Model interpretation using vectorized logistic regression (word list coefficients provided

model_interpretation.py

import numpy as np
from sklearn.feature_extraction.text import CountVectorizer

all_data = pd.read_table('alldata.tsv')
dict = {pair[1]['token']:pair[1]['coef'] for pair in all_vocab[['token','coef']].iterrows()}

vectorizer = CountVectorizer(vocabulary=all_vocab['token'].values, max_df=0.3, ngram_range=(1, 2), min_df=20)
cur_vocab = vectorizer.get_feature_names()

def get_tokens(review):
    a = vectorizer.transform([review])
    nz = np.nonzero(a)[1]
    return all_vocab.iloc[nz][['coef','token']]

review = all_data['review'].values[5]
print(review)
o = get_tokens(review)
sns.barplot(x="coef", y="token", data=o);