@wlearn/automl 0.1.0 → 0.2.0

package/src/portfolio.js

@@ -6,17 +6,16 @@
  * zeroshot portfolio approach (TabRepo).
  */
 
-import { stratifiedKFold, kFold, normalizeX, normalizeY,
-  ValidationError } from '@wlearn/core'
-import { Executor } from './executor.js'
-import { detectTask } from './common.js'
-import { makeCandidateId } from './common.js'
+const { stratifiedKFold, kFold, normalizeX, normalizeY,
+  ValidationError } = require('@wlearn/core')
+const { Executor } = require('./executor.js')
+const { detectTask, makeCandidateId } = require('./common.js')
 
 // ---------------------------------------------------------------------------
 // Portfolio configs: task -> model_name -> list of param dicts
 // ---------------------------------------------------------------------------
 
-export const PORTFOLIO = {
+const PORTFOLIO = {
   classification: {
     xgb: [
       { objective: 'multi:softprob', eta: 0.05, max_depth: 6, numRound: 200,
@@ -103,6 +102,66 @@ export const PORTFOLIO = {
       { task: 'classification', nClauses: 500, threshold: 100, s: 5.0, nEpochs: 100 },
       { task: 'classification', nClauses: 50, threshold: 25, s: 2.0, nEpochs: 60 },
     ],
+    rf: [
+      { nEstimators: 100, maxDepth: 10, maxFeatures: 'sqrt', criterion: 'gini',
+        minSamplesSplit: 2, minSamplesLeaf: 1 },
+      { nEstimators: 300, maxDepth: 0, maxFeatures: 'log2', criterion: 'gini',
+        minSamplesSplit: 5, minSamplesLeaf: 2 },
+      { nEstimators: 200, maxDepth: 15, maxFeatures: 'sqrt', criterion: 'entropy',
+        extraTrees: 1, minSamplesSplit: 2 },
+      { nEstimators: 200, maxDepth: 12, maxFeatures: 'sqrt', criterion: 'hellinger',
+        minSamplesSplit: 2, minSamplesLeaf: 1 },
+      { nEstimators: 150, maxDepth: 10, maxFeatures: 'sqrt', criterion: 'gini',
+        heterogeneous: 1, oobWeighting: 1, minSamplesLeaf: 2 },
+      { nEstimators: 200, maxDepth: 10, maxFeatures: 'sqrt', criterion: 'gini',
+        histogramBinning: 1, minSamplesSplit: 2, minSamplesLeaf: 1 },
+      { nEstimators: 150, maxDepth: 12, maxFeatures: 'sqrt', criterion: 'gini',
+        jarf: 1, minSamplesSplit: 2, minSamplesLeaf: 1 },
+    ],
+    mlp: [
+      { hidden_sizes: [64], activation: 'relu', lr: 0.01, epochs: 50,
+        optimizer: 'adam', batch_size: 32 },
+      { hidden_sizes: [128, 64], activation: 'gelu', lr: 0.001, epochs: 100,
+        optimizer: 'adam', batch_size: 32 },
+      { hidden_sizes: [256, 128], activation: 'silu', lr: 0.001, epochs: 150,
+        optimizer: 'adam', batch_size: 16 },
+    ],
+    tabm: [
+      { hidden_sizes: [128], activation: 'relu', n_ensemble: 32, lr: 0.005,
+        epochs: 100, optimizer: 'adam' },
+      { hidden_sizes: [64], activation: 'gelu', n_ensemble: 8, lr: 0.01,
+        epochs: 50, optimizer: 'adam' },
+      { hidden_sizes: [128, 64], activation: 'silu', n_ensemble: 16, lr: 0.001,
+        epochs: 150, optimizer: 'adam' },
+    ],
+    nam: [
+      { hidden_sizes: [64], activation: 'exu', lr: 0.01, epochs: 100,
+        optimizer: 'adam' },
+      { hidden_sizes: [128], activation: 'relu', lr: 0.001, epochs: 100,
+        optimizer: 'adam' },
+      { hidden_sizes: [64, 32], activation: 'exu', lr: 0.005, epochs: 150,
+        optimizer: 'adam' },
+    ],
+    gam: [
+      { family: 'binomial', penalty: 'elasticnet', alpha: 0.5, nLambda: 100, nFolds: 5 },
+      { family: 'binomial', penalty: 'lasso', alpha: 1.0, nLambda: 100, nFolds: 5 },
+      { family: 'binomial', penalty: 'ridge', alpha: 0.0, nLambda: 100, nFolds: 5 },
+      { family: 'binomial', penalty: 'scad', nLambda: 100, nFolds: 5 },
+    ],
+    bart: [
+      { numTrees: 200, numBurnin: 200, numSamples: 100, alpha: 0.95, beta: 2.0 },
