@numero-uno-jpn/rstats 0.5.0 → 0.6.1

package/dist/numerics.js

@@ -0,0 +1,2 @@
+/*! rStats.js 0.6.1 | MIT License | includes code ported from SciPy, Boost.Math, statsmodels, LAPACK and others: see THIRD_PARTY_NOTICES.md at https://github.com/numero-uno-jpn/rStats.js */
+import{a as H,b as T,c as D}from"./chunks/chunk-TPC6MTYC.js";import"./chunks/chunk-DRNMIEYH.js";import{a as M,f as j}from"./chunks/chunk-NOCEAM44.js";import{b as S}from"./chunks/chunk-7NK4SA6B.js";import"./chunks/chunk-UGUQYWRQ.js";function x(o,l,t="mean"){if(o.length!==l.length)throw new RangeError("'x' and 'y' lengths differ");let n=[];for(let r=0;r<o.length;r++)!Number.isNaN(o[r])&&!Number.isNaN(l[r])&&n.push(r);n.sort((r,s)=>o[r]-o[s]);let e=typeof t=="function"?t:t==="min"?r=>Math.min(...r):t==="max"?r=>Math.max(...r):S,a=[],f=[];for(let r=0;r<n.length;){let s=o[n[r]],c=r+1;for(;c<n.length&&o[n[c]]===s;)c++;a.push(s),f.push(c-r===1?l[n[r]]:e(n.slice(r,c).map(i=>l[i]))),r=c}return{x:a,y:f}}function F(o,l,t,n){let e=(n-o[t])/(o[t+1]-o[t]);return l[t]+(l[t+1]-l[t])*e}function O(o,l,t){let n=(e,a)=>e===0?0:e*a;return n(1-t,o[l])+n(t,o[l+1])}function X(o,l,t={}){let{method:n="linear",n:e=50,f:a=0,rule:f=1,ties:r="mean"}=t,s=x(o,l,r),c=s.x.length;if(c<=1&&n==="linear")throw new RangeError("need at least two non-NA values to interpolate");if(c===0)throw new RangeError("zero non-NA points");let[i,p]=typeof f=="number"?[f,f]:f,d=t.yleft??(i===1?NaN:s.y[0]),h=t.yright??(p===1?NaN:s.y[c-1]),m;if(t.xout!==void 0)m=Array.from(t.xout);else{if(e<=0)throw new RangeError("'approx' requires n >= 1");m=M(s.x[0],s.x[c-1],e)}let g=s.x,u=s.y,w=j(m,g),y=m.map((b,A)=>{let R=w[A];if(Number.isNaN(R))return NaN;if(R===0)return d;let E=R-1;return b===g[E]?u[E]:R===c?h:n==="linear"?F(g,u,E,b):O(u,E,a)});return{x:m,y}}function v(o,l){let t=o.length,n=new Array(t).fill(0),e=new Array(t).fill(0),a=new Array(t).fill(0);if(t<2)return{b:n,c:e,d:a};if(t<3)return n[0]=(l[1]-l[0])/(o[1]-o[0]),n[1]=n[0],{b:n,c:e,d:a};let f=t-1;a[0]=o[1]-o[0],e[1]=(l[1]-l[0])/a[0];for(let r=1;r<f;r++)a[r]=o[r+1]-o[r],n[r]=2*(a[r-1]+a[r]),e[r+1]=(l[r+1]-l[r])/a[r],e[r]=e[r+1]-e[r];n[0]=-a[0],n[f]=-a[t-2],e[0]=0,e[f]=0,t>3&&(e[0]=e[2]/(o[3]-o[1])-e[1]/(o[2]-o[0]),e[f]=e[t-2]/(o[f]-o[t-3])-e[t-3]/(o[t-2]-o[t-4]),e[0]=e[0]*a[0]*a[0]/(o[3]-o[0]),e[f]=-e[f]*a[t-2]*a[t-2]/(o[f]-o[t-4]));for(let r=1;r<t;r++){let