It's the combination of the 1M dataset and the Slope1 using the MemoryDiffStorage structure- MemoryDiffStorage seems to be in general a space-eater. Look at the javadoc for it: 'maxEntries' lets you trade space for item-item accuracy.
Lance On Sat, Jan 1, 2011 at 11:46 AM, Jud <[email protected]> wrote: > Lance Norskog <goksron <at> gmail.com> writes: > >> >> Try more memory. I had problems with the 1M Grouplens set and Slope1. >> >> You might also try the first N lines of the rating.dat file. "head >> -50000 < rating.dat > mini.dat" will give you a smaller dataset. >> >> It would be great to have a tool that isolates a heavily connected >> subgraph of users & ids. >> > > Hey Lance and thank you very much for this information. > > After days and nights of doing anything anyone could think of....inc.: > Swapfiles... > Swapfile manager > Reinstall > Java Heap size... > Installing Eclipse for manging Java Heap size > Trying Mahout-0.4, mahout-0.2 etc... > Reading endless articles about this problem.... > > LANCE IS RIGHT!!!! > > Owen - pay attention....! :-) > > Lance thanx again... > > Jud. > > -- Lance Norskog [email protected]
