with tf.Session(graph=lstm_graph) as sess:
tf.global_variables_initializer().run()
train_score = []
validation_score = []
for epoch in range(config.max_epoch):
print(
"epoch ****************************************************************************",
epoch)
# lr_decay = config.learing_rate_decay ** max(float(epoch + 1 - config.init_epoch), 0.0)
# current_lr = config.init_learning_rate * lr_decay
# print("learning_rate............................", current_lr)
# batch_index = 0
for batch_X, batch_Y in utils.generate_one_epoch(
train_data, train_labels, config.batch_size,
config.num_steps):
train_data_feed = {
inputs: batch_X,
targets: batch_Y,
learning_rate: config.learning_rate,
keep_prob: config.keep_prob
}
train_loss, _ = sess.run([loss, optimizer], train_data_feed)
# batch_index += 1
# print accuracy of train model for each 50 batch
# if batch_index % 1000 == 0:
# print("train_loss: ", train_loss)
# output training score and validation score for each epoch
],
[
validation_data, validation_labels, "validation",
validation_reader
]]:
data = each[0]
labels = each[1]
which_data = each[2]
reader = each[3]
final_predict = np.array([])
total_loss = 0
num_loss = 0
print(
"Data.....................................................................",
which_data)
for batch_X, batch_Y in utils.generate_one_epoch(
data, labels, config.batch_size, shuffle=False):
data_feed = {
inputs: batch_X,
targets: batch_Y,
keep_prob: 1.0
}
loss_p, predict = sess.run([loss, prediction], data_feed)
total_loss += loss_p
num_loss += 1
# predict = predict[0]
final_predict = np.append(final_predict, predict)
# print "losssssssssssssssssssssssssssssssssssssssssssssssssssssssssss", which_data, total_loss/num_loss
final_predict = np.array(final_predict).reshape(
len(final_predict), 1)
stages = [utils.get_state(x) for x in final_predict]
non_eruption_stage = [