Online LDA using Hoffman's Python Implementation.
Usage
$ python online_lda.py -h
usage: online_lda.py [-h] [-o OUTDIR] [-b BATCHSIZE] [-d NUM_DOCS]
[-k NUM_TOPICS] [-t TAU_0] [-l KAPPA] [-m MODEL_OUT_FREQ]
dataset vocab_file
positional arguments:
dataset Input dataset filename.
vocab_file Vocabulary filename.
optional arguments:
-h, --help show this help message and exit
-o OUTDIR, --outdir OUTDIR
Directory to place output files. (default='')
-b BATCHSIZE, --batchsize BATCHSIZE
Batch size. (default=256)
-d NUM_DOCS, --num_docs NUM_DOCS
Total # docs in dataset. (default=7990787)
-k NUM_TOPICS, --num_topics NUM_TOPICS
Number of topics. (default=100)
-t TAU_0, --tau_0 TAU_0
Tau learning parameter to downweight early documents
(default=1024)
-l KAPPA, --kappa KAPPA
Kappa learning parameter; decay factor for influence
of batches.(default=0.7)
-m MODEL_OUT_FREQ, --model_out_freq MODEL_OUT_FREQ
Number of iterations interval for outputting a model
file. (default=10000)Used in the blog post http://wellecks.wordpress.com/2014/10/26/ldaoverflow-with-online-lda/