# words.append(a+"_"+n+"_l2")
#runs pca on the smaller matrix we construct....
####################
B = []
for word in words:
if word in vecs:
B.append(vecs[word])
elif "avg" in word:
tokens = word.split('_')
avg = (vecs[tokens[0]] + vecs[tokens[1]]) / 2.0
B.append(avg)
elif "l2" in word:
tokens = word.split('_')
B.append(tb.local_knn_pred(vecs, tokens[0], tokens[1], "norm"))
elif "cos" in word:
tokens = word.split('_')
B.append(tb.local_knn_pred(vecs, tokens[0], tokens[1], "cos"))
elif "svr" in word:
tokens = word.split('_')
B.append(tnn.pred(vecs[tokens[0]], vecs[tokens[1]]))
B = np.matrix(B)
# add other non-plotted bigrams to the matrix before running PCA
B = np.concatenate((B, wordVecsMatrix[1:500, :]))
print "Running PCA"
pca.fit(B)
B = pca.transform(B)
#######################
# words.append(a+"_"+n+"_cos")
# words.append(a+"_"+n+"_l2")
#runs pca on the smaller matrix we construct....
####################
B = []
for word in words:
if word in vecs:
B.append(vecs[word])
elif "avg" in word:
tokens = word.split('_')
avg = (vecs[tokens[0]] + vecs[tokens[1]]) / 2.0
B.append(avg)
elif "l2" in word:
tokens = word.split('_')
B.append(tb.local_knn_pred(vecs, tokens[0], tokens[1], "norm"))
elif "cos" in word:
tokens = word.split('_')
B.append(tb.local_knn_pred(vecs, tokens[0], tokens[1], "cos"))
elif "svr" in word:
tokens = word.split('_')
B.append(tnn.pred(vecs[tokens[0]], vecs[tokens[1]]))
B = np.matrix(B)
# add other non-plotted bigrams to the matrix before running PCA
B = np.concatenate((B, wordVecsMatrix[1:500, :]))
print "Running PCA"
pca.fit(B)
B = pca.transform(B)
#######################
def predict(word1, word2):
return tb.local_knn_pred(word_vectors, word1, word2, "norm")
def predict (word1, word2): return tb.local_knn_pred(word_vectors, word1, word2, "norm")
