@stevenvo780/st-lang 4.7.0 → 4.9.0

package/dist/reasoning/probabilistic/examples.js

@@ -0,0 +1,85 @@
+"use strict";
+// ============================================================
+// Probabilistic Programming — Programas de ejemplo
+// ============================================================
+//
+// Mini-biblioteca de programas canónicos para exámenes/demos:
+//   - `coinExample()`           Bernoulli(0.5) sin conditioning.
+//   - `biasedCoinExample(p)`    Bernoulli(p) sin conditioning.
+//   - `twoCoinsExample()`       AND de dos monedas (sanity check).
+//   - `bayesianLinearRegression(xs, ys)`
+//                              priors gaussianos sobre slope/intercept
+//                              + likelihood gaussiana del residuo.
+//   - `gaussianMeanModel(data)` posterior sobre mu | data,
+//                              prior Normal(0, 10), σ conocido = 1.
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.coinExample = coinExample;
+exports.biasedCoinExample = biasedCoinExample;
+exports.twoCoinsExample = twoCoinsExample;
+exports.bayesianLinearRegression = bayesianLinearRegression;
+exports.gaussianMeanModel = gaussianMeanModel;
+const distributions_1 = require("./distributions");
+const types_1 = require("./types");
+/** Bernoulli simétrica: prior trivial sobre {true, false}. */
+function coinExample() {
+    return (s) => s.sample((0, types_1.bernoulli)(0.5));
+}
+/** Bernoulli con bias arbitrario p ∈ [0,1]. */
+function biasedCoinExample(p) {
+    return (s) => s.sample((0, types_1.bernoulli)(p));
+}
+/** AND de dos monedas Bernoulli(0.5): P(true) = 0.25 esperado. */
+function twoCoinsExample() {
+    return (s) => {
+        const a = s.sample((0, types_1.bernoulli)(0.5));
+        const b = s.sample((0, types_1.bernoulli)(0.5));
+        return a && b;
+    };
+}
+/**
+ * Bayesian linear regression univariado:
+ *   slope ~ Normal(0, 5)
+ *   intercept ~ Normal(0, 5)
+ *   yᵢ | slope, intercept ~ Normal(slope·xᵢ + intercept, 1)
+ *
+ * Devuelve la pareja (slope, intercept). Las observaciones se
+ * codifican como factor(logPdf(...)) sobre el residuo — equivale
+ * a una likelihood gaussiana, suitable para importance/MH.
+ */
+function bayesianLinearRegression(xs, ys) {
+    if (xs.length !== ys.length) {
+        throw new Error('bayesianLinearRegression: xs e ys deben tener la misma longitud');
+    }
+    return (s) => {
+        const slope = s.sample((0, types_1.normal)(0, 5));
+        const intercept = s.sample((0, types_1.normal)(0, 5));
+        for (let i = 0; i < xs.length; i++) {
+            const x = xs[i];
+            const y = ys[i];
+            const predicted = slope * x + intercept;
+            // Likelihood: y ~ Normal(predicted, σ=1). Usamos factor con
+            // log-pdf para que el peso de la traza refleje la calidad
+            // del fit sin descartar.
+            s.factor((0, distributions_1.logPdf)((0, types_1.normal)(predicted, 1), y));
+        }
+        return { slope, intercept };
+    };
+}
+/**
+ * Posterior sobre la media de una gaussiana con σ conocido.
+ *   mu ~ Normal(0, 10)
+ *   xᵢ | mu ~ Normal(mu, 1)
+ *
+ * Útil como caso de test analítico: el posterior verdadero es
+ * Normal((Σxᵢ) / (n + 1/100), 1/√(n + 1/100)).
+ */
+function gaussianMeanModel(data) {
+    return (s) => {
+        const mu = s.sample((0, types_1.normal)(0, 10));
+        for (const x of data) {
+            s.factor((0, distributions_1.logPdf)((0, types_1.normal)(mu, 1), x));
+        }
+        return mu;
+    };
+}
+//# sourceMappingURL=examples.js.map

