scoreBySigTerms.py

def scoreBySigTerms (test_docs, sig_words, n_sig_terns):
    scoresByLabel = {}  
    for label in [0,1]:
        useSigWords = sig_words[label][0:n_sig_terns]  
        vectorizer = CountVectorizer(analyzer=lambda x: x, min_df=1, vocabulary=useSigWords)
        test_doc_vectors = vectorizer.transform(test_docs)
        a = np.sum(test_doc_vectors,axis=1)
        b = []
        for i in range(len(a)):
            b.append(a[i,0])
        scoresByLabel[label] = b
 
    predicted_labels, probabilities = [], np.zeros((len(test_docs),2))
    for i, testDoc in enumerate(test_docs):
        docScores = np.array([scoresByLabel[label][i] for label in [0,1]])
        probabilities[i] = docScores / (np.sum(predList) + 1.0e-16)      # Linear normalization to 1. Can do softmax...
        predicted_label = np.argmax(probabilities[i])
        predicted_labels.append(predicted_label)