s=a[r-1]/n[r-1];n[r]=n[r]-s*a[r-1],e[r]=e[r]-s*e[r-1]}e[f]=e[f]/n[f];for(let r=t-2;r>=0;r--)e[r]=(e[r]-a[r]*e[r+1])/n[r];n[f]=(l[f]-l[t-2])/a[t-2]+a[t-2]*(e[t-2]+2*e[f]);for(let r=0;r<f;r++)n[r]=(l[r+1]-l[r])/a[r]-a[r]*(e[r+1]+2*e[r]),a[r]=(e[r+1]-e[r])/a[r],e[r]=3*e[r];return e[f]=3*e[f],a[f]=a[t-2],{b:n,c:e,d:a}}function z(o,l,t){let n=o.length,e=new Array(n).fill(0),a=new Array(n).fill(0),f=new Array(n).fill(0);for(let r=0;r<n-1;r++){let s=o[r+1]-o[r];e[r]=(l[r+1]-l[r])/s-s*(t[r+1]+2*t[r]),a[r]=3*t[r],f[r]=(t[r+1]-t[r])/s}return{b:e,c:a,d:f}}function J(o,l){let t=o.length;if(t<3)return v(o,l);let n=Array.from({length:t-1},(c,i)=>o[i+1]-o[i]),e=n.map((c,i)=>(l[i+1]-l[i])/c),a=new Array(t).fill(0),f=new Array(t).fill(0);for(let c=1;c<t-1;c++)a[c]=2*(n[c-1]+n[c]),f[c]=e[c]-e[c-1];for(let c=2;c<t-1;c++){let i=n[c-1]/a[c-1];a[c]=a[c]-i*n[c-1],f[c]=f[c]-i*f[c-1]}let r=new Array(t).fill(0);r[t-2]=f[t-2]/a[t-2];for(let c=t-3;c>=1;c--)r[c]=(f[c]-n[c]*r[c+1])/a[c];let s=z(o,l,r);return s.b[t-1]=e[t-2]+n[t-2]*r[t-2],s}function P(o,l){let t=o.length;if(l[0]!==l[t-1])throw new RangeError("periodic spline needs y[1] == y[n]");if(t<3)return{b:new Array(t).fill(0),c:new Array(t).fill(0),d:new Array(t).fill(0)};let n=t-1,e=Array.from({length:n},(u,w)=>o[w+1]-o[w]),a=e.map((u,w)=>(l[w+1]-l[w])/u),f=u=>u===0?n-1:u-1,r=Array.from({length:n},(u,w)=>2*(e[f(w)]+e[w])),s=Array.from({length:n},(u,w)=>a[w]-a[f(w)]),c=e.slice(),i=new Array(n).fill(0);i[0]=e[n-1],i[n-2]=i[n-2]+e[n-2];let p=i[0],d=r[n-1],h=s[n-1];for(let u=0;u<n-1;u++){if(u+1<n-1){let y=e[u]/r[u];r[u+1]=r[u+1]-y*c[u],i[u+1]=i[u+1]-y*i[u],s[u+1]=s[u+1]-y*s[u]}let w=p/r[u];d-=w*i[u],h-=w*s[u],u+1<n-1&&(p=(u+1===n-2?e[n-2]:0)-w*c[u])}let m=new Array(t).fill(0);m[n-1]=h/d;for(let u=n-2;u>=0;u--){let w=s[u]-i[u]*m[n-1]-(u<n-2?c[u]*m[u+1]:0);m[u]=w/r[u]}m[t-1]=m[0];let g=z(o,l,m);return g.b[t-1]=g.b[0],g.c[t-1]=g.c[0],g.d[t-1]=g.d[0],g}function Y(o,l){let t=o.length,{b:n,c:e,d:a}=v(o,l);if(t<2)return{b:n,c:e,d:a};let f=Array.from({length:t-1},(s,c)=>(l[c+1]-l[c])/(o[c+1]-o[c]));for(let s=0;s<t;s++){let c=f[s===0?0:s-1],i=f[s===t-1?t-2:s],p=Math.min(Math.abs(c),Math.abs(i)),d=c*i>0?i>0:n[s]>=0;n[s]=d?Math.min(Math.max(0,n[s]),3*p):Math.max(Math.min(0,n[s]),-3*p)}for(let s=0;s<t-1;s++){let c=o[s+1]-o[s],i=(l[s+1]-l[s])/c;e[s]=(3*i-2*n[s]-n[s+1])/c,a[s]=(n[s]+n[s+1]-2*i)/(c*c)}let r=o[t-1]-o[t-2];return e[t-1]=(n[t-2]+2*n[t-1]-3*(l[t-1]-l[t-2])/r)/r,a[t-1]=a[t-2],{b:n,c:e,d:a}}function q(o,l,t,n){let e=x(o,l,n),a=e.x.length;if(a===0)throw new RangeError("zero non-NA points");t==="periodic"&&e.y[a-1]!