package/dist/reasoning/probabilistic/examples.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"examples.js","sourceRoot":"","sources":["../../../src/reasoning/probabilistic/examples.ts"],"names":[],"mappings":";AAAA,+DAA+D;AAC/D,mDAAmD;AACnD,+DAA+D;AAC/D,EAAE;AACF,8DAA8D;AAC9D,iEAAiE;AACjE,+DAA+D;AAC/D,mEAAmE;AACnE,yCAAyC;AACzC,uEAAuE;AACvE,mEAAmE;AACnE,2DAA2D;AAC3D,oEAAoE;;AAMpE,kCAEC;AAGD,8CAEC;AAGD,0CAMC;AAYD,4DAqBC;AAUD,8CAQC;AAvED,mDAAyC;AACzC,mCAA2D;AAE3D,8DAA8D;AAC9D,SAAgB,WAAW;IACzB,OAAO,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,IAAA,iBAAS,EAAC,GAAG,CAAC,CAAC,CAAC;AACzC,CAAC;AAED,+CAA+C;AAC/C,SAAgB,iBAAiB,CAAC,CAAS;IACzC,OAAO,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,IAAA,iBAAS,EAAC,CAAC,CAAC,CAAC,CAAC;AACvC,CAAC;AAED,kEAAkE;AAClE,SAAgB,eAAe;IAC7B,OAAO,CAAC,CAAC,EAAE,EAAE;QACX,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,IAAA,iBAAS,EAAC,GAAG,CAAC,CAAC,CAAC;QACnC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,IAAA,iBAAS,EAAC,GAAG,CAAC,CAAC,CAAC;QACnC,OAAO,CAAC,IAAI,CAAC,CAAC;IAChB,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;;;GASG;AACH,SAAgB,wBAAwB,CACtC,EAAY,EACZ,EAAY;IAEZ,IAAI,EAAE,CAAC,MAAM,KAAK,EAAE,CAAC,MAAM,EAAE,CAAC;QAC5B,MAAM,IAAI,KAAK,CAAC,iEAAiE,CAAC,CAAC;IACrF,CAAC;IACD,OAAO,CAAC,CAAC,EAAE,EAAE;QACX,MAAM,KAAK,GAAG,CAAC,CAAC,MAAM,CAAC,IAAA,cAAM,EAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,SAAS,GAAG,CAAC,CAAC,MAAM,CAAC,IAAA,cAAM,EAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QACzC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACnC,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;YAChB,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;YAChB,MAAM,SAAS,GAAG,KAAK,GAAG,CAAC,GAAG,SAAS,CAAC;YACxC,4DAA4D;YAC5D,0DAA0D;YAC1D,yBAAyB;YACzB,CAAC,CAAC,MAAM,CAAC,IAAA,sBAAM,EAAC,IAAA,cAAM,EAAC,SAAS,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QAC5C,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC;IAC9B,CAAC,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,SAAgB,iBAAiB,CAAC,IAAc;IAC9C,OAAO,CAAC,CAAC,EAAE,EAAE;QACX,MAAM,EAAE,GAAG,CAAC,CAAC,MAAM,CAAC,IAAA,cAAM,EAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;QACnC,KAAK,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;YACrB,CAAC,CAAC,MAAM,CAAC,IAAA,sBAAM,EAAC,IAAA,cAAM,EAAC,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QACrC,CAAC;QACD,OAAO,EAAE,CAAC;IACZ,CAAC,CAAC;AACJ,CAAC"}

package/dist/reasoning/probabilistic/index.d.ts

@@ -0,0 +1,6 @@
+export { enumerate, importanceSample, metropolisHastings, rejectionSample } from './inference';
+export { enumerateSupport, logPdf, sampleFrom } from './distributions';
+export { bayesianLinearRegression, biasedCoinExample, coinExample, gaussianMeanModel, twoCoinsExample, } from './examples';
+export { bernoulli, categorical, discrete, normal, poisson, uniform } from './types';
+export type { BernoulliDist, CategoricalDist, DiscreteDist, Distribution, InferenceOptions, NormalDist, PProgram, PoissonDist, PosteriorSummary, Sampler, UniformDist, } from './types';
+//# sourceMappingURL=index.d.ts.map

package/dist/reasoning/probabilistic/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/reasoning/probabilistic/index.ts"],"names":[],"mappings":"AAiBA,OAAO,EAAE,SAAS,EAAE,gBAAgB,EAAE,kBAAkB,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAE/F,OAAO,EAAE,gBAAgB,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAEvE,OAAO,EACL,wBAAwB,EACxB,iBAAiB,EACjB,WAAW,EACX,iBAAiB,EACjB,eAAe,GAChB,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAErF,YAAY,EACV,aAAa,EACb,eAAe,EACf,YAAY,EACZ,YAAY,EACZ,gBAAgB,EAChB,UAAU,EACV,QAAQ,EACR,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,WAAW,GACZ,MAAM,SAAS,CAAC"}

