Extremely slow and running out of memory, but not using Java memory optimaly
Contributor IHi,
During handling not thet extensive datasets datasets (6000 records with microtext), Rapidminer signals it runs out of memory and on others it's just extremely slow (have a knn-2 running for 22hrs now). I have 36Gb of ram, perform quite some preprocessing (including strong pruning while creating TF-IDF vector, not keeping any text either), and should have 8Gb available for Java. However, for some reason, Rapidminer only uses only 256Mb, while no other tasks are taking away more than 50Mb at any time. Is there a setting hidden somewhere, where this can be adapted? I'm guessing that is the bottleneck...
I'm using the latest version of RapidMiner Studio Educational in a Windows10 environment (64bit). Already went through the .bat file for MAX_JAVA_MEMORY as mentioned here in some other posts, but that string is simpely not in there. Did this move maybe?
Answers
The way we handle memory has changed from v5.3 to the current v7.5, so I don't know what versions you are comparing. I don't know what your process looks like but if you generating n-grams, you could easily run out of memory depending on your set up. It's really hard to troubleshoot without looking at how your process is built along with where it chokes.