==e.y[0]&&(e.y[a-1]=e.y[0]),t==="hyman"&&B(e.y);let f=t==="fmm"?v(e.x,e.y):t==="natural"?J(e.x,e.y):t==="periodic"?P(e.x,e.y):t==="hyman"?Y(e.x,e.y):(()=>{throw new RangeError(`invalid spline method '${String(t)}'`)})();return{method:t,...e,...f}}function B(o){let l=!0,t=!0;for(let n=1;n<o.length;n++){let e=o[n]-o[n-1];e>=0||(l=!1),e<=0||(t=!1)}if(!l&&!t)throw new RangeError("'y' must be increasing or decreasing")}function C(o){let l=Math.trunc(o);if(!(l>=0&&l<=3))throw new RangeError("'deriv' must be between 0 and 3");return l}function I(o,l,t){let n=C(t),{x:e,y:a,b:f,c:r,d:s,method:c}=o,i=e.length,p=c==="natural",d=e[i-1]-e[0],h=0;return Array.from(l,m=>{let g=m;if(c==="periodic"&&i>1){let y=(g-e[0])%d;y<0&&(y+=d),g=e[0]+y}if(p&&n>0&&g<=e[0])return n===1?f[0]:0;if(!(e[h]<=g&&(h===i-1||g<=e[h+1]))){let y=0,b=i;for(;b-y>1;){let A=y+b>>1;e[A]<=g?y=A:b=A}h=y}let u=g-e[h],w=p&&g<e[0]?0:s[h];switch(n){case 0:return a[h]+u*(f[h]+u*(r[h]+u*w));case 1:return f[h]+u*(2*r[h]+u*(3*w));case 2:return 2*r[h]+u*(6*w);default:return 6*w}})}function G(o,l,t={}){let{method:n="fmm",ties:e="mean"}=t,a=q(o,l,n,e),f;if(t.xout!==void 0)f=Array.from(t.xout);else{let r=a.x.length,s=t.n??3*r;if(s<=0)throw new RangeError("'spline' requires n >= 1");f=M(t.xmin??a.x[0],t.xmax??a.x[r-1],s)}return{x:f,y:I(a,f,0)}}function K(o,l,t={}){let{method:n="fmm",ties:e="mean"}=t;if(n==="monoH.FC"){let f=x(o,l,e);if(f.x.length===0)throw new RangeError("zero non-NA points");let r=Q(f.x,f.y);return(s,c=0)=>L(f.x,f.y,r,s,c)}let a=q(o,l,n,e);return(f,r=0)=>I(a,f,r)}function Q(o,l){let t=o.length;if(t<2)throw new RangeError("length(m) must be at least two");let n=Array.from({length:t-1},(a,f)=>(l[f+1]-l[f])/(o[f+1]-o[f])),e=Array.from({length:t},(a,f)=>f===0?n[0]:f===t-1?n[t-2]:(n[f-1]+n[f])/2);for(let a=0;a<t-1;a++){let f=n[a];if(f===0){e[a]=0,e[a+1]=0;continue}let r=e[a]/f,s=e[a+1]/f,c=2*r+s-3,i=r+2*s-3;if(c>0&&i>0&&r*(c+i)<c*c){let p=3*f/Math.sqrt(r*r+s*s);e[a]=p*r,e[a+1]=p*s}}return e}function L(o,l,t,n,e){let a=C(e),f=o.length;return Array.from(n,r=>{if(Number.isNaN(r))return NaN;let s=0,c=f;for(;s<c;){let A=s+c>>1;o[A]<=r?s=A+1:c=A}let i=s;if(i===0||i===f){let A=i===0?0:f-1;return