for epoch in range(epochs):
t0 = time.time()
loss_train = 0
loss_test = 0
acc_train = 0
acc_test = 0
#print("epoch: {}\t".format(epoch), end="")
# training
n_batch = len(X_train) // batch_size
#for _ in range(n_batch):
for _ in tqdm(range(n_batch), total=n_batch):
X_batch, y_batch = utils.get_batch_data(X_train_doc, y_train,
batch_size)
X_batch_emb = embedding_bert.get_batch_emb(X_batch, doc_len,
sen_len, tokenizer,
estimator)
X_batch_emb = X_batch_emb[:, :, 0, :]
X_batch_seq, _ = embedding_bert.get_batch_seq(X_batch,
doc_len,
sen_len,
tokenizer,
tol=2)
_, loss, acc, atten_w = sess.run(
[optimizer, cost, accuracy, soft_atten_weights],
feed_dict={
X_emb: X_batch_emb,
y: y_batch,
seq_length: X_batch_seq,
x = utils.token_sens(xx_clean, sentence_size = 30, word_to_idx = word_to_idx) x.shape # In[ ]: # In[16]: a, b = utils.get_batch_data(X_train, y_train, batch_size = 5) a_clean = utils.replace_contr(a, contractions) a_clean.shape, b.shape # In[17]: a, a_clean, b # In[ ]:
# In[ ]:
# ## Predict section
# In[30]:
with tf.Session() as sess:
saver.restore(sess, savepath)
X_batch, y_batch = utils.get_batch_data(X_test, y_test, 10)
X_batch_emb = embedding_bert.get_batch_emb(X_batch, doc_len, sen_len, tokenizer, estimator)
X_batch_emb = np.squeeze(X_batch_emb, axis = 1)
_, X_batch_seq = embedding_bert.get_batch_seq(X_batch, doc_len, sen_len, tokenizer, tol = 2)
atten_w, y_proba = sess.run([soft_atten_weights, Y_proba], feed_dict={X_emb: X_batch_emb,
seq_length: X_batch_seq, is_training:False})
y_pred = np.argmax(y_proba, axis = 1)
# In[31]:
X_batch, y_batch, y_pred