For the vectorizer, what's the output feature dimension and are you creating sparse vectors or dense vectors? The model on the driver consists of numClasses * numFeatures doubles. However, the driver needs more memory in order to receive the task result (of the same size) from executors. So you need to control the feature dimension (this is why people use the hashing trick) and reduce the number of partitions. -Xiangrui
On Wed, Sep 24, 2014 at 10:59 AM, jatinpreet <jatinpr...@gmail.com> wrote: > Hi, > > I was able to get the training running in local mode with default settings, > there was a problem with document labels which were quite large(not 20 as > suggested earlier). > > I am currently training 175000 documents on a single node with 2GB of > executor memory and 5GB of driver memory successfully. If I increase the > number of documents, I get the OOM error. I wish to understand what > generally the bottlenecks are for naive bayes, is it the executor or the > driver memory? Also, what are the things to keep in mind while training huge > sets of data so that I can have a bullet proof classification system, > slowing down in case of low memory is fine but not exceptions. > > As a side note, is there any classification algorithm in MLlib which can > just append the new training data to an existing model? With naive bayes, I > need to have all the data available at once for training. > > Thanks, > Jatin > > > > ----- > Novice Big Data Programmer > -- > View this message in context: > http://apache-spark-user-list.1001560.n3.nabble.com/Out-of-memory-exception-in-MLlib-s-naive-baye-s-classification-training-tp14809p15052.html > Sent from the Apache Spark User List mailing list archive at Nabble.com. > > --------------------------------------------------------------------- > To unsubscribe, e-mail: user-unsubscr...@spark.apache.org > For additional commands, e-mail: user-h...@spark.apache.org > --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