a===0?l[A]+t[A]*(r-o[A]):a===1?t[A]:0}let p=o[i-1],d=o[i]-p,h=(r-p)/d,m=h-1,g=l[i-1],u=l[i],w=t[i-1],y=t[i],b=(u-g)/d;switch(a){case 0:{let A=h*h*(3-2*h);return g*(1-A)+d*w*(h*m*m)+u*A+d*y*(h*h*m)}case 1:return b*(-6*h*m)+w*(m*(3*h-1))+y*(h*(3*h-2));case 2:return(b*(6-12*h)+w*(6*h-4)+y*(6*h-2))/d;default:return(-12*b+6*w+6*y)/(d*d)}})}function U(o,l,t){let n=o.length;if(l.length!==n||t.length!==n)throw new RangeError("'x', 'y' and 'm' must have the same length");let e=Array.from({length:n},(s,c)=>c).sort((s,c)=>o[s]-o[c]),a=e.map(s=>o[s]),f=e.map(s=>l[s]),r=e.map(s=>t[s]);for(let s=1;s<n;s++)if(!(a[s]-a[s-1]>0))throw new RangeError("'x' must be *strictly* increasing (non - NA)");return(s,c=0)=>L(a,f,r,s,c)}function V(o,l,t){let n=Array.from(o),e=Array.from(l),a=n.length;if(e.length!==a)throw new RangeError("'xi' and 'yi' must have the same length");let f=Array.from(t),r=Math.min(...n),s=Math.max(...n);if(f.some(i=>i<r||i>s))throw new RangeError("Some interpolation points outside the nodes.");let c=n.map((i,p)=>{let d=1;for(let h=0;h<a;h++)h!==p&&(d*=i-n[h]);return 1/d});return f.map(i=>{let p=0,d=0,h=n.lastIndexOf(i);if(h>=0)return e[h];for(let m=0;m<a;m++){let g=i-n[m],u=c[m]/g;p+=u*e[m],d+=u}return p/d})}function W(o,l,t,n){let e=Array.from(o),a=Array.from(l),f=Array.from(t),r=e.length;if(a.length!==r||f.length!==r)throw new RangeError("'xi', 'yi' and 'di' must have the same length");let s=e.flatMap(h=>[h,h]),c=2*r,i=a.flatMap(h=>[h,h]),p=[i[0]],d=i.slice();for(let h=1;h<c;h++){let m=[];for(let g=0;g<c-h;g++)h===1&&s[g+1]===s[g]?m.push(f[g>>1]):m.push((d[g+1]-d[g])/(s[g+h]-s[g]));p.push(m[0]),d=m}return Array.from(n,h=>{let m=p[c-1];for(let g=c-2;g>=0;g--)m=m*(h-s[g])+p[g];return m})}function Z(o,l,t,n=[],e={}){let a=e.method??"rk4";if(a!=="euler"&&a!=="rk2"&&a!=="rk4")throw new RangeError(`unknown method '${String(a)}'`);let f=[...n],r=o.length,s=(d,h)=>{let m=Array.from(t(d,h,f));if(m.length!==r)throw new RangeError(`The number of derivatives returned by func() (${m.length}) must equal the length of the initial conditions vector (${r})`);return m},c=(d,h,m)=>d.map((g,u)=>g+h*m[u]),i=[...o],p=[];if(l.length===0)return p;p.push([l[0],...i]);for(let d=1;d<l.length;d++){let h=l[d-1],m=l[d]-h,g=s(h,i);if(a==="euler")i=c(i,m,g);else if(a==="rk2"){let u=s(h+m,c(i,m,g));i=i.map((w,y)=>w+m*(.5*g[y]+.5*u[y]))}else{let u=s(h+m/2,c(i,m/2,g)),w=s(h+m/2,c(i,m/2,u)),y=s(h+m,c(i,m,w));i=i.map((b,A)=>b+m/6*(g[A]+2*u[A]+2*w[A]+y[A]))}p.push([l[d],...i])}return p}function _(o,l,t){let n=l.eps??1e-4,e=l.d??t,a=l.zeroTol??Math.sqrt(Number.EPSILON/7e-7),f=l.r??4;if(!