Re: why decision trees do binary split?
Hello, There is a big compelling reason for binary splits in general for trees: a split is made if the difference between the two resulting branches is significant.You also want to compare the significance of this candidate split vs all the other candidate splits. There are many statistical tests to compare two groups. You can even generate something like p-values that, according to some, allow you to compare different candidate splits. If you introduce multibranch splits... things become much more messy. Also, mind that breaking categorical variables into as many groups as there are levels would in some cases separate subgroups of variables which are not that different. Successive binary splits could potentially provide you with the required homogeneous subsets. Best, Carlos J. Gil Bellosta http://www.datanalytics.com 2014-11-06 10:46 GMT+01:00 Sean Owen so...@cloudera.com: I haven't seen that done before, which may be most of the reason - I am not sure that is common practice. I can see upsides - you need not pick candidate splits to test since there is only one N-way rule possible. The binary split equivalent is N levels instead of 1. The big problem is that you are always segregating the data set entirely, and making the equivalent of those N binary rules, even when you would not otherwise bother because they don't add information about the target. The subsets matching each child are therefore unnecessarily small and this makes learning on each independent subset weaker. On Nov 6, 2014 9:36 AM, jamborta jambo...@gmail.com wrote: I meant above, that in the case of categorical variables it might be more efficient to create a node on each categorical value. Is there a reason why spark went down the binary route? thanks, -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/why-decision-trees-do-binary-split-tp18188p18265.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
Re: SparkR: split, apply, combine strategy for dataframes?
Thanks for your reply. I think that the problem was that SparkR tried to serialize the whole environment. Mind that the large dataframe was part of it. So every worker received their slice / partition (which is very small) plus the whole thing! So I deleted the large dataframe and list before parallelizing and the cluster ran without memory issues. Best, Carlos J. Gil Bellosta http://www.datanalytics.com 2014-08-15 3:53 GMT+02:00 Shivaram Venkataraman shiva...@eecs.berkeley.edu: Could you try increasing the number of slices with the large data set ? SparkR assumes that each slice (or partition in Spark terminology) can fit in memory of a single machine. Also is the error happening when you do the map function or does it happen when you combine the results ? Thanks Shivaram On Thu, Aug 14, 2014 at 3:53 PM, Carlos J. Gil Bellosta gilbello...@gmail.com wrote: Hello, I am having problems trying to apply the split-apply-combine strategy for dataframes using SparkR. I have a largish dataframe and I would like to achieve something similar to what ddply(df, .(id), foo) would do, only that using SparkR as computing engine. My df has a few million records and I would like to split it by id and operate on the pieces. These pieces are quite small in size: just a few hundred records. I do something along the following lines: 1) Use split to transform df into a list of dfs. 2) parallelize the resulting list as a RDD (using a few thousand slices) 3) map my function on the pieces using Spark. 4) recombine the results (do.call, rbind, etc.) My cluster works and I can perform medium sized batch jobs. However, it fails with my full df: I get a heap space out of memory error. It is funny as the slices are very small in size. Should I send smaller batches to my cluster? Is there any recommended general approach to these kind of split-apply-combine problems? Best, Carlos J. Gil Bellosta http://www.datanalytics.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