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: [HIVEMALL-91] Implement Online LDA
Date Mon, 17 Apr 2017 10:30:43 GMT
Github user myui commented on a diff in the pull request:

    https://github.com/apache/incubator-hivemall/pull/66#discussion_r111728179
  
    --- Diff: core/src/main/java/hivemall/topicmodel/OnlineLDAModel.java ---
    @@ -350,66 +348,80 @@ private float computeApproxBoundForMiniBatch() {
             float score = 0.f;
     
             // prepare
    -        final float[] gammaSum = new float[_miniBatchSize];
    -        for (int d = 0; d < gammaSum.length; d++) {
    -            gammaSum[d] += MathUtils.sum(_gamma[d], _K);
    +        float[] gammaSum = new float[_miniBatchSize];
    +        Arrays.fill(gammaSum, 0.f);
    +        for (int d = 0; d < _miniBatchSize; d++) {
    +            for (int k = 0; k < _K; k++) {
    +                gammaSum[d] += _gamma[d][k];
    +            }
             }
    -        final float[] lambdaSum = new float[_K];
    -        for (float[] lambda_l : _lambda.values()) {
    -            MathUtils.add(lambdaSum, lambda_l, _K);
    +        float[] lambdaSum = new float[_K];
    +        Arrays.fill(lambdaSum, 0.f);
    +        for (int k = 0; k < _K; k++) {
    +            for (String label : _lambda.keySet()) {
    +                lambdaSum[k] += _lambda.get(label)[k];
    +            }
             }
     
    -        // E[log p(docs | theta, beta)]
             for (int d = 0; d < _miniBatchSize; d++) {
    -            // for each word in the document
    -            for (String label : _miniBatchMap.get(d).keySet()) {
    -                float wordCount = _miniBatchMap.get(d).get(label);
    -
    -                float tmp = 0.f;
    +            // E[log p(doc | theta, beta)]
    +            Map<String, Float> doc = _miniBatchMap.get(d);
    +            for (String label : doc.keySet()) { // for each word in the document
    +                float[] temp = new float[_K];
    +                float max = Float.MIN_VALUE;
                     for (int k = 0; k < _K; k++) {
                         float eLogTheta_dk = (float) (Gamma.digamma(_gamma[d][k]) - Gamma.digamma(gammaSum[d]));
    -                    float eLogBeta_kw = 0.f;
    -                    if (_lambda.containsKey(d)) {
    -                        eLogBeta_kw = (float) (Gamma.digamma(_lambda.get(d)[k]
    -                                - Gamma.digamma(lambdaSum[k])));
    +                    float eLogBeta_kw = (float) (Gamma.digamma(_lambda.get(label)[k])
    --- End diff --
    
    hash lookup `_lambda.get(label)` should be placed out of loop.


---
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