Using a random seed in Rmallet

Question

Is there an option (or a workaround) in Rmallet to use a random seed, as is possible through the mallet command line (i.e. --random-seed 1)?

score 2 · Accepted Answer · answered May 21 '16 at 14:02

2

Yes, via the rJava interface to the underlying ParallelTopicModel.setRandomSeed method (see here)

library(mallet)
library(rJava)
m <- MalletLDA(num.topics=20, alpha.sum=5, beta=0.1)
m$model$setRandomSeed(42L)

The seed has to be an explicit integer (hence the L in 42L). If you are using the development version of RMallet from github, you'd need

m$setRandomSeed(42L)

answered May 21 '16 at 14:02

agoldst

Great answer; thanks a bunch! m$model$setRandomSeed(42L) worked great as long as it was called before documents were loaded. – shackett May 21 '16 at 17:54
1

You're welcome. Glad it worked. You're right about setting the seed before loading documents. I had to look at [the source](https://github.com/mimno/Mallet/blob/master/src/cc/mallet/topics/ParallelTopicModel.java#L224) to understand why there was any randomness there: the initial random assignment of tokens to topics is done when you load documents. – agoldst May 22 '16 at 14:48

1 Answers1