def test_predict():
global model, X_tr, Y_tr, word_to_ix, tag_to_ix
torch.manual_seed(711);
best_tags = model.predict(bilstm.prepare_sequence(X_tr[5], word_to_ix))
eq_(best_tags[0:5],['NOUN', 'ADJ', 'CONJ', 'ADV', 'ADJ'])
best_tags = model.predict(bilstm.prepare_sequence(X_tr[0], word_to_ix))
eq_(best_tags[0:5],['X', 'NUM', 'INTJ', 'PART', 'AUX'])
def test_neg_log_likelihood():
global model, X_tr, Y_tr, word_to_ix, tag_to_ix
torch.manual_seed(711);
lstm_feats = model.forward(bilstm.prepare_sequence(X_tr[5], word_to_ix))
loss = model.neg_log_likelihood(lstm_feats, bilstm.prepare_sequence(Y_tr[5], tag_to_ix))
assert_almost_equal(loss.item(),50.326389, places=4)
lstm_feats = model.forward(bilstm.prepare_sequence(X_tr[0], word_to_ix))
loss = model.neg_log_likelihood(lstm_feats, bilstm.prepare_sequence(Y_tr[0], tag_to_ix))
assert_almost_equal(loss.item(),102.239616, places=4)
def test_forward_alg():
global model, X_tr, word_to_ix
torch.manual_seed(711);
lstm_feats = model.forward(bilstm.prepare_sequence(X_tr[0], word_to_ix))
alpha = model.forward_alg(lstm_feats)
assert_almost_equal(alpha.item(), 104.916992, places=4)
lstm_feats = model.forward(bilstm.prepare_sequence(X_tr[1], word_to_ix))
alpha = model.forward_alg(lstm_feats)
assert_almost_equal(alpha.item(), 65.290924, places=4)
def test_score_sentence():
global model, X_tr, Y_tr, word_to_ix, tag_to_ix
torch.manual_seed(711);
lstm_feats = model.forward(bilstm.prepare_sequence(X_tr[0], word_to_ix))
score = model.score_sentence(lstm_feats, bilstm.prepare_sequence(Y_tr[0], tag_to_ix))
print(tag_to_ix)
assert_almost_equal(score.item(), 2.659940, places=4)
lstm_feats = model.forward(bilstm.prepare_sequence(X_tr[1], word_to_ix))
score = model.score_sentence(lstm_feats, bilstm.prepare_sequence(Y_tr[1], tag_to_ix))
assert_almost_equal(score.item(), -2.397999, places=4)
def apply_model(model,
outfilename,
word_to_ix,
all_tags=None,
trainfile=TRAIN_FILE,
testfile=DEV_FILE):
"""
applies the model on the data and writes the best sequence of tags to the outfile
"""
if all_tags is None:
all_tags = set()
# this is slow
for i, (words,
tags) in enumerate(preproc.conll_seq_generator(trainfile)):
for tag in tags:
all_tags.add(tag)
with open(outfilename, 'w') as outfile:
for words, _ in preproc.conll_seq_generator(testfile):
seq_words = bilstm.prepare_sequence(words, word_to_ix)
pred_tags = model.predict(seq_words)
for i, tag in enumerate(pred_tags):
outfile.write(tag + '\n')
outfile.write('\n')
def test_dlmodel_forward():
global model, X_tr, word_to_ix
torch.manual_seed(711)
lstm_feats = model(bilstm.prepare_sequence(X_tr[0], word_to_ix))
assert_almost_equal(lstm_feats[0].data.numpy()[0], 0.044214, places=4)
assert_almost_equal(lstm_feats[0].data.numpy()[1], -0.071416, places=4)
assert_almost_equal(lstm_feats[0].data.numpy()[2], -0.125279, places=4)
