hivemall-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From myui <...@git.apache.org>
Subject [GitHub] incubator-hivemall pull request #66: [WIP][HIVEMALL-91] Implement Online LDA
Date Tue, 04 Apr 2017 01:03:22 GMT
Github user myui commented on a diff in the pull request:

    https://github.com/apache/incubator-hivemall/pull/66#discussion_r109558404
  
    --- Diff: core/src/main/java/hivemall/lda/OnlineLDAModel.java ---
    @@ -0,0 +1,497 @@
    +/*
    + * Licensed to the Apache Software Foundation (ASF) under one
    + * or more contributor license agreements.  See the NOTICE file
    + * distributed with this work for additional information
    + * regarding copyright ownership.  The ASF licenses this file
    + * to you under the Apache License, Version 2.0 (the
    + * "License"); you may not use this file except in compliance
    + * with the License.  You may obtain a copy of the License at
    + *
    + *   http://www.apache.org/licenses/LICENSE-2.0
    + *
    + * Unless required by applicable law or agreed to in writing,
    + * software distributed under the License is distributed on an
    + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
    + * KIND, either express or implied.  See the License for the
    + * specific language governing permissions and limitations
    + * under the License.
    + */
    +package hivemall.lda;
    +
    +import hivemall.utils.lang.Preconditions;
    +
    +import java.util.ArrayList;
    +import java.util.Arrays;
    +import java.util.Collections;
    +import java.util.Map;
    +import java.util.HashMap;
    +import java.util.SortedMap;
    +import java.util.TreeMap;
    +
    +import org.apache.commons.math3.distribution.GammaDistribution;
    +import org.apache.commons.math3.special.Gamma;
    +
    +public final class OnlineLDAModel {
    +
    +    private static final boolean printLambda = false;
    +    private static final boolean printGamma = false;
    +    private static final boolean printPhi = false;
    +
    +    // number of topics
    +    private int K_;
    +
    +    // prior on weight vectors "theta ~ Dir(alpha_)"
    +    private float alpha_ = 1 / 2.f;
    +
    +    // prior on topics "beta"
    +    private float eta_ = 1 / 20.f;
    +
    +    // total number of documents
    +    // in the truly online setting, this can be an estimate of the maximum number of
documents that could ever seen
    +    private int D_ = 11102;
    +
    +    // defined by (tau0 + countEMStep)^(-kappa_)
    +    private double rhot;
    +
    +    // positive value which downweights early iterations
    +    private double tau0_ = 1020;
    +
    +    // exponential decay rate (i.e., learning rate) which must be in (0.5, 1] to guarantee
convergence
    +    private double kappa_ = 0.7;
    +
    +    // random number generator
    +    GammaDistribution gd;
    +    private double SHAPE = 100d;
    +    private double SCALE = 1.d / SHAPE;
    +
    +    // parameters
    +    private ArrayList<HashMap<String, float[]>> phi_;
    +    private float[][] gamma_;
    +    private HashMap<String, float[]> lambda_;
    +
    +    // check convergence in the expectation (E) step
    +    private double DELTA = 1E-5;
    +
    +    // for update size of lambda_
    +    private int dummySize = 100;
    +    private float[][] dummyLambdas;
    +
    +    private ArrayList<HashMap<String, Float>> miniBatchMap;
    +    private int miniBatchSize;
    +
    +    private int accumDocCount = 0;
    +    private int accumWordCount = 0;
    +
    +    public OnlineLDAModel(int K, float alpha, float eta, int D, double tau0, double kappa)
{
    +        Preconditions.checkArgument(0.d < tau0, "tau0 MUST be positive: " + tau0);
    +        Preconditions.checkArgument(0.5 < kappa && kappa <= 1.d,
    +            "kappa MUST be in (0.5, 1.0]: " + kappa);
    +
    +        K_ = K;
    +        alpha_ = alpha;
    +        eta_ = eta;
    +        D_ = D;
    +        tau0_ = tau0;
    +        kappa_ = kappa;
    +
    +        // initialize a random number generator
    +        gd = new GammaDistribution(SHAPE, SCALE);
    +        gd.reseedRandomGenerator(1001);
    +
    +        // initialize the parameters
    +        lambda_ = new HashMap<String, float[]>();
    +
    +        setDummyLambda();