(Number.isInteger(f)&&f>=1))throw new RangeError("'r' must be a positive integer");return{eps:n,r:f,h0:o.map(r=>Math.abs(e*r)+(Math.abs(r)<a?n:0))}}function k(o){let l=[...o];for(let t=1;t<l.length;t++){let n=4**t;for(let e=0;e<l.length-t;e++)l[e]=(l[e+1]*n-l[e])/(n-1)}return l[0]}function N(o,l,t){let n=[...o];return n[l]=n[l]+t,n}function $(o){let l=o.method??"Richardson";if(l!=="Richardson"&&l!=="simple")throw new RangeError(`unknown method "${String(l)}" (expected "Richardson" or "simple")`);return l}function tt(o,l,t={}){let n=[...l],e=n.length;if($(t)==="simple"){let s=t.eps??1e-4,c=o([...n]);return n.map((i,p)=>(o(N(n,p,s))-c)/s)}let{r:a,h0:f}=_(n,t,1e-4),r=[];for(let s=0;s<e;s++){let c=[],i=f[s];for(let p=0;p<a;p++){if(p>0&&Math.abs(c[p-1])<1e-20?c.push(0):c.push((o(N(n,s,i))-o(N(n,s,-i)))/(2*i)),Number.isNaN(c[p]))throw new RangeError(`func returned NaN at step ${i} from x`);i/=2}r.push(k(c))}return r}function nt(o,l,t={}){let n=[...l],e=n.length,a=Array.from(o([...n])),f=a.map(()=>new Array(e).fill(NaN));if($(t)==="simple"){let c=t.eps??1e-4;for(let i=0;i<e;i++){let p=o(N(n,i,c));a.forEach((d,h)=>f[h][i]=(p[h]-d)/c)}return f}let{r,h0:s}=_(n,t,1e-4);for(let c=0;c<e;c++){let i=a.map(()=>[]),p=s[c];for(let d=0;d<r;d++){let h=o(N(n,c,p)),m=o(N(n,c,-p));a.forEach((g,u)=>i[u].push((h[u]-m[u])/(2*p))),p/=2}a.forEach((d,h)=>f[h][c]=k(i[h]))}return f}function rt(o,l,t={}){if((t.method??"Richardson")!=="Richardson")throw new RangeError('hessian supports only method "Richardson"');let n=[...l],e=n.length,{r:a,h0:f}=_(n,t,.1),r=o([...n]),s=[];for(let i=0;i<e;i++){let p=[],d=f[i];for(let h=0;h<a;h++)p.push((o(N(n,i,d))-2*r+o(N(n,i,-d)))/d**2),d/=2;s.push(k(p))}let c=n.map(()=>new Array(e).fill(NaN));for(let i=0;i<e;i++){c[i][i]=s[i];for(let p=0;p<i;p++){let d=[],h=f[i],m=f[p];for(let g=0;g<a;g++){let u=N(N(n,i,h),p,m),w=N(N(n,i,-h),p,-m);d.push((o(u)-2*r+o(w)-s[i]*h**2-s[p]*m**2)/(2*h*m)),h/=2,m/=2}c[i][p]=c[p][i]=k(d)}}return c}export{X as approx,V as barylag,tt as grad,W as hermiteInterp,rt as hessian,H as integrate,nt as jacobian,Z as ode,T as optimize,G as spline,K as splinefun,U as splinefunH,D as uniroot};

