def train_and_model(model):
# TODO LOOK ME this make show all tensor belong cpu or gpu
# with tf.Session(config=tf.ConfigProto(log_device_placement=True)) as sess, tf.device("/CPU:0"):
with tf.Session() as sess:
# checkpoint val
saver = tf.train.Saver()
# tensorboard
train_writer = tf.summary.FileWriter(FLAGS.dir_train_tensorboard,
sess.graph)
test_writer = tf.summary.FileWriter(FLAGS.dir_test_tensorboard)
# train step
print("Train Start...")
train_batch_config = Batch.Config()
train_batch = Batch.Batch(train_batch_config)
sess.run(model["init_op"])
for step in range(FLAGS.max_train_step + 1):
key_list = [
Batch.INPUT_DATA, Batch.OUTPUT_LABEL, Batch.OUTPUT_DATA
]
data = train_batch.next_batch(FLAGS.batch_size, key_list)
feed_dict = {
model["X"]: data[Batch.INPUT_DATA],
model["Y"]: data[Batch.OUTPUT_DATA],
model["Y_label"]: data[Batch.OUTPUT_LABEL]
}
sess.run(model["train_op"], feed_dict)
# print log
if step % FLAGS.print_log_step_size == 0:
summary_train, _acc, _cost = sess.run(
[model["summary"], model["batch_acc"], model["cost"]],
feed_dict=feed_dict)
print(datetime.datetime.utcnow(), "train step: %d" % step,
"batch_acc:", _acc, "cost:", _cost)
# checkpoint
if step % FLAGS.checkpoint_step_size == 0:
saver.save(sess, FLAGS.dir_train_checkpoint, global_step=step)
# summary tensorboard
if step % FLAGS.summary_step_size:
train_writer.add_summary(summary=summary_train,
global_step=step)
# test step
print("Test Start...")
test_batch_config = Batch.Config()
test_batch = Batch.Batch(test_batch_config)
total_acc = 0.
for step in range(FLAGS.max_test_step + 1):
key_list = [
Batch.INPUT_DATA, Batch.OUTPUT_LABEL, Batch.OUTPUT_DATA
]
data = test_batch.next_batch(FLAGS.batch_size, key_list)
feed_dict = {
model["X"]: data[Batch.INPUT_DATA],
model["Y"]: data[Batch.OUTPUT_DATA],
model["Y_label"]: data[Batch.OUTPUT_LABEL]
}
# print("input:", data[Batch.INPUT_DATA])
# print("output:", data[Batch.OUTPUT_DATA])
# print("label:", data[Batch.OUTPUT_LABEL])
summary_test, _acc = sess.run(
[model["summary"], model["batch_acc"]], feed_dict=feed_dict)
print(datetime.datetime.utcnow(), "test step: %d" % step,
"batch_acc: ", _acc)
total_acc += _acc
if step % FLAGS.print_log_step_size == 0:
summary_test, _acc = sess.run(
[model["summary"], model["batch_acc"]],
feed_dict=feed_dict)
# print(datetime.datetime.utcnow(), "test step: %d" % step
# , "batch_acc: ", _acc)
test_writer.add_summary(summary=summary_test, global_step=step)
print("test complete: total acc =",
total_acc / (FLAGS.max_test_step + 1))
return
