? 1e-8,\n      weightDecay: options.weightDecay ?? 0,\n      momentumDecay: options.momentumDecay ?? 0.004,\n    };\n\n    super(params, defaults);\n\n    // Validate hyperparameters\n    assertFiniteNonNegative(\"learning rate\", defaults.lr);\n    assertInRange(\"beta1\", defaults.beta1, 0, 1);\n    assertInRange(\"beta2\", defaults.beta2, 0, 1);\n    assertFinitePositive(\"epsilon\", defaults.eps);\n    assertFiniteNonNegative(\"weight_decay value\", defaults.weightDecay);\n    assertFiniteNonNegative(\"momentum_decay\", defaults.momentumDecay);\n  }\n\n  /**\n   * Get the current learning rate.\n   *\n   * @param groupIdx - Parameter group index (default: 0)\n   * @returns Current learning rate\n   */\n  getLearningRate(groupIdx = 0): number {\n    const group = this.paramGroups[groupIdx];\n    if (!group) {\n      throw new InvalidParameterError(\n        `Invalid group index: ${groupIdx} (valid range: [0, ${this.paramGroups.length}))`,\n        \"groupIdx\",\n        groupIdx\n      );\n    }\n    return group.options.lr;\n  }\n\n  /**\n   * Set the learning rate for all parameter groups.\n   *\n   * @param lr - New learning rate\n   */\n  setLearningRate(lr: number): void {\n    assertFiniteNonNegative(\"learning rate\", lr);\n    for (const group of this.paramGroups) {\n      group.options.lr = lr;\n    }\n  }\n\n  protected isState(state: Record<string, unknown>): state is NadamState {\n    return (\n      typeof state[\"step\"] === \"number\" &&\n      state[\"expAvg\"] instanceof Float64Array &&\n      state[\"expAvgSq\"] instanceof Float64Array &&\n      typeof state[\"muProduct\"] === \"number\"\n    );\n  }\n\n  step(closure?: () => number): number | undefined {\n    let loss: number | undefined;\n\n    if (closure) {\n      loss = closure();\n    }\n\n    // Increment global step counter\n    this._stepCount++;\n\n    for (const group of this.paramGroups) {\n      const { lr, beta1, beta2, eps, weightDecay, momentumDecay } = group.options;\n\n      // Re-validate hyperparameters\n      assertFiniteNonNegative(\"learning rate\", lr);\n      assertInRange(\"beta1\", beta1, 0, 1);\n      assertInRange(\"beta2\", beta2, 0, 1);\n      assertFinitePositive(\"epsilon\", eps);\n      assertFiniteNonNegative(\"weight_decay value\", weightDecay);\n      assertFiniteNonNegative(\"momentum_decay\", momentumDecay);\n\n      for (const param of group.params) {\n        const {\n          grad: gradData,\n          gradOffset: gOff,\n          param: pData,\n          paramOffset: pOff,\n        } = assertHasGradFloat(param, \"Nadam\");\n        const size = param.tensor.size;\n\n        // Initialize state if needed\n        let state = this.state.get(param);\n        if (!state) {\n          state = {\n            step: 0,\n            expAvg: new Float64Array(size),\n            expAvgSq: new Float64Array(size),\n            muProduct: 1,\n          };\n          this.state.set(param, state);\n        }\n\n        // Validate state buffer sizes\n        assertBufferSize(state.expAvg, size, \"Nadam expAvg\");\n        assertBufferSize(state.expAvgSq, size, \"Nadam expAvgSq\");\n\n        state.step++;\n        const t = state.step;\n\n        const biasCorrection2 = 1 - beta2 ** t;\n        const mu = beta1 * (1 - 0.5 * 0.96 ** (t * momentumDecay));\n        const muNext = beta1 * (1 - 0.5 * 0.96 ** ((t + 1) * momentumDecay));\n        const muProduct = state.muProduct * mu;\n        const muProductNext = muProduct * muNext;\n        state.muProduct = muProduct;\n\n        for (let i = 0; i < size; i++) {\n          const gi0 = safeArrayAccess(gradData, gOff + i, \"Nadam gradient\");\n          const pi = safeArrayAccess(pData, pOff + i, \"Nadam parameter\");\n          assertFinite(\"gradient\", gi0);\n          assertFinite(\"parameter\", pi);\n\n          // Apply weight decay\n          const gi = weightDecay !== 0 ? gi0 + weightDecay * pi : gi0;\n\n          // Update biased first moment estimate: m(t) = β1 * m(t-1) + (1 - β1) * g(t)\n          const m = safeArrayAccess(state.expAvg, i, \"Nadam expAvg\");\n          const mNew = beta1 * m + (1 - beta1) * gi;\n          state.expAvg[i] = mNew;\n\n          // Update biased second moment estimate: v(t) = β2 * v(t-1) + (1 - β2) * g(t)²\n          const v = safeArrayAccess(state.expAvgSq, i, \"Nadam expAvgSq\");\n          const vNew = beta2 * v + (1 - beta2) * gi * gi;\n          state.expAvgSq[i] = vNew;\n\n          const denom = Math.sqrt(vNew / biasCorrection2) + eps;\n          const mHatNext = mNew / (1 - muProductNext);\n          const gHat = gi / (1 - muProduct);\n          const mNesterov = muNext * mHatNext + (1 - mu) * gHat;\n          pData[pOff + i] = pi - (lr * mNesterov) / denom;\n        }\n      }\n    }\n\n    return loss;\n  }\n}\n","import { InvalidParameterError } from \"../../core\";\nimport type { GradTensor } from \"../../ndarray\";\nimport {\n  assertBufferSize,\n  assertFinite,\n  assertFiniteNonNegative,\n  assertFinitePositive,\n  assertHasGradFloat,\n  safeArrayAccess,\n} from \"../_internal\";\nimport { Optimizer, type ParamGroup } from \"../Optimizer\";\n\n/**\n * Options for the RMSprop optimizer.\n *\n * @property lr - Learning rate (step size)\n * @property alpha - Smoothing constant for moving average of squared gradients\n * @property eps - Small constant for numerical stability\n * @property weightDecay - Weight decay coefficient (L2 penalty)\n * @property momentum - Momentum factor\n * @property centered - Whether to use centered RMSprop variant\n */\ntype RMSpropOptions = {\n  lr: number;\n  alpha: number;\n  eps: number;\n  weightDecay: number;\n  momentum: number;\n  centered: boolean;\n};\n\n/**\n * State maintained per parameter by RMSprop.