valid_predictions = []
n_batches = valid_X.shape[0] // VALID_BATCH_SIZE
for i in range(n_batches):
start = i * VALID_BATCH_SIZE
end = start + VALID_BATCH_SIZE
pred, valid_cost = sess.run([valid, loss],
feed_dict={
x: valid_X[start:end],
t: valid_y[start:end]
})
valid_predictions.extend(pred)
valid_costs.append(valid_cost)
# f1_score = f1_score(np.argmax(valid_y, 1).astype('int32'), valid_predictions, average='macro')
accuracy = accuracy_score(
np.argmax(valid_y, 1).astype('int32'), valid_predictions)
if epoch % 5 == 0:
print(
'EPOCH: {epoch}, Training cost: {train_cost}, Validation cost: {valid_cost}, Validation Accuracy: {accuracy} '
.format(epoch=epoch,
train_cost=np.mean(train_costs),
valid_cost=np.mean(valid_costs),
accuracy=accuracy))
if early_stopping.check(np.mean(valid_costs)):
break
print("save model...")
saver = tf.train.Saver()
saver.save(sess, SAVE_PATH, global_step=epoch)
n_batches = train_X.shape[0] // hp.BATCH_SIZE
# train
train_costs = []
train_acc = []
for i in tqdm(range(n_batches)):
# print(i)
start = i * hp.BATCH_SIZE
end = start + hp.BATCH_SIZE
_, _acc, _loss = sess.run([train, accuracy, loss], feed_dict={x: train_X[start:end], y: ty[start:end], is_training: True})
train_costs.append(_loss)
train_acc.append(_acc)
print('EPOCH: {epoch}, Training avg cost: {train_cost}, Training avg acc: {train_acc} '
.format(epoch=epoch, train_cost=np.mean(train_costs), train_acc=np.mean(train_acc)))
if early_stopping.check(np.mean(train_costs)):
print("epoch:",epoch)
break
# valid
valid_costs = []
valid_predictions = []
valid_scores = []
n_batches = X_test.shape[0] // hp.VALID_BATCH_SIZE
print("test n_batch", n_batches)
for i in range(n_batches):
start = i * hp.BATCH_SIZE
end = start + hp.BATCH_SIZE
pred, valid_cost, score = sess.run([valid, loss, pre_score], feed_dict={x: X_test[start:end], y: ey[start:end], is_training: False})
valid_predictions.extend(pred)