package/dist/optim/localmin.d.ts

@@ -0,0 +1,16 @@
+/** Default tolerance .Machine$double.eps^0.25, as in R's optimize */
+export declare const LOCALMIN_TOL: number;
+export interface LocalMinResult {
+    /** Location of the minimum */
+    x: number;
+    /** f at x (the last value computed there) */
+    fx: number;
+    /** Number of evaluations of f */
+    evaluations: number;
+}
+/**
+ * A local minimum of f in [lower, upper], to within about 3 tol + 2 sqrt(eps) |x| (Brent's localmin).
+ *
+ * @param tol - Tolerance (default .Machine$double.eps^0.25).
+ */
+export declare function localMin(f: (x: number) => number, lower: number, upper: number, tol?: number): LocalMinResult;

package/dist/optim/localmin.js

@@ -0,0 +1,97 @@
+// One-dimensional minimisation without derivatives (localmin from Brent 1973, "Algorithms for Minimization without
+// Derivatives", chapter 5), the algorithm behind R's optimize(). Golden-section search combined with successive
+// parabolic interpolation. The procedure follows the ALGOL 60 procedure in Brent's book (the same as netlib's
+// fmin.f, public domain), written for rStats.js.
+/** Default tolerance .Machine$double.eps^0.25, as in R's optimize */
+export const LOCALMIN_TOL = Math.pow(Number.EPSILON, 0.25);
+/**
+ * A local minimum of f in [lower, upper], to within about 3 tol + 2 sqrt(eps) |x| (Brent's localmin).
+ *
+ * @param tol - Tolerance (default .Machine$double.eps^0.25).
+ */
+export function localMin(f, lower, upper, tol = LOCALMIN_TOL) {
+    const c = 0.5 * (3 - Math.sqrt(5));
+    const eps = Math.sqrt(Number.EPSILON);
+    let a = lower;
+    let b = upper;
+    let v = a + c * (b - a);
+    let w = v;
+    let x = v;
+    let d = 0;
+    let e = 0;
+    let fx = f(x);
+    let fv = fx;
+    let fw = fx;
+    let evaluations = 1;
+    const tol3 = tol / 3;
+    for (;;) {
+        const xm = 0.5 * (a + b);
+        const tol1 = eps * Math.abs(x) + tol3;
+        const t2 = 2 * tol1;
+        // stopping criterion
+        if (Math.abs(x - xm) <= t2 - 0.5 * (b - a))
+            break;
+        let golden = true;
+        if (Math.abs(e) > tol1) {
+            // fit a parabola through (v, fv), (w, fw), (x, fx)
+            let r = (x - w) * (fx - fv);
+            let q = (x - v) * (fx - fw);
+            let p = (x - v) * q - (x - w) * r;
+            q = 2 * (q - r);
+            if (q > 0)
+                p = -p;
+            else
+                q = -q;
+            r = e;
+            e = d;
+            if (Math.abs(p) < Math.abs(0.5 * q * r) && p > q * (a - x) && p < q * (b - x)) {
+                // parabolic interpolation step
+                d = p / q;
+                const u = x + d;
+                // f must not be evaluated too close to a or b
+                if (u - a < t2 || b - u < t2)
+                    d = x <= xm ? tol1 : -tol1; // sign(tol1, xm - x), + for 0
+                golden = false;
+            }
+        }
+        if (golden) {
+            // golden-section step
+            e = x < xm ? b - x : a - x;
+            d = c * e;
+        }
+        // f must not be evaluated too close to x
+        const u = Math.abs(d) >= tol1 ? x + d : d >= 0 ? x + tol1 : x - tol1; // x + sign(tol1, d)
+        const fu = f(u);
+        evaluations++;
+        // update a, b, v, w and x
+        if (fu <= fx) {
+            if (u < x)
+                b = x;
+            else
+                a = x;
+            v = w;
+            fv = fw;
+            w = x;
+            fw = fx;
+            x = u;
+            fx = fu;
+        }
+        else {
+            if (u < x)
+                a = u;
+            else
+                b = u;
+            if (fu <= fw || w === x) {
+                v = w;
+                fv = fw;
+                w = u;
+                fw = fu;
+            }
+            else if (fu <= fv || v === x || v === w) {
+                v = u;
+                fv = fu;
+            }
+        }
+    }
+    return { x, fx, evaluations };
+}

package/dist/random/beta.js

@@ -3,7 +3,7 @@
 // Cheng (1978) "Generating beta variates with nonintegral shape parameters"
 // (Comm. ACM 21(4)): algorithm BB (min(a, b) > 1) and BC (min(a, b) <= 1).
 // These are the algorithms cited in R's ?rbeta, with the constants from the paper. The order in which uniforms
-// are consumed and the parameter handling follow observed R output. R's source was not consulted.
+// are consumed and the parameter handling follow observed R output.
 import { unifRand } from "./rng.js";
 import { generate } from "./common.js";
 import { rchisq } from "./gamma.js";

package/dist/random/binomial.js

@@ -3,7 +3,7 @@
 // Kachitvichyanukul & Schmeiser (1988) "Binomial random variate generation"
 // (Comm. ACM 31(2)): algorithm BTPE. When n min(p, 1-p) < 30, inversion (BINV in the paper) is used.
 // These are the algorithms cited in R's ?rbinom, with the constants from the paper. The order in which uniforms
-// are consumed and the parameter handling follow observed R output. R's source was not consulted.
+// are consumed and the parameter handling follow observed R output.
 import { unifRand } from "./rng.js";
 import { generate, powInt } from "./common.js";
 import { qbinom } from "../distributions/binomial.js";

package/dist/random/continuous.js

@@ -1,7 +1,7 @@
 // Continuous random variates generated by inversion and similar: runif, rnorm, rexp, rlnorm, rlogis, rcauchy, rweibull.
 //
 // Parameter handling (whether a uniform is consumed for NaN, 0 or Inf) follows observed R output
-// (verified by next in test/fixtures/random*.json). R's source was not consulted.
+// (verified by next in test/fixtures/random*.json).
 import { expRand, normRand, unifRand } from "./rng.js";
 import { generate } from "./common.js";
 /**