package/dist/reasoning/probabilistic/index.js

@@ -0,0 +1,42 @@
+"use strict";
+// ============================================================
+// ST Probabilistic Programming — Barrel
+// ============================================================
+//
+// PPL minimal: define un `PProgram<T>` como `(sampler) => T`, donde
+// `sampler` ofrece `sample(dist)`, `observe(condition)` y
+// `factor(logWeight)`. Cuatro motores de inferencia:
+//
+//   - `enumerate`           Exacto, soporta solo discreto finito.
+//   - `rejectionSample`     Muestreo del prior + rechazo por observe.
+//   - `importanceSample`    Muestreo del prior + log-weights.
+//   - `metropolisHastings`  Single-site MCMC (proposal = prior).
+//
+// Más distribuciones (`sampleFrom`, `logPdf`, `enumerateSupport`) y
+// programas de ejemplo (`coinExample`, `bayesianLinearRegression`,
+// `gaussianMeanModel`).
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.uniform = exports.poisson = exports.normal = exports.discrete = exports.categorical = exports.bernoulli = exports.twoCoinsExample = exports.gaussianMeanModel = exports.coinExample = exports.biasedCoinExample = exports.bayesianLinearRegression = exports.sampleFrom = exports.logPdf = exports.enumerateSupport = exports.rejectionSample = exports.metropolisHastings = exports.importanceSample = exports.enumerate = void 0;
+var inference_1 = require("./inference");
+Object.defineProperty(exports, "enumerate", { enumerable: true, get: function () { return inference_1.enumerate; } });
+Object.defineProperty(exports, "importanceSample", { enumerable: true, get: function () { return inference_1.importanceSample; } });
+Object.defineProperty(exports, "metropolisHastings", { enumerable: true, get: function () { return inference_1.metropolisHastings; } });
+Object.defineProperty(exports, "rejectionSample", { enumerable: true, get: function () { return inference_1.rejectionSample; } });
+var distributions_1 = require("./distributions");
+Object.defineProperty(exports, "enumerateSupport", { enumerable: true, get: function () { return distributions_1.enumerateSupport; } });
+Object.defineProperty(exports, "logPdf", { enumerable: true, get: function () { return distributions_1.logPdf; } });
+Object.defineProperty(exports, "sampleFrom", { enumerable: true, get: function () { return distributions_1.sampleFrom; } });
+var examples_1 = require("./examples");
+Object.defineProperty(exports, "bayesianLinearRegression", { enumerable: true, get: function () { return examples_1.bayesianLinearRegression; } });
+Object.defineProperty(exports, "biasedCoinExample", { enumerable: true, get: function () { return examples_1.biasedCoinExample; } });
+Object.defineProperty(exports, "coinExample", { enumerable: true, get: function () { return examples_1.coinExample; } });
+Object.defineProperty(exports, "gaussianMeanModel", { enumerable: true, get: function () { return examples_1.gaussianMeanModel; } });
+Object.defineProperty(exports, "twoCoinsExample", { enumerable: true, get: function () { return examples_1.twoCoinsExample; } });
+var types_1 = require("./types");
+Object.defineProperty(exports, "bernoulli", { enumerable: true, get: function () { return types_1.bernoulli; } });
+Object.defineProperty(exports, "categorical", { enumerable: true, get: function () { return types_1.categorical; } });
+Object.defineProperty(exports, "discrete", { enumerable: true, get: function () { return types_1.discrete; } });
+Object.defineProperty(exports, "normal", { enumerable: true, get: function () { return types_1.normal; } });
+Object.defineProperty(exports, "poisson", { enumerable: true, get: function () { return types_1.poisson; } });
+Object.defineProperty(exports, "uniform", { enumerable: true, get: function () { return types_1.uniform; } });
+//# sourceMappingURL=index.js.map

package/dist/reasoning/probabilistic/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/reasoning/probabilistic/index.ts"],"names":[],"mappings":";AAAA,+DAA+D;AAC/D,wCAAwC;AACxC,+DAA+D;AAC/D,EAAE;AACF,oEAAoE;AACpE,0DAA0D;AAC1D,qDAAqD;AACrD,EAAE;AACF,kEAAkE;AAClE,sEAAsE;AACtE,8DAA8D;AAC9D,iEAAiE;AACjE,EAAE;AACF,oEAAoE;AACpE,mEAAmE;AACnE,wBAAwB;;;AAExB,yCAA+F;AAAtF,sGAAA,SAAS,OAAA;AAAE,6GAAA,gBAAgB,OAAA;AAAE,+GAAA,kBAAkB,OAAA;AAAE,4GAAA,eAAe,OAAA;AAEzE,iDAAuE;AAA9D,iHAAA,gBAAgB,OAAA;AAAE,uGAAA,MAAM,OAAA;AAAE,2GAAA,UAAU,OAAA;AAE7C,uCAMoB;AALlB,oHAAA,wBAAwB,OAAA;AACxB,6GAAA,iBAAiB,OAAA;AACjB,uGAAA,WAAW,OAAA;AACX,6GAAA,iBAAiB,OAAA;AACjB,2GAAA,eAAe,OAAA;AAGjB,iCAAqF;AAA5E,kGAAA,SAAS,OAAA;AAAE,oGAAA,WAAW,OAAA;AAAE,iGAAA,QAAQ,OAAA;AAAE,+FAAA,MAAM,OAAA;AAAE,gGAAA,OAAO,OAAA;AAAE,gGAAA,OAAO,OAAA"}

