virtualenv:
source /home/dfried/.virtualenvs/ufldl/bin/activate
files you can actually run:
==========================
admm_trainer.py - used to train ADMM using distance in wordnet
- requires access to a dump of wordnet pairwise similarities, one is available at /cl/nldata/books_google_ngrams_eng/wordnet_similarities_max.npy
- by default will copy this sim file to a shared memory file in TMP in case multiple procs are running
relational_admm_trainer.py - used to train ADMM using wordnet relations
evaluation/ - contains interfaces to semeval, grefenstette, and wordsim eval sets, plus one script evaluation.py that will run all
usage:
python evaluation/evaluation.py MODEL_DIRECTORIES --plot_interval PLOT_INTEVAL --limit LIMIT
- caches stats in .pkl files in each model directory, reads these if necessary
- flag --no_new will only read existing cached stats
plot_admm_stats.py - plots cached ADMM stats from stats.pkl files in model directories
usage:
python plot_admm_stats.py MODEL_DIRECTORIES --limit LIMIT
some dependent files
================
models.py - contains all Theano models
policies.py - gradient descent schedules
corpus_creator.py - build an hd5 file from Berkeley format corpus
dump_wordnet_similarities.py - used to dump the pairwise similarities needed by admm_trainer.py
ngrams.py - used to read an hd5 file created by corpus_creator.py