package/dist/random/extra.d.ts

@@ -0,0 +1,33 @@
+/**
+ * Random numbers from the inverse gamma distribution. Equivalent to `extraDistr::rinvgamma(n, alpha, beta = 1)`.
+ *
+ * @param n Number of values to generate.
+ * @param alpha Shape.
+ * @param options.beta Scale. Default `1`.
+ */
+export declare function rinvgamma(n: number, alpha: number, options?: {
+    beta?: number;
+}): number[];
+/** Random numbers from the Kumaraswamy distribution. Equivalent to `extraDistr::rkumar(n, a, b)`. */
+export declare function rkumar(n: number, a: number, b: number): number[];
+/** Random numbers from the Pareto distribution. Equivalent to `extraDistr::rpareto(n, a = 1, b = 1)`. */
+export declare function rpareto(n: number, options?: {
+    a?: number;
+    b?: number;
+}): number[];
+/** Random numbers from the triangular distribution. Equivalent to `extraDistr::rtriang(n, a = -1, b = 1, c = (a + b)/2)`. */
+export declare function rtriang(n: number, options?: {
+    a?: number;
+    b?: number;
+    c?: number;
+}): number[];
+/** Random numbers from the Laplace distribution. Equivalent to `extraDistr::rlaplace(n, mu = 0, sigma = 1)`. */
+export declare function rlaplace(n: number, options?: {
+    mu?: number;
+    sigma?: number;
+}): number[];
+/** Random numbers from the arcsine distribution on (a, b), by inversion: `qarcsine(runif(n), a, b)`. */
+export declare function rarcsine(n: number, options?: {
+    a?: number;
+    b?: number;
+}): number[];

package/dist/random/extra.js

@@ -0,0 +1,62 @@
+// Random variates of the distributions outside base R: rinvgamma, rkumar, rpareto, rtriang, rlaplace (as extraDistr
+// draws them, so the same seed gives the same numbers) and rarcsine.
+//
+// Written for rStats.js (no ported code); the ways of drawing were taken from extraDistr's output and documentation:
+// rinvgamma is 1 / rgamma(alpha, rate = beta); rkumar and rpareto invert the distribution function at 1 - u; rtriang
+// mixes the minimum and maximum of two uniforms; rlaplace draws an exponential and then a uniform for the sign.
+// Invalid parameters give NaN without drawing. rarcsine (not in extraDistr) inverts the distribution function at u.
+import { qarcsine1 } from "../distributions/arcsine.js";
+import { qkumar1 } from "../distributions/kumaraswamy.js";
+import { qpareto1 } from "../distributions/pareto.js";
+import { generate } from "./common.js";
+import { rgamma1 } from "./gamma.js";
+import { expRand, unifRand } from "./rng.js";
+/**
+ * Random numbers from the inverse gamma distribution. Equivalent to `extraDistr::rinvgamma(n, alpha, beta = 1)`.
+ *
+ * @param n Number of values to generate.
+ * @param alpha Shape.
+ * @param options.beta Scale. Default `1`.
+ */
+export function rinvgamma(n, alpha, options = {}) {
+    const { beta = 1 } = options;
+    return generate(n, () => (alpha > 0 && beta > 0 && Number.isFinite(alpha) && Number.isFinite(beta) ? 1 / rgamma1(alpha, 1 / beta) : NaN));
+}
+/** Random numbers from the Kumaraswamy distribution. Equivalent to `extraDistr::rkumar(n, a, b)`. */
+export function rkumar(n, a, b) {
+    return generate(n, () => (a > 0 && b > 0 && Number.isFinite(a) && Number.isFinite(b) ? qkumar1(1 - unifRand(), a, b, true, false) : NaN));
+}
+/** Random numbers from the Pareto distribution. Equivalent to `extraDistr::rpareto(n, a = 1, b = 1)`. */
+export function rpareto(n, options = {}) {
+    const { a = 1, b = 1 } = options;
+    return generate(n, () => (a > 0 && b > 0 && Number.isFinite(a) && Number.isFinite(b) ? qpareto1(1 - unifRand(), a, b, true, false) : NaN));
+}
+/** Random numbers from the triangular distribution. Equivalent to `extraDistr::rtriang(n, a = -1, b = 1, c = (a + b)/2)`. */
+export function rtriang(n, options = {}) {
+    const { a = -1, b = 1 } = options;
+    const c = options.c ?? (a + b) / 2;
+    return generate(n, () => {
+        if (!(Number.isFinite(a) && Number.isFinite(b) && a < b && a <= c && c <= b))
+            return NaN;
+        const u = unifRand();
+        const v = unifRand();
+        const cc = (c - a) / (b - a);
+        return a + (b - a) * ((1 - cc) * Math.min(u, v) + cc * Math.max(u, v));
+    });
+}
+/** Random numbers from the Laplace distribution. Equivalent to `extraDistr::rlaplace(n, mu = 0, sigma = 1)`. */
+export function rlaplace(n, options = {}) {
+    const { mu = 0, sigma = 1 } = options;
+    return generate(n, () => {
+        if (!(Number.isFinite(mu) && sigma > 0 && Number.isFinite(sigma)))
+            return NaN;
+        const e = expRand();
+        const s = unifRand() > 0.5 ? 1 : -1;
+        return mu + sigma * s * e;
+    });
+}
+/** Random numbers from the arcsine distribution on (a, b), by inversion: `qarcsine(runif(n), a, b)`. */
+export function rarcsine(n, options = {}) {
+    const { a = 0, b = 1 } = options;
+    return generate(n, () => (Number.isFinite(a) && Number.isFinite(b) && a < b ? qarcsine1(unifRand(), a, b, true, false) : NaN));
+}