\n *\n * @property squareAvg - Exponentially weighted average of squared gradients\n * @property momentumBuffer - Momentum buffer (if momentum > 0)\n * @property gradAvg - Exponentially weighted average of gradients (centered variant only)\n */\ntype RMSpropState = {\n  squareAvg: Float64Array;\n  momentumBuffer?: Float64Array;\n  gradAvg?: Float64Array;\n};\n\n/**\n * RMSprop (Root Mean Square Propagation) optimizer.\n *\n * RMSprop adapts the learning rate for each parameter by dividing by a running\n * average of recent gradient magnitudes. This helps with non-stationary objectives\n * and is particularly effective for RNNs.\n *\n * @example\n * ```ts\n * import { RMSprop } from 'deepbox/optim';\n *\n * const optimizer = new RMSprop(model.parameters(), {\n *   lr: 0.01,\n *   alpha: 0.99,\n *   momentum: 0.9,\n *   centered: true\n * });\n *\n * // Training loop\n * for (let epoch = 0; epoch < numEpochs; epoch++) {\n *   optimizer.zeroGrad();\n *   // ...\n *   optimizer.step();\n * }\n * ```\n *\n * @category Optimizers\n */\nexport class RMSprop extends Optimizer<RMSpropOptions, RMSpropState> {\n  /** Internal counter tracking total number of optimization steps */\n  private _stepCount = 0;\n\n  /**\n   * Get the total number of optimization steps performed.\n   *\n   * @returns Number of steps taken\n   */\n  get stepCount(): number {\n    return this._stepCount;\n  }\n  /**\n   * Create a new RMSprop optimizer.\n   *\n   * @param params - Iterable of parameters or parameter groups to optimize\n   * @param options - Optimization options\n   * @param options.lr - Learning rate (default: 0.01)\n   * @param options.alpha - Smoothing constant (default: 0.99)\n   * @param options.eps - Numerical stability constant (default: 1e-8)\n   * @param options.weightDecay - Weight decay coefficient (default: 0)\n   * @param options.momentum - Momentum factor (default: 0)\n   * @param options.centered - Use centered variant (default: false)\n   * @throws {InvalidParameterError} If a parameter is invalid\n   */\n  constructor(\n    params: Iterable<GradTensor> | ReadonlyArray<ParamGroup<RMSpropOptions>>,\n    options: {\n      readonly lr?: number;\n      readonly alpha?: number;\n      readonly eps?: number;\n      readonly weightDecay?: number;\n      readonly momentum?: number;\n      readonly centered?: boolean;\n    } = {}\n  ) {\n    // Set default values for all options\n    const defaults = {\n      lr: options.lr ?? 0.01,\n      alpha: options.alpha ?? 0.99,\n      eps: options.eps ?? 1e-8,\n      weightDecay: options.weightDecay ?? 0,\n      momentum: options.momentum ?? 0,\n      centered: options.centered ?? false,\n    };\n\n    super(params, defaults);\n\n    // Validate all hyperparameters\n    assertFiniteNonNegative(\"learning rate\", defaults.lr);\n    if (!Number.isFinite(defaults.alpha) || defaults.alpha < 0 || defaults.alpha > 1) {\n      throw new InvalidParameterError(\n        `Invalid alpha: ${defaults.alpha} (must be in range [0, 1])`,\n        \"alpha\",\n        defaults.alpha\n      );\n    }\n    assertFinitePositive(\"epsilon\", defaults.eps);\n    assertFiniteNonNegative(\"weight_decay value\", defaults.weightDecay);\n    assertFiniteNonNegative(\"momentum value\", defaults.momentum);\n  }\n\n  /**\n   * Get the current learning rate.\n   *\n   * @param groupIdx - Parameter group index (default: 0)\n   * @returns Current learning rate\n   */\n  getLearningRate(groupIdx = 0): number {\n    const group = this.paramGroups[groupIdx];\n    if (!group) {\n      throw new InvalidParameterError(\n        `Invalid group index: ${groupIdx} (valid range: [0, ${this.paramGroups.length}))`,\n        \"groupIdx\",\n        groupIdx\n      );\n    }\n    return group.options.lr;\n  }\n\n  /**\n   * Set the learning rate for all parameter groups.\n   *\n   * @param lr - New learning rate\n   */\n  setLearningRate(lr: number): void {\n    assertFiniteNonNegative(\"learning rate\", lr);\n    for (const group of this.paramGroups) {\n      group.options.lr = lr;\n    }\n  }\n\n  protected isState(state: Record<string, unknown>): state is RMSpropState {\n    if (!(state[\"squareAvg\"] instanceof Float64Array)) return false;\n    if (\n      state[\"momentumBuffer\"] !== undefined &&\n      !(state[\"momentumBuffer\"] instanceof Float64Array)\n    ) {\n      return false;\n    }\n    if (state[\"gradAvg\"] !== undefined && !(state[\"gradAvg\"] instanceof Float64Array)) {\n      return false;\n    }\n    return true;\n  }\n\n  step(closure?: () => number): number | undefined {\n    let loss: number | undefined;\n\n    // Evaluate closure if provided\n    if (closure) {\n      loss = closure();\n    }\n\n    // Increment global step counter\n    this._stepCount++;\n\n    // Update each parameter group\n    for (const group of this.paramGroups) {\n      const { lr, alpha, eps, weightDecay, momentum, centered } = group.options;\n\n      // Re-validate hyperparameters\n      assertFiniteNonNegative(\"learning rate\", lr);\n      if (!Number.isFinite(alpha) || alpha < 0 || alpha > 1) {\n        throw new InvalidParameterError(\n          `Invalid alpha: ${alpha} (must be in range [0, 1])`,\n          \"alpha\",\n          alpha\n        );\n      }\n      assertFinitePositive(\"epsilon\", eps);\n      assertFiniteNonNegative(\"weight_decay value\", weightDecay);\n      assertFiniteNonNegative(\"momentum value\", momentum);\n\n      // Update each parameter in the group\n      for (const param of group.params) {\n        // Get gradient and validate\n        const {\n          grad: gradData,\n          gradOffset: gOff,\n          param: pData,\n          paramOffset: pOff,\n        } = assertHasGradFloat(param, \"RMSprop\");\n        const size = param.tensor.size;\n\n        // Get or initialize optimizer state for this parameter\n        let state = this.state.get(param);\n        if (!state) {\n          state = {\n            squareAvg: new Float64Array(size),\n          };\n          this.state.set(param, state);\n        }\n\n        // Initialize momentum buffer if needed\n        if (momentum > 0 && !state.momentumBuffer) {\n          state.momentumBuffer = new Float64Array(size);\n        }\n\n        // Initialize gradient average buffer if centered variant is used\n        if (centered && !state.gradAvg) {\n          state.gradAvg = new Float64Array(size);\n        }\n\n        // Validate state buffer sizes\n        assertBufferSize(state.squareAvg, size, \"RMSprop squareAvg\");\n        if (momentum > 0 && state.momentumBuffer) {\n          assertBufferSize(state.momentumBuffer, size, \"RMSprop momentumBuffer\");\n        }\n        if (centered && state.gradAvg) {\n          assertBufferSize(state.gradAvg, size, \"RMSprop gradAvg\");\n        }\n\n        // Update each element of the parameter\n        for (let i = 0; i < size; i++) {\n          // Get current gradient and parameter values\n          const gi = safeArrayAccess(gradData, gOff + i, \"RMSprop gradient\");\n          const pi = safeArrayAccess(pData, pOff + i, \"RMSprop parameter\");\n\n          // Validate values are finite\n          assertFinite(\"gradient\", gi);\n          assertFinite(\"parameter\", pi);\n\n          // Apply weight decay to gradient if specified\n          let grad = gi;\n          if (weightDecay !== 0) {\n            grad = grad + weightDecay * pi;\n          }\n\n          // Update exponentially weighted average of squared gradients\n          // v(t) = α * v(t-1) + (1 - α) * g(t)^2\n          const sqAvg = safeArrayAccess(state.squareAvg, i, \"RMSprop squareAvg\");\n          const sqAvgNew = alpha * sqAvg + (1 - alpha) * grad * grad;\n          state.squareAvg[i] = sqAvgNew;\n\n          // Compute adaptive learning rate denominator\n          let avg = sqAvgNew;\n\n          // Centered variant: subtract squared mean of gradients\n          if (centered) {\n            const gAvg = state.gradAvg ? safeArrayAccess(state.gradAvg, i, \"RMSprop gradAvg\") : 0;\n            // Update exponentially weighted average of gradients\n            const gAvgNew = alpha * gAvg + (1 - alpha) * grad;\n            if (state.gradAvg) state.gradAvg[i] = gAvgNew;\n            // Use variance instead of second moment\n            avg = sqAvgNew - gAvgNew * gAvgNew;\n          }\n\n          const denom = centered ? Math.sqrt(Math.max(avg, 0) + eps) : Math.sqrt(avg) + eps;\n          const normalizedGrad = grad / denom;\n\n          // Apply momentum if specified\n          if (momentum > 0) {\n            const buf = state.momentumBuffer\n              ? safeArrayAccess(state.momentumBuffer, i, \"RMSprop momentumBuffer\")\n              : 0;\n            const bufNew = momentum * buf + normalizedGrad;\n            if (state.momentumBuffer) state.momentumBuffer[i] = bufNew;\n            // Update parameter with momentum\n            pData[pOff + i] = pi - lr * bufNew;\n          } else {\n            // Update parameter without momentum\n            pData[pOff + i] = pi - lr * normalizedGrad;\n          }\n        }\n      }\n    }\n\n    return loss;\n  }\n}\n","import { InvalidParameterError } from \"../../core\";\nimport type { GradTensor } from \"../../ndarray\";\nimport {\n  assertFinite,\n  assertFiniteNonNegative,\n  assertHasGradFloat,\n  safeArrayAccess,\n} from \"../_internal\";\nimport { Optimizer, type ParamGroup } from \"../Optimizer\";\n\ntype SGDOptions = {\n  lr: number;\n  momentum: number;\n  dampening: number;\n  weightDecay: number;\n  nesterov: boolean;\n};\n\ntype SGDState = {\n  momentumBuffer?: Float64Array;\n};\n\n/**\n * Stochastic Gradient Descent (SGD) optimizer.\n *\n * Implements vanilla SGD with optional momentum, weight decay, and Nesterov acceleration.\n *\n * @example\n * ```ts\n * import { SGD } from 'deepbox/optim';\n * import { Module } from 'deepbox/nn';\n *\n * const model: Module = ...;\n * const optimizer = new SGD(model.parameters(), {\n *   lr: 0.01,\n *   momentum: 0.9,\n *   weightDecay: 5e-4,\n *   nesterov: true\n * });\n *\n * // Training loop\n * for (let epoch = 0; epoch < numEpochs; epoch++) {\n *   for (const [inputs, targets] of dataLoader) {\n *     optimizer.zeroGrad();\n *     const outputs = model.forward(inputs);\n *     const loss = criterion(outputs, targets);\n *     loss.backward();\n *     optimizer.step();\n *   }\n * }\n * ```\n *\n * @category Optimizers\n */\nexport class SGD extends Optimizer<SGDOptions, SGDState> {\n  /** Internal counter tracking total number of optimization steps */\n  private _stepCount = 0;\n\n  get stepCount(): number {\n    return this._stepCount;\n  }\n  /**\n   * Create a new SGD optimizer.