Github user witgo commented on the pull request: https://github.com/apache/spark/pull/2376#issuecomment-55913971 I tried to reproduce it with the way the test case, but was unsuccessful.But like the code below does appear the problem, do not know why ```scala def runGibbsSampling( data: RDD[Document], initModel: TopicModel, totalIter: Int, burnInIter: Int ): (TopicModel, RDD[Document]) = { require(totalIter > burnInIter, "totalIter is less than burnInIter") require(totalIter > 0, "totalIter is less than 0") require(burnInIter > 0, "burnInIter is less than 0") val (numTopics, numTerms, alpha, beta) = (initModel.topicCounts_.size, initModel.topicTermCounts_.head.size, initModel.alpha, initModel.beta) val probModel = TopicModel(numTopics, numTerms, alpha, beta) logInfo("Start initialization") var (topicModel, corpus) = sampleTermAssignment(data, initModel) for (iter <- 1 to totalIter) { logInfo("Start Gibbs sampling (Iteration %d/%d)".format(iter, totalIter)) val broadcastModel = data.context.broadcast(topicModel) val previousCorpus = corpus corpus = corpus.mapPartitions { docs => val rand = new Random val topicModel = broadcastModel.value val topicThisTerm = BDV.zeros[Double](numTopics) docs.map { doc => val content = doc.content val topics = doc.topics val topicsDist = doc.topicsDist for (i <- 0 until content.length) { val term = content(i) val topic = topics(i) val chosenTopic = topicModel.dropOneDistSampler(topicsDist, topicThisTerm, rand, term, topic) if (topic != chosenTopic) { topics(i) = chosenTopic topicsDist(topic) += -1 topicsDist(chosenTopic) += 1 topicModel.update(term, topic, -1) topicModel.update(term, chosenTopic, 1) } } doc } }.setName(s"LDA-$iter").persist(StorageLevel.MEMORY_AND_DISK) if (iter % 5 == 0 && data.context.getCheckpointDir.isDefined) { corpus.checkpoint() } topicModel = collectTopicCounters(corpus, numTerms, numTopics) if (iter > burnInIter) { probModel.merge(topicModel) } previousCorpus.unpersist() broadcastModel.unpersist() } val burnIn = (totalIter - burnInIter).toDouble probModel.topicCounts_ :/= burnIn probModel.topicTermCounts_.foreach(_ :/= burnIn) (probModel, corpus) } ```
--- 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 infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org