+      { numTrees: 500, numBurnin: 300, numSamples: 100, alpha: 0.95, beta: 2.0 },
+      { numTrees: 100, maxDepth: 5, numBurnin: 100, numSamples: 100,
+        alpha: 0.8, beta: 1.0 },
+    ],
+    fm: [
+      { k: 4, lr: 0.01, epoch: 10, opt: 'adagrad' },
+      { k: 8, lr: 0.005, epoch: 20, opt: 'adagrad', lambda: 0.001 },
+    ],
+    xlr: [
+      { lr: 0.01, opt: 'ftrl', epoch: 10 },
+      { lr: 0.01, lambda: 0.1, opt: 'adagrad', epoch: 20 },
+    ],
   },
   regression: {
     xgb: [
@@ -190,6 +249,66 @@ export const PORTFOLIO = {
       { task: 'regression', nClauses: 500, threshold: 100, s: 5.0, nEpochs: 100 },
       { task: 'regression', nClauses: 50, threshold: 25, s: 2.0, nEpochs: 60 },
     ],
+    rf: [
+      { nEstimators: 100, maxDepth: 10, maxFeatures: 'sqrt', criterion: 'mse',
+        minSamplesSplit: 2, minSamplesLeaf: 1 },
+      { nEstimators: 300, maxDepth: 0, maxFeatures: 'log2', criterion: 'mse',
+        minSamplesSplit: 5, minSamplesLeaf: 2 },
+      { nEstimators: 200, maxDepth: 15, maxFeatures: 'sqrt', criterion: 'mae',
+        extraTrees: 1, minSamplesSplit: 2 },
+      { nEstimators: 100, maxDepth: 10, maxFeatures: 'sqrt', criterion: 'mse',
+        leafModel: 1, minSamplesLeaf: 5 },
+      { nEstimators: 150, maxDepth: 10, maxFeatures: 'sqrt', criterion: 'mse',
+        heterogeneous: 1, oobWeighting: 1, minSamplesLeaf: 2 },
+      { nEstimators: 200, maxDepth: 10, maxFeatures: 'sqrt', criterion: 'mse',
+        histogramBinning: 1, minSamplesSplit: 2, minSamplesLeaf: 1 },
+      { nEstimators: 150, maxDepth: 12, maxFeatures: 'sqrt', criterion: 'mse',
+        jarf: 1, minSamplesSplit: 2, minSamplesLeaf: 1 },
+    ],
+    mlp: [
+      { hidden_sizes: [64], activation: 'relu', lr: 0.01, epochs: 50,
+        optimizer: 'adam', batch_size: 32 },
+      { hidden_sizes: [128, 64], activation: 'gelu', lr: 0.001, epochs: 100,
+        optimizer: 'adam', batch_size: 32 },
+      { hidden_sizes: [256, 128], activation: 'silu', lr: 0.001, epochs: 150,
+        optimizer: 'adam', batch_size: 16 },
+    ],
+    tabm: [
+      { hidden_sizes: [128], activation: 'relu', n_ensemble: 32, lr: 0.005,
+        epochs: 100, optimizer: 'adam' },
+      { hidden_sizes: [64], activation: 'gelu', n_ensemble: 8, lr: 0.01,
+        epochs: 50, optimizer: 'adam' },
+      { hidden_sizes: [128, 64], activation: 'silu', n_ensemble: 16, lr: 0.001,
+        epochs: 150, optimizer: 'adam' },
+    ],
+    nam: [
+      { hidden_sizes: [64], activation: 'exu', lr: 0.01, epochs: 100,
+        optimizer: 'adam' },
+      { hidden_sizes: [128], activation: 'relu', lr: 0.001, epochs: 100,
+        optimizer: 'adam' },
+      { hidden_sizes: [64, 32], activation: 'exu', lr: 0.005, epochs: 150,
+        optimizer: 'adam' },
+    ],
+    gam: [
+      { family: 'gaussian', penalty: 'elasticnet', alpha: 0.5, nLambda: 100, nFolds: 5 },
+      { family: 'gaussian', penalty: 'lasso', alpha: 1.0, nLambda: 100, nFolds: 5 },
+      { family: 'gaussian', penalty: 'ridge', alpha: 0.0, nLambda: 100, nFolds: 5 },
+      { family: 'gaussian', penalty: 'scad', nLambda: 100, nFolds: 5 },
+    ],
+    bart: [
+      { numTrees: 200, numBurnin: 200, numSamples: 100, alpha: 0.95, beta: 2.0 },
+      { numTrees: 500, numBurnin: 300, numSamples: 100, alpha: 0.95, beta: 2.0 },
+      { numTrees: 100, maxDepth: 5, numBurnin: 100, numSamples: 100,
+        alpha: 0.8, beta: 1.0 },
+    ],
+    fm: [
+      { k: 4, lr: 0.01, epoch: 10, opt: 'adagrad' },
+      { k: 8, lr: 0.005, epoch: 20, opt: 'adagrad', lambda: 0.001 },
+    ],
+    xlr: [
+      { lr: 0.01, opt: 'ftrl', epoch: 10 },
+      { lr: 0.01, lambda: 0.1, opt: 'adagrad', epoch: 20 },
+    ],
   },
 }
 
