def fit(self, X, Y, valid_X = None, valid_Y = None):
self.sess.run(tf.initialize_all_variables())
summary_train = Summary('train_loss')
summary_valid = Summary('valid_loss')
summary_valid_acc = Summary('valid_acc')
summary_train.start()
summary_valid.start()
summary_valid_acc.start()
for iter in range(NUM_ITER):
total_loss = 0.0
for i in range(0,len(X), BATCH_SIZE):
tail = min(i+BATCH_SIZE, len(X))
batch_xs = X[i:tail,:]
batch_ys = Y[i:tail,:]
_, loss = self.sess.run([self.train_step, self.cross_entropy], feed_dict={self.x: batch_xs, self.y: batch_ys})
total_loss += loss*(tail-i+1)
total_loss /= float(len(X))
print('Avg training loss on {}th epoch: {}'.format(iter, total_loss))
summary_train.log(total_loss)
if valid_X is not None and valid_Y is not None and iter % 100 ==0:
valid_loss, valid_acc = self.eva(valid_X, valid_Y)
summary_valid.log(valid_loss)
summary_valid_acc.log(valid_acc)
