rare_we

1. requirements Option 1: Docker Dockerfile found in /Docker/Dockerfile Bash ./start source activate rare_we_clean

   Option 2: Conda Requirements can be found in /Docker/environment_rare_we_clean.yml

   We train context2vec models from https://www.github.com/melamud/context2vec/

   Training corpus and the skipgram models can be found from https://github.com/minimalparts/nonce2vec

2. chimeras/nonce/crw evaluation

   Usage:python eval/eval_script.py [-h] [--f MODEL_PARAM_FILE] [--m MODEL_TYPE] [--w WEIGHTS [WEIGHTS ...]] [--d DATA] [--g GPU] [--ws W2SALIENCE_F] [--n_result N_RESULT]

   WEIGHTS (integers in WEIGHT_DICT):

   WEIGHT_DICT={0:False,1:TOP_MUTUAL_SIM,2:LDA,3:INVERSE_S_FREQ,4:INVERSE_W_FREQ,5:TOP_CLUSTER_DENSITY, 6:SUBSTITUTE_PROB}

    For our experiments, choose 0 or 3 for skipgram; choose 0 for the substitutes-based method
   

   MODEL_TYPE:

    skipgram (context skipgram input vector without stop words)
    
    context2vec (word embedding in context2vec space)
    
    context2vec-skipgram(context2vec substitutes in skipgram space) (the substitute method)
    
    context2vec-skipgram?skipgram (context2vec substitutes in skipgram space plus skipgram context words)