package/dist/reasoning/probabilistic/inference.d.ts

@@ -0,0 +1,45 @@
+import type { InferenceOptions, PProgram, PosteriorSummary } from './types';
+/**
+ * Enumera todas las trayectorias del programa, multiplicando la
+ * probabilidad de cada decisión discreta y descartando ramas con
+ * `observe(false)`. Soporta `factor(logW)` agregando exp(logW) al
+ * peso. No soporta distribuciones continuas (uniform, normal) ni
+ * `poisson` — lanza error si las encuentra.
+ *
+ * `maxStates` protege de explosión combinatoria: corta la
+ * enumeración cuando el número de ramas pendientes la supera.
+ */
+export declare function enumerate<T>(program: PProgram<T>, maxStates?: number): PosteriorSummary<T>;
+/**
+ * Muestrea N trazas del prior; cualquier traza con `observe(false)`
+ * o `factor(-Infinity)` se descarta y se reintenta.
+ *
+ * `factor(logW)` con `logW < 0` se trata como aceptación
+ * probabilística: con prob `exp(logW)` la traza se acepta, en
+ * caso contrario se descarta. `logW > 0` es legal pero raro;
+ * se acepta siempre (rejection no maneja pesos > 1).
+ */
+export declare function rejectionSample<T>(program: PProgram<T>, opts?: InferenceOptions): PosteriorSummary<T>;
+/**
+ * Muestrea trazas del prior y acumula log-pesos de `observe`
+ * (treated as -Infinity si false, 0 si true) y `factor`. Devuelve
+ * los samples + sus pesos normalizados; el histograma y mean/std
+ * reflejan estos pesos.
+ *
+ * Reporta ESS = (Σwᵢ)² / Σwᵢ² como diagnóstico de degeneración.
+ */
+export declare function importanceSample<T>(program: PProgram<T>, opts?: InferenceOptions): PosteriorSummary<T>;
+/**
+ * MCMC Metropolis-Hastings sobre el espacio de trazas del programa.
+ *
+ * Convenciones:
+ *   - `burnIn` itera sin contar (default 1000).
+ *   - `thin` toma 1 de cada N (default 1).
+ *   - Single-site: en cada iteración, re-muestrea UN sample site
+ *     (elegido uniforme) y re-ejecuta. Si el número de sample sites
+ *     cambia entre trazas, igual funciona (re-ejecutamos forzando
+ *     el valor anterior en los sites que coinciden).
+ *   - Reporta `acceptanceRate` para diagnóstico.
+ */
+export declare function metropolisHastings<T>(program: PProgram<T>, opts?: InferenceOptions): PosteriorSummary<T>;
+//# sourceMappingURL=inference.d.ts.map

package/dist/reasoning/probabilistic/inference.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"inference.d.ts","sourceRoot":"","sources":["../../../src/reasoning/probabilistic/inference.ts"],"names":[],"mappings":"AA0BA,OAAO,KAAK,EAAgB,gBAAgB,EAAE,QAAQ,EAAE,gBAAgB,EAAW,MAAM,SAAS,CAAC;AA8HnG;;;;;;;;;GASG;AACH,wBAAgB,SAAS,CAAC,CAAC,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,EAAE,SAAS,SAAU,GAAG,gBAAgB,CAAC,CAAC,CAAC,CAiD3F;AAsED;;;;;;;;GAQG;AACH,wBAAgB,eAAe,CAAC,CAAC,EAC/B,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,EACpB,IAAI,GAAE,gBAAqB,GAC1B,gBAAgB,CAAC,CAAC,CAAC,CAoBrB;AAkCD;;;;;;;GAOG;AACH,wBAAgB,gBAAgB,CAAC,CAAC,EAChC,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,EACpB,IAAI,GAAE,gBAAqB,GAC1B,gBAAgB,CAAC,CAAC,CAAC,CAkCrB;AA+CD;;;;;;;;;;;GAWG;AACH,wBAAgB,kBAAkB,CAAC,CAAC,EAClC,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,EACpB,IAAI,GAAE,gBAAqB,GAC1B,gBAAgB,CAAC,CAAC,CAAC,CAqErB"}