    +    }
    +
    +    private void setDummyLambda() {
    +        dummyLambdas = new float[dummySize][];
    +        double[] tmpDArray;
    +        for (int b = 0; b < dummySize; b++) {
    +            float[] tmpDummyLambda = new float[K_];
    +            tmpDArray = gd.sample(K_);
    +            for (int k = 0; k < K_; k++) {
    +                tmpDummyLambda[k] = (float) tmpDArray[k];
    +            }
    +            dummyLambdas[b] = tmpDummyLambda;
    +        }
    +    }
    +
    +    public void train(String[][] miniBatch, int time) {
    +        miniBatchSize = miniBatch.length;
    +
    +        rhot = Math.pow(tau0_ + time, -kappa_);
    +
    +        if (printLambda) {
    +            System.out.println("lambda:");
    +            for (String key : lambda_.keySet()) {
    +                System.out.println(Arrays.toString(lambda_.get(key)));
    +            }
    +        }
    +
    +        // get the number of words(Nd) for each documents
    +        getMiniBatchParams(miniBatch);
    +        accumDocCount += miniBatchSize;
    +
    +        makeMiniBatchMap(miniBatch);
    +
    +        updateSizeOfParameterForMiniBatch();
    +
    +        // Expectation
    +        float[][] lastGamma;
    +        do {
    +            // (deep) copy the last gamma values
    +            lastGamma = new float[gamma_.length][];
    +            for (int d = 0; d < gamma_.length; d++) {
    +                lastGamma[d] = gamma_[d].clone();
    +            }
    +
    +            stepE();
    +        } while (!checkGammaDiff(lastGamma, gamma_));
    +
    +        // Maximization
    +        stepM();
    +
    +        if (printGamma) {
    +            System.out.println("gamma:");
    +            for (int d = 0; d < miniBatchSize; d++) {
    +                System.out.println(Arrays.toString(gamma_[d]));
    +            }
    +        }
    +
    +        if (printPhi) {
    +            System.out.println("phi");
    +            for (int d = 0; d < miniBatchSize; d++) {
    +                for (String label : miniBatchMap.get(d).keySet()) {
    +                    System.out.println(Arrays.toString(phi_.get(d).get(label)));
    +                }
    +            }
    +        }
    +    }
    +
    +    private void stepE() {
    +        // 1) Updating phi_
    +
    +        // dirichlet_expectation_2d(gamma_)
    +        float[][] ElogTheta = new float[miniBatchSize][K_];
    +        for (int d = 0; d < miniBatchSize; d++) {
    +            float gammaSum_d = 0.f;
    +            for (int k = 0; k < K_; k++) {
    +                gammaSum_d += gamma_[d][k];
    +            }
    +            for (int k = 0; k < K_; k++) {
    +                ElogTheta[d][k] = (float) (Gamma.digamma(gamma_[d][k]) - Gamma.digamma(gammaSum_d));
    +            }
    +        }
    +
    +        // dirichlet_expectation_2d(lambda_)
    +        HashMap<String, float[]> ElogBeta = new HashMap<String, float[]>();
    +        for (int k = 0; k < K_; k++) {
    +            float lambdaSum_k = 0.f;
    +            for (String label : lambda_.keySet()) {
    +                lambdaSum_k += lambda_.get(label)[k];
    +            }
    +            for (int d = 0; d < miniBatchSize; d++) {
    +                for (String label : miniBatchMap.get(d).keySet()) {
    +                    float[] ElogBeta_label;
    +                    if (ElogBeta.containsKey(label)) {
    +                        ElogBeta_label = ElogBeta.get(label);
    +                    } else {
    +                        ElogBeta_label = new float[K_];
    +                        Arrays.fill(ElogBeta_label, 0.f);
    +                    }
    +
    +                    ElogBeta_label[k] = (float) (Gamma.digamma(lambda_.get(label)[k])
- Gamma.digamma(lambdaSum_k));
    +                    ElogBeta.put(label, ElogBeta_label);
    +                }
    +            }
    +        }
    +
    +        // updating phi_ w/ normalization
    +        for (int d = 0; d < miniBatchSize; d++) {
    +            for (String label : miniBatchMap.get(d).keySet()) {
    +                float normalizer = 0.f;
    +                for (int k = 0; k < K_; k++) {
    +                    phi_.get(d).get(label)[k] = (float) Math.exp(ElogTheta[d][k]
