def prepare():
if not os.path.exists(args.save_dir):
os.makedirs(args.save_dir)
print('Vocab processing ...')
q1, q2, y = get_q2q_label(args.merged_files)
start_time = time.time()
vocab_processor = tc.learn.preprocessing.VocabularyProcessor(
max_document_length=args.max_q_len,
min_frequency=5,
tokenizer_fn=chinese_tokenizer)
q1_pad = np.array(list(vocab_processor.fit_transform(q1)))
q2_pad = np.array(list(vocab_processor.fit_transform(q2)))
del q1, q1_pad, q2, q2_pad, y
print('Vocab size: {}'.format(len(vocab_processor.vocabulary_)))
vocab_processor.save(os.path.join(args.save_dir, "vocab"))
# split
# split_data(args.merged_files, os.path.join(args.save_dir, "vocab"), args.pkl_files, mask=True)
# no mask
split_data(args.merged_files, os.path.join(args.save_dir, "vocab"),
args.pkl_files)
time_dif = get_time_dif(start_time)
print('Vocab processing time usage:', time_dif)
def predict():
print('Loading test data ...')
start_time = time.time()
q1_test, q2_test, y_test = get_q2q_label(args.test_data_files)
vocab_path = os.path.join(args.save_dir, 'vocab')
vocab_processor = tc.learn.preprocessing.VocabularyProcessor.restore(
vocab_path)
# MVLSTM model init
model = MVLSTM(sequence_length=args.max_q_len,
num_classes=args.num_classes,
embedding_dim=args.embedding_dim,
vocab_size=len(vocab_processor.vocabulary_),
max_length=args.max_q_len,
hidden_dim=args.hidden_size,
learning_rate=args.learning_rate)
q1_pad = np.array(list(vocab_processor.transform(q1_test)))
q2_pad = np.array(list(vocab_processor.transform(q2_test)))
session = tf.Session()
session.run(tf.global_variables_initializer())
saver = tf.train.Saver()
saver.restore(session, save_path=save_path)
print('Testing ...')
loss_test, acc_test = evaluate(q1_pad,
q2_pad,
y_test,
session,
model=model)
print('Test loss:{0:>6.2}, Test acc:{1:7.2%}'.format(loss_test, acc_test))
test_batches = batch_iter_per_epoch(q1_pad, q2_pad, y_test, shuffle=False)
all_predictions = []
all_predict_prob = []
count = 0 # concatenate第一次不能为空，需要一个判断来赋all_predict_prob
for q1_test_batch, q2_test_batch, y_test_batch in test_batches:
batch_predictions, batch_predict_probs = session.run(
[model.y_pred, model.probs],
feed_dict={
model.input_q1: q1_test_batch,
model.input_q2: q2_test_batch,
model.dropout_keep_prob: 1.0
})
all_predictions = np.concatenate([all_predictions, batch_predictions])
if count == 0:
all_predict_prob = batch_predict_probs
else:
all_predict_prob = np.concatenate(
[all_predict_prob, batch_predict_probs])
count = 1
y_test = [float(temp) for temp in y_test]
# Evaluation indices
print('Precision, Recall, F1-Score ...')
print(
metrics.classification_report(y_test,
all_predictions,
target_names=['not match', 'match']))
# Confusion Matrix
print('Confusion Matrix ...')
print(metrics.confusion_matrix(y_test, all_predictions))
# Write probability to csv
out_dir = os.path.join(args.save_dir, 'predict_prob_csv')
print('Saving evaluation to {0}'.format(out_dir))
with open(out_dir, 'w') as f:
csv.writer(f).writerows(all_predict_prob)
time_dif = get_time_dif(start_time)
print('Time usage:', time_dif)