\n   *\n   * @param params - Iterable of parameters or parameter groups to optimize\n   * @param options - Optimization options\n   * @param options.lr - Learning rate (default: 0.01)\n   * @param options.momentum - Momentum factor (default: 0)\n   * @param options.dampening - Dampening for momentum (default: 0)\n   * @param options.weightDecay - Weight decay (L2 penalty) (default: 0)\n   * @param options.nesterov - Enable Nesterov momentum (default: false)\n   */\n  constructor(\n    params: Iterable<GradTensor> | ReadonlyArray<ParamGroup<SGDOptions>>,\n    options: {\n      readonly lr?: number;\n      readonly momentum?: number;\n      readonly dampening?: number;\n      readonly weightDecay?: number;\n      readonly nesterov?: boolean;\n    } = {}\n  ) {\n    const defaults = {\n      lr: options.lr ?? 0.01,\n      momentum: options.momentum ?? 0,\n      dampening: options.dampening ?? 0,\n      weightDecay: options.weightDecay ?? 0,\n      nesterov: options.nesterov ?? false,\n    };\n\n    super(params, defaults);\n\n    // Validate options\n    assertFiniteNonNegative(\"learning rate\", defaults.lr);\n    assertFiniteNonNegative(\"momentum value\", defaults.momentum);\n    assertFiniteNonNegative(\"dampening\", defaults.dampening);\n    assertFiniteNonNegative(\"weight_decay value\", defaults.weightDecay);\n    if (defaults.nesterov && (defaults.momentum <= 0 || defaults.dampening !== 0)) {\n      throw new InvalidParameterError(\n        \"Nesterov momentum requires a momentum and zero dampening\",\n        \"nesterov\",\n        {\n          momentum: defaults.momentum,\n          dampening: defaults.dampening,\n          nesterov: defaults.nesterov,\n        }\n      );\n    }\n  }\n\n  /**\n   * Perform a single optimization step.\n   *\n   * Implements the SGD update rule with optional momentum and weight decay.\n   *\n   * @param closure - Optional closure that reevaluates the model and returns the loss\n   * @returns Loss value if closure is provided\n   */\n  protected isState(state: Record<string, unknown>): state is SGDState {\n    if (\n      state[\"momentumBuffer\"] !== undefined &&\n      !(state[\"momentumBuffer\"] instanceof Float64Array)\n    ) {\n      return false;\n    }\n    return true;\n  }\n\n  step(closure?: () => number): number | undefined {\n    let loss: number | undefined;\n\n    // Evaluate loss if closure provided\n    if (closure) {\n      loss = closure();\n    }\n\n    // Increment global step counter\n    this._stepCount++;\n\n    // Update each parameter group\n    for (const group of this.paramGroups) {\n      const { lr, momentum, dampening, weightDecay, nesterov } = group.options;\n\n      assertFiniteNonNegative(\"learning rate\", lr);\n      assertFiniteNonNegative(\"momentum value\", momentum);\n      assertFiniteNonNegative(\"dampening\", dampening);\n      assertFiniteNonNegative(\"weight_decay value\", weightDecay);\n\n      if (nesterov && (momentum <= 0 || dampening !== 0)) {\n        throw new InvalidParameterError(\n          \"Nesterov momentum requires a momentum and zero dampening\",\n          \"nesterov\",\n          { momentum, dampening, nesterov }\n        );\n      }\n\n      for (const param of group.params) {\n        const {\n          grad: gradData,\n          gradOffset,\n          param: paramData,\n          paramOffset,\n        } = assertHasGradFloat(param, \"SGD\");\n        const size = param.tensor.size;\n\n        let state = this.state.get(param);\n        if (!state) {\n          state = {};\n          this.state.set(param, state);\n        }\n\n        // Momentum buffer is stored densely (one value per element).\n        let momentumBuffer: Float64Array | undefined;\n        if (momentum !== 0) {\n          if (!state.momentumBuffer) {\n            state.momentumBuffer = new Float64Array(size);\n          }\n          momentumBuffer = state.momentumBuffer;\n        }\n\n        for (let i = 0; i < size; i++) {\n          const gi = safeArrayAccess(gradData, gradOffset + i, \"SGD gradient\");\n          const pi = safeArrayAccess(paramData, paramOffset + i, \"SGD parameter\");\n          assertFinite(\"gradient\", gi);\n          assertFinite(\"parameter\", pi);\n\n          // d_p = grad + weightDecay * param\n          let d = gi;\n          if (weightDecay !== 0) {\n            d = d + weightDecay * pi;\n          }\n\n          if (momentumBuffer) {\n            const bPrev = safeArrayAccess(momentumBuffer, i, \"SGD momentum buffer\");\n            const bNew = momentum * bPrev + (1 - dampening) * d;\n            momentumBuffer[i] = bNew;\n            d = nesterov ? d + momentum * bNew : bNew;\n          }\n\n          // param -= lr * d\n          paramData[paramOffset + i] = pi - lr * d;\n        }\n      }\n    }\n\n    return loss;\n  }\n\n  /**\n   * Get the current learning rate.\n   *\n   * @param groupIdx - Parameter group index (default: 0)\n   * @returns Current learning rate\n   */\n  getLearningRate(groupIdx = 0): number {\n    const group = this.paramGroups[groupIdx];\n    if (!group) {\n      throw new InvalidParameterError(\n        `Invalid group index: ${groupIdx} (valid range: [0, ${this.paramGroups.length}))`,\n        \"groupIdx\",\n        groupIdx\n      );\n    }\n    return group.options.lr;\n  }\n\n  /**\n   * Set the learning rate for all parameter groups.\n   *\n   * @param lr - New learning rate\n   */\n  setLearningRate(lr: number): void {\n    assertFiniteNonNegative(\"learning rate\", lr);\n    for (const group of this.paramGroups) {\n      group.options.lr = lr;\n    }\n  }\n}\n","// Learning rate schedulers for optimizers\nimport { InvalidParameterError } from \"../core\";\n\n/**\n * Interface for optimizer-like objects that schedulers can work with.\n * This allows schedulers to work with different optimizer implementations.\n * Parameter groups may expose `lr` directly or via `options.lr`.