    +                            + ElogBeta.get(label)[k]) + 1E-20f;
    +                    normalizer += phi_.get(d).get(label)[k];
    +                }
    +
    +                // normalize
    +                for (int k = 0; k < K_; k++) {
    +                    phi_.get(d).get(label)[k] /= normalizer;
    +                }
    +            }
    +        }
    +
    +        // 2) Updating gamma_
    +        for (int d = 0; d < miniBatchSize; d++) {
    +            for (int k = 0; k < K_; k++) {
    +                float gamma_dk = alpha_;
    +                for (String label : miniBatchMap.get(d).keySet()) {
    +                    gamma_dk += phi_.get(d).get(label)[k] * miniBatchMap.get(d).get(label);
    +                }
    +                gamma_[d][k] = gamma_dk;
    +            }
    +        }
    +    }
    +
    +    private void stepM() {
    +        // calculate lambdaBar
    +        HashMap<String, float[]> lambdaBar = new HashMap<String, float[]>();
    +
    +        float docRatio = (float) D_ / (float) miniBatchSize;
    +
    +        for (int d = 0; d < miniBatchSize; d++) {
    +            for (String label : miniBatchMap.get(d).keySet()) {
    +                float[] lambdaBar_label;
    +                if (lambdaBar.containsKey(label)) {
    +                    lambdaBar_label = lambdaBar.get(label);
    +                } else {
    +                    lambdaBar_label = new float[K_];
    +                    Arrays.fill(lambdaBar_label, eta_);
    +                }
    +                for (int k = 0; k < K_; k++) {
    +                    lambdaBar_label[k] += docRatio * phi_.get(d).get(label)[k];
    +                }
    +                lambdaBar.put(label, lambdaBar_label);
    +            }
    +        }
    +
    +        // update lambda_
    +        for (String label : lambda_.keySet()) {
    +            float[] lambda_label = lambda_.get(label);
    +            float[] lambdaBar_label;
    +            if (lambdaBar.containsKey(label)) {
    +                lambdaBar_label = lambdaBar.get(label);
    +            } else {
    +                lambdaBar_label = new float[K_];
    +                Arrays.fill(lambdaBar_label, eta_);
    +            }
    +            for (int k = 0; k < K_; k++) {
    +                lambda_label[k] = (float) ((1.d - rhot) * lambda_label[k] + rhot * lambdaBar_label[k]);
    +            }
    +            lambda_.put(label, lambda_label);
    +        }
    +    }
    +
    +    private void getMiniBatchParams(String[][] miniBatch) {
    +        miniBatchSize = miniBatch.length;
    +
    +        for (int d = 0; d < miniBatchSize; d++) {
    +            accumWordCount += miniBatch[d].length;
    +        }
    +    }
    +
    +    private void makeMiniBatchMap(String[][] miniBatch) {
    +        miniBatchMap = new ArrayList<HashMap<String, Float>>(); // initialize
    +
    +        // parse document
    +        for (int d = 0; d < miniBatchSize; d++) {
    +            HashMap<String, Float> docMap = new HashMap<String, Float>();
    +
    +            // parse features
    +            for (int w = 0; w < miniBatch[d].length; w++) {
    +                String fv = miniBatch[d][w];
    +                String[] parsedFeature = fv.split(":"); // [`label`, `value`]
    +                if (parsedFeature.length == 1) { // wrong format
    +                    continue;
    +                }
    +                String label = parsedFeature[0];
    +                float value = Float.parseFloat(parsedFeature[1]);
    +                docMap.put(label, value);
    +            }
    +
    +            miniBatchMap.add(docMap);
    +        }
    +    }
    +
    +    private void updateSizeOfParameterForMiniBatch() {
    +        phi_ = new ArrayList<HashMap<String, float[]>>();
    +        gamma_ = new float[miniBatchSize][];
    +
    +        for (int d = 0; d < miniBatchSize; d++) {
    +            float[] gammad = getRandomGammaArray();
    +            gamma_[d] = gammad;
    +
    +            // phi_ not needed to be initialized
    +            HashMap<String, float[]> phid = new HashMap<String, float[]>();
    --- End diff --
    
    HashMap => Map


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

Mime
View raw message