package/dist/random/gamma.js

@@ -4,7 +4,7 @@
 // technique" (Comm. ACM 25(1)); for shape < 1, algorithm GS from Ahrens & Dieter (1974)
 // "Computer methods for sampling from gamma, beta, Poisson and binomial distributions"
 // (Computing 12). These are the algorithms cited in R's ?rgamma, with the constants from the paper.
-// The order in which uniforms are consumed and the parameter handling follow observed R output. R's source was not consulted.
+// The order in which uniforms are consumed and the parameter handling follow observed R output.
 import { expRand, normRand, unifRand } from "./rng.js";
 import { gammaScale } from "../distributions/gamma.js";
 import { generate } from "./common.js";

package/dist/random/hyper.js

@@ -3,7 +3,7 @@
 // Kachitvichyanukul & Schmeiser (1985) "Computer generation of hypergeometric random
 // variates" (J. Statist. Comput. Simul. 22): algorithm HIN (inversion when the mode is near the lower end) and
 // H2PE (rejection over three regions). These are the algorithms cited in R's ?rhyper, with the constants from the paper.
-// The order in which uniforms are consumed and the parameter handling follow observed R output. R's source was not consulted.
+// The order in which uniforms are consumed and the parameter handling follow observed R output.
 import { unifRand } from "./rng.js";
 import { generate } from "./common.js";
 import { rbinom1 } from "./binomial.js";

package/dist/random/mt.js

@@ -4,7 +4,6 @@
 // equidistributed uniform pseudo-random number generator" (ACM TOMACS 8(1)).
 // How set.seed() expands the seed and maps to [0, 1) was verified against R's ?RNG documentation and
 // observations of R's .Random.seed and runif() output (test/fixtures/random*.json).
-// R's source was not consulted.
 const N = 624;
 const M = 397;
 const MATRIX_A = 0x9908b0df;

package/dist/random/poisson.js

@@ -4,7 +4,6 @@
 // distributions" (ACM TOMS 8(2)): algorithm PD. For mu < 10, table-based inversion (Case B in the paper);
 // for mu >= 10, normal approximation with rejection (Case A). These are the algorithms cited in R's ?rpois,
 // with the constants from the paper. The order in which uniforms are consumed and the parameter handling follow observed R output.
-// R's source was not consulted.
 import { expRand, normRand, unifRand } from "./rng.js";
 import { generate } from "./common.js";
 import { rgamma1 } from "./gamma.js";

package/dist/random/rank.js

@@ -1,6 +1,6 @@
 // Random Wilcoxon rank statistics: rsignrank and rwilcox.
 // Generated by definition (signs chosen by uniforms; ranks sampled without replacement).
-// Uniform consumption follows observed R output. R's source was not consulted.
+// Uniform consumption follows observed R output.
 import { unifIndex, unifRand } from "./rng.js";
 import { generate } from "./common.js";
 /**

package/dist/random/rng.js

@@ -2,7 +2,7 @@
 //
 // As in R, there is a single state (set by setSeed), corresponding to R's default
 // RNGkind("Mersenne-Twister", "Inversion", "Rejection").
-// The procedures were verified against R's ?RNG and ?Random documentation and observed output. R's source was not consulted.
+// The procedures were verified against R's ?RNG and ?Random documentation and observed output.
 import { MersenneTwister } from "./mt.js";
 import { ppnd16 } from "./as241.js";
 /** Value of .Random.seed[1]: uniform 3 (Mersenne-Twister) + 100 × normal 4 (Inversion) + 10000 × sample 1 (Rejection) */

package/dist/random/sample.js

@@ -4,7 +4,7 @@
 // Weighted sampling sorts the weights in decreasing order and selects by cumulative probability (with replacement,
 // when more than 200 candidates have large weights, Walker's (1977) alias method). The sort is a heapsort,
 // and the order of equal weights also matches R.
-// All of this was verified against observed R output. R's source was not consulted.
+// All of this was verified against observed R output.
 import { unifIndex, unifRand } from "./rng.js";
 /**
  * Random sample of integers from `0..n-1`. Equivalent to R's `sample.int()`, but 0-based.