\n */\ninterface SchedulerOptimizer {\n  paramGroups: SchedulerParamGroup[];\n}\n\ntype SchedulerParamGroup = {\n  params: unknown[];\n  lr?: number;\n  options?: Record<string, unknown>;\n};\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n  return typeof value === \"object\" && value !== null;\n}\n\nfunction resolveGroupLr(group: SchedulerParamGroup, index: number) {\n  const options = isRecord(group.options) ? group.options : undefined;\n  const lrValue = group.lr ?? options?.[\"lr\"];\n  if (typeof lrValue !== \"number\" || !Number.isFinite(lrValue) || lrValue < 0) {\n    throw new InvalidParameterError(\n      `optimizer.paramGroups[${index}].lr must be finite and >= 0`,\n      `optimizer.paramGroups[${index}].lr`,\n      lrValue\n    );\n  }\n  return lrValue;\n}\n\nfunction setGroupLr(group: SchedulerParamGroup, lr: number) {\n  if (isRecord(group.options)) {\n    group.options[\"lr\"] = lr;\n  }\n  if (\"lr\" in group) {\n    group.lr = lr;\n  }\n  if (!(\"lr\" in group) && !isRecord(group.options)) {\n    group.lr = lr;\n  }\n}\n\nfunction validateLastEpoch(value: number) {\n  if (!Number.isInteger(value) || value < -1) {\n    throw new InvalidParameterError(\"lastEpoch must be an integer >= -1\", \"lastEpoch\", value);\n  }\n  return value;\n}\n\nfunction validateFiniteNumber(value: number, name: string) {\n  if (!Number.isFinite(value)) {\n    throw new InvalidParameterError(`${name} must be finite`, name, value);\n  }\n  return value;\n}\n\nfunction validatePositiveNumber(value: number, name: string) {\n  if (!Number.isFinite(value) || value <= 0) {\n    throw new InvalidParameterError(`${name} must be > 0`, name, value);\n  }\n  return value;\n}\n\nfunction validatePositiveInteger(value: number, name: string) {\n  if (!Number.isInteger(value) || value <= 0) {\n    throw new InvalidParameterError(`${name} must be a positive integer`, name, value);\n  }\n  return value;\n}\n\nfunction validateNonNegativeNumber(value: number, name: string) {\n  if (!Number.isFinite(value) || value < 0) {\n    throw new InvalidParameterError(`${name} must be >= 0`, name, value);\n  }\n  return value;\n}\n\nfunction validateNonNegativeInteger(value: number, name: string) {\n  if (!Number.isInteger(value) || value < 0) {\n    throw new InvalidParameterError(`${name} must be a non-negative integer`, name, value);\n  }\n  return value;\n}\n\nfunction validateOptimizer(optimizer: SchedulerOptimizer) {\n  if (!optimizer || typeof optimizer !== \"object\" || !Array.isArray(optimizer.paramGroups)) {\n    throw new InvalidParameterError(\n      \"optimizer must expose paramGroups array\",\n      \"optimizer\",\n      optimizer\n    );\n  }\n  if (optimizer.paramGroups.length === 0) {\n    throw new InvalidParameterError(\n      \"optimizer.paramGroups must contain at least one group\",\n      \"optimizer.paramGroups\",\n      optimizer.paramGroups\n    );\n  }\n  for (let i = 0; i < optimizer.paramGroups.length; i++) {\n    const group = optimizer.paramGroups[i];\n    if (!group || typeof group !== \"object\") {\n      throw new InvalidParameterError(\n        `optimizer.paramGroups[${i}] must be an object`,\n        \"optimizer.paramGroups\",\n        group\n      );\n    }\n    if (!Array.isArray(group.params)) {\n      throw new InvalidParameterError(\n        `optimizer.paramGroups[${i}].params must be an array`,\n        `optimizer.paramGroups[${i}].params`,\n        group.params\n      );\n    }\n    resolveGroupLr(group, i);\n  }\n}\n\nfunction validateMilestones(milestones: number[]) {\n  if (!Array.isArray(milestones) || milestones.length === 0) {\n    throw new InvalidParameterError(\n      \"milestones must be a non-empty array of non-negative integers\",\n      \"milestones\",\n      milestones\n    );\n  }\n\n  const sorted = [...milestones].sort((a, b) => a - b);\n  for (let i = 0; i < sorted.length; i++) {\n    const value = sorted[i];\n    if (value === undefined || !Number.isInteger(value) || value < 0) {\n      throw new InvalidParameterError(\n        \"milestones must contain non-negative integers only\",\n        \"milestones\",\n        milestones\n      );\n    }\n    if (i > 0) {\n      const prev = sorted[i - 1];\n      if (prev !== undefined && value <= prev) {\n        throw new InvalidParameterError(\n          \"milestones must be strictly increasing\",\n          \"milestones\",\n          milestones\n        );\n      }\n    }\n  }\n  return sorted;\n}\n\n/**\n * Base class for learning rate schedulers.\n *\n * Learning rate schedulers adjust the learning rate during training according\n * to a predefined schedule. This can help improve convergence and prevent\n * overshooting optimal solutions.\n *\n * @example\n * ```ts\n * import { SGD, StepLR } from 'deepbox/optim';\n *\n * const optimizer = new SGD(model.parameters(), { lr: 0.1 });\n * const scheduler = new StepLR(optimizer, { stepSize: 10, gamma: 0.1 });\n *\n * for (let epoch = 0; epoch < 100; epoch++) {\n *   train();\n *   scheduler.step();\n * }\n * ```\n *\n * @category Optimization\n */\nexport abstract class LRScheduler {\n  protected optimizer: SchedulerOptimizer;\n  protected lastEpoch: number;\n  protected baseLrs: number[];\n\n  constructor(optimizer: SchedulerOptimizer, lastEpoch: number = -1) {\n    validateOptimizer(optimizer);\n    this.lastEpoch = validateLastEpoch(lastEpoch);\n    this.optimizer = optimizer;\n\n    // Store base learning rates from all parameter groups\n    this.baseLrs = optimizer.paramGroups.map((group, index) => resolveGroupLr(group, index));\n  }\n\n  protected initializeFromLastEpoch(lastEpoch: number): void {\n    const validated = validateLastEpoch(lastEpoch);\n    if (validated < 0) {\n      return;\n    }\n    this.lastEpoch = -1;\n    for (let i = 0; i <= validated; i++) {\n      this.step();\n    }\n  }\n\n  /**\n   * Compute the learning rate for the current epoch.