entropix 1.0.0

Sampling SVD singular vectors for alternative DSM

entropix

Generate count-based Distributional Semantic Models by sampling SVD singular vectors instead of using top components.

Install

pip install entropix

or, after a git clone:

python3 setup.py install

Use

Sequential mode

entropix sample \
--model /abs/path/to/dense/numpy/model.npy \
--vocab /abs/path/to/corresponding/model.vocab \
--dataset dataset_to_optimize_on \  # men, simlex or simverb
--shuffle \
--mode seq \
--kfold-size .2 \  # size of kfold, between 0 and .5
--metric pearson \  # spr(spearman), pearson, rmse or both (spr+rmse)
--num-threads 5

Limit mode

entropix sample \
--model /abs/path/to/dense/numpy/model.npy \
--vocab /abs/path/to/corresponding/model.vocab \
--dataset dataset_to_optimize_on \  # men, simlex or simverb
--mode limit \
--metric pearson \
--limit 10  # number of dimensions to sample