@@ -198,7 +317,7 @@ export const PORTFOLIO = {
  * @param {string} task - 'classification' or 'regression'
  * @returns {Object} model name -> config list
  */
-export function getPortfolio(task = 'classification') {
+function getPortfolio(task = 'classification') {
   return PORTFOLIO[task] || PORTFOLIO.classification
 }
 
@@ -210,7 +329,7 @@ export function getPortfolio(task = 'classification') {
  * Yields pre-tuned configs from the zeroshot portfolio.
  * Same interface as RandomStrategy / HalvingStrategy.
  */
-export class PortfolioStrategy {
+class PortfolioStrategy {
   #queue = []
   #index = 0
   #total = 0
@@ -258,7 +377,7 @@ export class PortfolioStrategy {
 /**
  * Evaluate pre-tuned portfolio configs with cross-validation.
  */
-export class PortfolioSearch {
+class PortfolioSearch {
   #models
   #opts
   #leaderboard = null
@@ -330,3 +449,5 @@ export class PortfolioSearch {
   get leaderboard() { return this.#leaderboard }
   get bestResult() { return this.#bestResult }
 }
+
+module.exports = { PORTFOLIO, getPortfolio, PortfolioStrategy, PortfolioSearch }

package/src/progressive.js

@@ -1,8 +1,8 @@
-import { stratifiedKFold, kFold, normalizeX, normalizeY,
-  ValidationError } from '@wlearn/core'
-import { Executor } from './executor.js'
-import { ProgressiveStrategy } from './strategy-progressive.js'
-import { detectTask, scorerGreaterIsBetter } from './common.js'
+const { stratifiedKFold, kFold, normalizeX, normalizeY,
+  ValidationError } = require('@wlearn/core')
+const { Executor } = require('./executor.js')
+const { ProgressiveStrategy } = require('./strategy-progressive.js')
+const { detectTask, scorerGreaterIsBetter } = require('./common.js')
 
 /**
  * Progressive search: probe all candidates cheaply (1 fold + subsample),
@@ -12,7 +12,7 @@ import { detectTask, scorerGreaterIsBetter } from './common.js'
  * The probe phase filters out bad configs quickly, saving time
  * for thorough evaluation of promising candidates.
  */
-export class ProgressiveSearch {
+class ProgressiveSearch {
   #models
   #opts
   #leaderboard = null
@@ -154,3 +154,5 @@ export class ProgressiveSearch {
   get leaderboard() { return this.#leaderboard }
   get bestResult() { return this.#bestResult }
 }
+
+module.exports = { ProgressiveSearch }

package/src/sampler.js

@@ -1,11 +1,11 @@
-import { makeLCG } from '@wlearn/core'
+const { makeLCG } = require('@wlearn/core')
 
 const { floor, round, log, exp, min, max } = Math
 
 /**
  * Sample a single value from a SearchParam definition.
  */
-export function sampleParam(param, rng) {
+function sampleParam(param, rng) {
   const { type } = param
   switch (type) {
     case 'categorical':
@@ -26,7 +26,7 @@ export function sampleParam(param, rng) {
 /**
  * Sample a complete config from a SearchSpace, respecting conditions.
  */
-export function sampleConfig(space, rng) {
+function sampleConfig(space, rng) {
   const config = {}
   const keys = Object.keys(space)
 