\n   * Must be implemented by subclasses.\n   *\n   * @returns Array of learning rates for each parameter group\n   */\n  abstract getLr(): number[];\n\n  /**\n   * Perform a scheduler step, updating learning rates.\n   *\n   * Should be called once per epoch after the optimizer step.\n   */\n  step(): void {\n    this.lastEpoch++;\n    const newLrs = this.getLr();\n\n    for (let i = 0; i < this.optimizer.paramGroups.length; i++) {\n      const group = this.optimizer.paramGroups[i];\n      if (group) {\n        const next = newLrs[i];\n        if (next !== undefined) {\n          setGroupLr(group, next);\n        }\n      }\n    }\n  }\n\n  /**\n   * Get the current learning rates for all parameter groups.\n   */\n  getLastLr(): number[] {\n    return this.optimizer.paramGroups.map((group, index) => resolveGroupLr(group, index));\n  }\n\n  /**\n   * Get current epoch number.\n   */\n  get epoch() {\n    return this.lastEpoch;\n  }\n}\n\n/**\n * Step learning rate scheduler.\n *\n * Decays the learning rate by gamma every stepSize epochs.\n * lr = baseLr * gamma^(epoch // stepSize)\n *\n * @example\n * ```ts\n * const scheduler = new StepLR(optimizer, { stepSize: 30, gamma: 0.1 });\n * // lr = 0.1 for epochs 0-29\n * // lr = 0.01 for epochs 30-59\n * // lr = 0.001 for epochs 60-89\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.StepLR.html | PyTorch StepLR}\n */\nexport class StepLR extends LRScheduler {\n  private stepSize: number;\n  private gamma: number;\n\n  constructor(\n    optimizer: SchedulerOptimizer,\n    options: { stepSize: number; gamma?: number; lastEpoch?: number }\n  ) {\n    const stepSize = validatePositiveInteger(options.stepSize, \"stepSize\");\n    const gamma = validatePositiveNumber(options.gamma ?? 0.1, \"gamma\");\n    const lastEpoch = validateLastEpoch(options.lastEpoch ?? -1);\n    super(optimizer, -1);\n    this.stepSize = stepSize;\n    this.gamma = gamma;\n    this.initializeFromLastEpoch(lastEpoch);\n  }\n\n  getLr(): number[] {\n    const factor = this.gamma ** Math.floor(this.lastEpoch / this.stepSize);\n    return this.baseLrs.map((lr) => lr * factor);\n  }\n}\n\n/**\n * Exponential learning rate scheduler.\n *\n * Decays the learning rate exponentially every epoch.\n * lr = baseLr * gamma^epoch\n *\n * @example\n * ```ts\n * const scheduler = new ExponentialLR(optimizer, { gamma: 0.95 });\n * // lr *= 0.95 each epoch\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.ExponentialLR.html | PyTorch ExponentialLR}\n */\nexport class ExponentialLR extends LRScheduler {\n  private gamma: number;\n\n  constructor(optimizer: SchedulerOptimizer, options: { gamma: number; lastEpoch?: number }) {\n    const gamma = validatePositiveNumber(options.gamma, \"gamma\");\n    const lastEpoch = validateLastEpoch(options.lastEpoch ?? -1);\n    super(optimizer, -1);\n    this.gamma = gamma;\n    this.initializeFromLastEpoch(lastEpoch);\n  }\n\n  getLr(): number[] {\n    return this.baseLrs.map((lr) => lr * this.gamma ** this.lastEpoch);\n  }\n}\n\n/**\n * Cosine annealing learning rate scheduler.\n *\n * Sets the learning rate using a cosine annealing schedule.\n * lr = etaMin + (baseLr - etaMin) * (1 + cos(π * epoch / T_max)) / 2\n *\n * @example\n * ```ts\n * const scheduler = new CosineAnnealingLR(optimizer, { T_max: 100, etaMin: 0.001 });\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.CosineAnnealingLR.html | PyTorch CosineAnnealingLR}\n */\nexport class CosineAnnealingLR extends LRScheduler {\n  private T_max: number;\n  private etaMin: number;\n\n  constructor(\n    optimizer: SchedulerOptimizer,\n    options: { T_max: number; etaMin?: number; lastEpoch?: number }\n  ) {\n    const tMax = validatePositiveInteger(options.T_max, \"T_max\");\n    const etaMin = validateNonNegativeNumber(options.etaMin ?? 0, \"etaMin\");\n    const lastEpoch = validateLastEpoch(options.lastEpoch ?? -1);\n    super(optimizer, -1);\n    this.T_max = tMax;\n    this.etaMin = etaMin;\n    this.initializeFromLastEpoch(lastEpoch);\n  }\n\n  getLr(): number[] {\n    return this.baseLrs.map((baseLr) => {\n      return (\n        this.etaMin +\n        ((baseLr - this.etaMin) * (1 + Math.cos((Math.PI * this.lastEpoch) / this.T_max))) / 2\n      );\n    });\n  }\n}\n\n/**\n * Multi-step learning rate scheduler.\n *\n * Decays the learning rate by gamma once the epoch reaches one of the milestones.\n *\n * @example\n * ```ts\n * const scheduler = new MultiStepLR(optimizer, { milestones: [30, 80], gamma: 0.1 });\n * // lr = 0.1 for epochs 0-29\n * // lr = 0.01 for epochs 30-79\n * // lr = 0.001 for epochs 80+\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.MultiStepLR.html | PyTorch MultiStepLR}\n */\nexport class MultiStepLR extends LRScheduler {\n  private sortedMilestones: number[];\n  private gamma: number;\n\n  constructor(\n    optimizer: SchedulerOptimizer,\n    options: { milestones: number[]; gamma?: number; lastEpoch?: number }\n  ) {\n    const milestones = validateMilestones(options.milestones);\n    const gamma = validatePositiveNumber(options.gamma ?? 