hivemall-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From takuti <...@git.apache.org>
Subject [GitHub] incubator-hivemall pull request #66: [WIP][HIVEMALL-91] Implement Online LDA
Date Tue, 04 Apr 2017 09:06:02 GMT
Github user takuti commented on a diff in the pull request:

    https://github.com/apache/incubator-hivemall/pull/66#discussion_r109611010
  
    --- Diff: core/src/main/java/hivemall/lda/OnlineLDAModel.java ---
    @@ -0,0 +1,464 @@
    +/*
    + * 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.List;
    +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 {
    +
    +    // 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 List<Map<String, float[]>> phi_;
    +    private float[][] gamma_;
    +    private Map<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 List<Map<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_);
    +
    +        // 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();
    +    }
    +
    +    private void stepE() {
    +        // 1) Updating phi_
    +
    +        // dirichlet_expectation_2d(gamma_)
    +        float[][] ElogTheta = new float[miniBatchSize][K_];
    --- End diff --
    
    It's true, but plz wait until the final push after WIP condition. I will later check if
the algorithm is implemented correctly, and a new variable e.g. `expElogTheta` might be introduced.
If so, I guess `ElogTheta` is better in terms of both readability. It should be easy to understand
a relationship between `ElogTheta` and its exponential. 


---
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