@@ -56,7 +56,7 @@ export function sampleConfig(space, rng) {
 /**
  * Generate n random configs from a SearchSpace.
  */
-export function randomConfigs(space, n, { seed = 42 } = {}) {
+function randomConfigs(space, n, { seed = 42 } = {}) {
   const rng = makeLCG(seed)
   const configs = []
   for (let i = 0; i < n; i++) {
@@ -69,7 +69,7 @@ export function randomConfigs(space, n, { seed = 42 } = {}) {
  * Enumerate grid points from a SearchSpace.
  * Continuous params discretized to `steps` values.
  */
-export function gridConfigs(space, { steps = 5 } = {}) {
+function gridConfigs(space, { steps = 5 } = {}) {
   const keys = Object.keys(space)
   if (keys.length === 0) return [{}]
 
@@ -160,3 +160,5 @@ function _discretize(param, steps) {
       throw new Error(`Unknown SearchParam type: "${type}"`)
   }
 }
+
+module.exports = { sampleParam, sampleConfig, randomConfigs, gridConfigs }

package/src/search.js

@@ -1,13 +1,13 @@
-import { stratifiedKFold, kFold, normalizeX, normalizeY,
-  ValidationError } from '@wlearn/core'
-import { Executor } from './executor.js'
-import { RandomStrategy } from './strategy-random.js'
-import { detectTask, scorerGreaterIsBetter } from './common.js'
+const { stratifiedKFold, kFold, normalizeX, normalizeY,
+  ValidationError } = require('@wlearn/core')
+const { Executor } = require('./executor.js')
+const { RandomStrategy } = require('./strategy-random.js')
+const { detectTask, scorerGreaterIsBetter } = require('./common.js')
 
 /**
  * Random hyperparameter search with cross-validation.
  */
-export class RandomSearch {
+class RandomSearch {
   #models
   #opts
   #leaderboard = null
@@ -91,3 +91,5 @@ export class RandomSearch {
   get leaderboard() { return this.#leaderboard }
   get bestResult() { return this.#bestResult }
 }
+
+module.exports = { RandomSearch }

package/src/strategy-halving.js

@@ -1,6 +1,6 @@
-import { makeLCG } from '@wlearn/core'
-import { sampleConfig } from './sampler.js'
-import { makeCandidateId } from './common.js'
+const { makeLCG } = require('@wlearn/core')
+const { sampleConfig } = require('./sampler.js')
+const { makeCandidateId } = require('./common.js')
 
 const { ceil, log, max, min, floor } = Math
 
@@ -14,7 +14,7 @@ const { ceil, log, max, min, floor } = Math
  * in the current round have been evaluated. next() returns null
  * only when fully done.
  */
-export class HalvingStrategy {
+class HalvingStrategy {
   #candidates       // all candidates for current round
   #roundIndex = 0   // index within current round's candidates
   #round = 0        // current round number
@@ -155,3 +155,5 @@ export class HalvingStrategy {
     return this.#rounds
   }
 }
+
+module.exports = { HalvingStrategy }

package/src/strategy-progressive.js

@@ -1,6 +1,6 @@
-import { makeLCG } from '@wlearn/core'
-import { sampleConfig } from './sampler.js'
-import { makeCandidateId } from './common.js'
+const { makeLCG } = require('@wlearn/core')
+const { sampleConfig } = require('./sampler.js')
+const { makeCandidateId } = require('./common.js')
 
 const { max, ceil } = Math
 
@@ -14,7 +14,7 @@ const { max, ceil } = Math
  * This pairs with ProgressiveSearch which manages two Executors
  * (probe executor with 1 fold, full executor with K folds).
  */
-export class ProgressiveStrategy {
+class ProgressiveStrategy {
   #allCandidates = []
   #promotedCandidates = []
   #phase = 'probe'
@@ -124,3 +124,5 @@ export class ProgressiveStrategy {
     return false
   }
 }
+
+module.exports = { ProgressiveStrategy }

package/src/strategy-random.js

@@ -1,12 +1,12 @@
-import { makeLCG } from '@wlearn/core'
-import { sampleConfig } from './sampler.js'
-import { makeCandidateId } from './common.js'
+const { makeLCG } = require('@wlearn/core')
+const { sampleConfig } = require('./sampler.js')
+const { makeCandidateId } = require('./common.js')
 
 /**
  * Random search strategy: generates nIter random configs per model,
  * yields them one at a time. No adaptive behavior.
  */
-export class RandomStrategy {
+class RandomStrategy {
   #queue = []
   #index = 0
   #total = 0
@@ -65,3 +65,5 @@ export class RandomStrategy {
     return this.#index >= this.#total
   }
 }
+
+module.exports = { RandomStrategy }