0.1, \"gamma\");\n    const lastEpoch = validateLastEpoch(options.lastEpoch ?? -1);\n    super(optimizer, -1);\n    this.sortedMilestones = milestones;\n    this.gamma = gamma;\n    this.initializeFromLastEpoch(lastEpoch);\n  }\n\n  getLr(): number[] {\n    // Count how many milestones we've passed\n    let numDecays = 0;\n    for (const milestone of this.sortedMilestones) {\n      if (this.lastEpoch >= milestone) {\n        numDecays++;\n      }\n    }\n    const factor = this.gamma ** numDecays;\n    return this.baseLrs.map((lr) => lr * factor);\n  }\n}\n\n/**\n * Linear learning rate scheduler.\n *\n * Linearly interpolates the learning rate multiplicative factor from startFactor\n * to endFactor over totalIters epochs. After totalIters, the factor remains at endFactor.\n *\n * lr = baseLr * (startFactor + (endFactor - startFactor) * epoch / totalIters)\n *\n * @example\n * ```ts\n * const scheduler = new LinearLR(optimizer, {\n *   startFactor: 0.1,\n *   endFactor: 0.01,\n *   totalIters: 100\n * });\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.LinearLR.html | PyTorch LinearLR}\n */\nexport class LinearLR extends LRScheduler {\n  private startFactor: number;\n  private endFactor: number;\n  private totalIters: number;\n\n  constructor(\n    optimizer: SchedulerOptimizer,\n    options: {\n      startFactor?: number;\n      endFactor?: number;\n      totalIters: number;\n      lastEpoch?: number;\n    }\n  ) {\n    const startFactor = validatePositiveNumber(options.startFactor ?? 1 / 3, \"startFactor\");\n    const endFactor = validatePositiveNumber(options.endFactor ?? 1.0, \"endFactor\");\n    const totalIters = validatePositiveInteger(options.totalIters, \"totalIters\");\n    const lastEpoch = validateLastEpoch(options.lastEpoch ?? -1);\n    super(optimizer, -1);\n    this.startFactor = startFactor;\n    this.endFactor = endFactor;\n    this.totalIters = totalIters;\n    this.initializeFromLastEpoch(lastEpoch);\n  }\n\n  getLr(): number[] {\n    if (this.lastEpoch >= this.totalIters) {\n      return this.baseLrs.map((lr) => lr * this.endFactor);\n    }\n\n    const factor =\n      this.startFactor + (this.endFactor - this.startFactor) * (this.lastEpoch / this.totalIters);\n    return this.baseLrs.map((lr) => lr * factor);\n  }\n}\n\n/**\n * Reduce learning rate on plateau.\n *\n * Reduces learning rate when a metric has stopped improving.\n * This scheduler reads a metric value and if no improvement is seen\n * for 'patience' epochs, the learning rate is reduced.\n *\n * @example\n * ```ts\n * const scheduler = new ReduceLROnPlateau(optimizer, {\n *   mode: 'min',\n *   factor: 0.1,\n *   patience: 10\n * });\n *\n * for (let epoch = 0; epoch < 100; epoch++) {\n *   const valLoss = validate();\n *   scheduler.step(valLoss);\n * }\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.ReduceLROnPlateau.html | PyTorch ReduceLROnPlateau}\n */\nexport class ReduceLROnPlateau {\n  private optimizer: SchedulerOptimizer;\n  private mode: \"min\" | \"max\";\n  private factor: number;\n  private patience: number;\n  private threshold: number;\n  private cooldown: number;\n  private minLr: number;\n  private best: number;\n  private numBadEpochs: number;\n  private cooldownCounter: number;\n\n  constructor(\n    optimizer: SchedulerOptimizer,\n    options: {\n      mode?: \"min\" | \"max\";\n      factor?: number;\n      patience?: number;\n      threshold?: number;\n      cooldown?: number;\n      minLr?: number;\n    } = {}\n  ) {\n    this.optimizer = optimizer;\n    validateOptimizer(optimizer);\n    this.mode = options.mode ?? \"min\";\n    if (this.mode !== \"min\" && this.mode !== \"max\") {\n      throw new InvalidParameterError(\"mode must be 'min' or 'max'\", \"mode\", options.mode);\n    }\n    this.factor = validateFiniteNumber(options.factor ?? 0.1, \"factor\");\n    if (this.factor <= 0 || this.factor >= 1) {\n      throw new InvalidParameterError(\n        \"factor must be in the interval (0, 1)\",\n        \"factor\",\n        this.factor\n      );\n    }\n    this.patience = validateNonNegativeInteger(options.patience ?? 10, \"patience\");\n    this.threshold = validateNonNegativeNumber(options.threshold ?? 1e-4, \"threshold\");\n    this.cooldown = validateNonNegativeInteger(options.cooldown ?? 0, \"cooldown\");\n    this.minLr = validateNonNegativeNumber(options.minLr ?? 0, \"minLr\");\n    this.best = this.mode === \"min\" ? Infinity : -Infinity;\n    this.numBadEpochs = 0;\n    this.cooldownCounter = 0;\n  }\n\n  /**\n   * Check if metric improved.\n   */\n  private isBetter(current: number): boolean {\n    if (this.mode === \"min\") {\n      return current < this.best - this.threshold;\n    }\n    return current > this.best + this.threshold;\n  }\n\n  /**\n   * Perform a scheduler step based on the metric value.\n   *\n   * @param metric - Current value of the metric being monitored\n   */\n  step(metric: number): void {\n    if (!Number.isFinite(metric)) {\n      throw new InvalidParameterError(\"metric must be finite\", \"metric\", metric);\n    }\n    if (this.cooldownCounter > 0) {\n      this.cooldownCounter--;\n      this.numBadEpochs = 0;\n    }\n\n    if (this.isBetter(metric)) {\n      this.best = metric;\n      this.numBadEpochs = 0;\n    } else if (this.cooldownCounter === 0) {\n      this.numBadEpochs++;\n    }\n\n    if (this.numBadEpochs > this.patience) {\n      this.reduceLr();\n      this.cooldownCounter = this.cooldown;\n      this.numBadEpochs = 0;\n    }\n  }\n\n  /**\n   * Reduce learning rate for all parameter groups.\n   */\n  private reduceLr(): void {\n    for (let i = 0; i < this.optimizer.paramGroups.length; i++) {\n      const group = this.optimizer.paramGroups[i];\n      if (!group) {\n        throw new InvalidParameterError(\n          `optimizer.paramGroups[${i}] is missing`,\n          \"optimizer.paramGroups\",\n          group\n        );\n      }\n      const currentLr = resolveGroupLr(group, i);\n      const newLr = Math.max(currentLr * this.factor, this.minLr);\n      setGroupLr(group, newLr);\n    }\n  }\n\n  /**\n   * Get the current learning rates for all parameter groups.\n   */\n  getLastLr(): number[] {\n    return this.optimizer.paramGroups.map((group, index) => resolveGroupLr(group, index));\n  }\n}\n\n/**\n * Warmup scheduler that wraps another scheduler.\n *\n * Linearly increases the learning rate from 0 to the base lr over warmupEpochs,\n * then delegates to the wrapped scheduler.\n *\n * @example\n * ```ts\n * const baseScheduler = new CosineAnnealingLR(optimizer, { T_max: 100 });\n * const scheduler = new WarmupLR(optimizer, baseScheduler, { warmupEpochs: 5 });\n * ```\n */\nexport class WarmupLR extends LRScheduler {\n  private warmupEpochs: number;\n  private afterScheduler: LRScheduler | null;\n\n  constructor(\n    optimizer: SchedulerOptimizer,\n    afterScheduler: LRScheduler | null,\n    options: { warmupEpochs: number; lastEpoch?: number }\n  ) {\n    const warmupEpochs = validatePositiveInteger(options.warmupEpochs, \"warmupEpochs\");\n    const lastEpoch = validateLastEpoch(options.lastEpoch ?? -1);\n    super(optimizer, -1);\n    this.warmupEpochs = warmupEpochs;\n    this.afterScheduler = afterScheduler;\n    this.initializeFromLastEpoch(lastEpoch);\n  }\n\n  getLr(): number[] {\n    if (this.lastEpoch < this.warmupEpochs) {\n      // Linear warmup\n      const factor = (this.lastEpoch + 1) / this.warmupEpochs;\n      return this.baseLrs.map((lr) => lr * factor);\n    }\n\n    if (this.afterScheduler) {\n      // Delegate to wrapped scheduler\n      return this.afterScheduler.getLr();\n    }\n\n    return this.baseLrs;\n  }\n\n  override step(): void {\n    super.step();\n\n    // Also step the after scheduler once warmup is complete\n    if (this.lastEpoch >= this.warmupEpochs && this.afterScheduler) {\n      this.afterScheduler.step();\n    }\n  }\n}\n\n/**\n * One-cycle learning rate scheduler.\n *\n * Implements the 1cycle policy: lr starts at maxLr/divFactor, increases to maxLr\n * over pctStart of the training, then decreases to maxLr/finalDivFactor.\n *\n * @example\n * ```ts\n * const scheduler = new OneCycleLR(optimizer, {\n *   maxLr: 0.1,\n *   totalSteps: 1000,\n *   pctStart: 0.3\n * });\n * ```\n *\n * @see {@link https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.OneCycleLR.html | PyTorch OneCycleLR}\n */\nexport class OneCycleLR extends LRScheduler {\n  private maxLr: number;\n  private totalSteps: number;\n  private pctStart: number;\n  private divFactor: number;\n  private finalDivFactor: number;\n  private annealStrategy: \"cos\" | \"linear\";\n\n  constructor(\n    optimizer: SchedulerOptimizer,\n    options: {\n      maxLr: number;\n      totalSteps: number;\n      pctStart?: number;\n      divFactor?: number;\n      finalDivFactor?: number;\n      annealStrategy?: \"cos\" | \"linear\";\n      lastEpoch?: number;\n    }\n  ) {\n    const maxLr = validatePositiveNumber(options.maxLr, \"maxLr\");\n    const totalSteps = validatePositiveInteger(options.totalSteps, \"totalSteps\");\n    const pctStart = validateFiniteNumber(options.pctStart ?? 0.3, \"pctStart\");\n    if (pctStart <= 0 || pctStart >= 1) {\n      throw new InvalidParameterError(\n        \"pctStart must be in the interval (0, 1)\",\n        \"pctStart\",\n        pctStart\n      );\n    }\n    const divFactor = validatePositiveNumber(options.divFactor ?? 25, \"divFactor\");\n    const finalDivFactor = validatePositiveNumber(options.finalDivFactor ?? 1e4, \"finalDivFactor\");\n    const annealStrategy = options.annealStrategy ?? \"cos\";\n    if (annealStrategy !== \"cos\" && annealStrategy !== \"linear\") {\n      throw new InvalidParameterError(\n        \"annealStrategy must be 'cos' or 'linear'\",\n        \"annealStrategy\",\n        annealStrategy\n      );\n    }\n    const lastEpoch = validateLastEpoch(options.lastEpoch ?? -1);\n    super(optimizer, -1);\n    this.maxLr = maxLr;\n    this.totalSteps = totalSteps;\n    this.pctStart = pctStart;\n    this.divFactor = divFactor;\n    this.finalDivFactor = finalDivFactor;\n    this.annealStrategy = annealStrategy;\n    this.initializeFromLastEpoch(lastEpoch);\n  }\n\n  getLr(): number[] {\n    const stepNum = this.lastEpoch;\n    const upSteps = Math.max(1, Math.floor(this.totalSteps * this.pctStart));\n    const downSteps = Math.max(1, this.totalSteps - upSteps);\n\n    const initialLr = this.maxLr / this.divFactor;\n    const minLr = this.maxLr / this.finalDivFactor;\n\n    let lr: number;\n\n    if (stepNum >= this.totalSteps) {\n      lr = minLr;\n    } else if (stepNum < upSteps) {\n      // Increasing phase\n      const pct = stepNum / upSteps;\n      lr = initialLr + (this.maxLr - initialLr) * pct;\n    } else {\n      // Decreasing phase\n      const pct = (stepNum - upSteps) / downSteps;\n      if (this.annealStrategy === \"cos\") {\n        lr = minLr + ((this.maxLr - minLr) * (1 + Math.cos(Math.PI * pct))) / 2;\n      } else {\n        lr = this.maxLr - (this.maxLr - minLr) * pct;\n      }\n    }\n\n    // Scale for each param group based on their base lr ratio\n    const baseRef = this.baseLrs[0] ?? 0;\n    return this.baseLrs.map((baseLr) => {\n      if (baseRef === 0) {\n        return baseLr === 0 ? 0 : lr;\n      }\n      return lr * (baseLr / baseRef);\n    });\n  }\n}\n"]}