import data_reader as reader

import tensorflow as tf

import datetime

import model

import config

import os

TRAIN_PATH = config.load_train_path()

LOG_DIR = 'log'

MODEL_DIR = 'model_data'

N_CLASS = 43

IMAGE_WIDTH = 50

IMAGE_HEIGHT = 50

BATCH_SIZE = 64

VALIDATE_RATE = 0.01

EPOCH = 40

learning_rate = 1e-3

def load_data():

return len(t_images), t_image, t_label, v_image, v_label

if __name__ == '__main__':

# load train and validate data

train_size, train_image, train_label, validate_image, validate_label = load_data()

# ready something about training

train_logits = model.inference(train_image, True)

train_loss = model.calculate_loss(train_logits, train_label)

train_accuracy = model.calculate_accuracy(train_logits, train_label)

#validate_accuracy = model.calculate_accuracy_width_images(validate_image, validate_label, True)

learning_rate = tf.train.exponential_decay(learning_rate, 1000, 1, 0.99, staircase=True, name=None)

train_step = model.get_train_step(train_loss, learning_rate)

# initialize summary

summary_merge = tf.summary.merge_all()

graph = tf.get_default_graph()

train_writer = tf.summary.FileWriter(LOG_DIR, graph)

# validate_writer = tf.summary.FileWriter(LOG_DIR, graph)

# initialize session and saver

sess = tf.InteractiveSession()

sess.run(tf.global_variables_initializer())

saver = tf.train.Saver()

checkpoint = tf.train.get_checkpoint_state(MODEL_DIR)

if checkpoint and checkpoint.model_checkpoint_path:

saver.restore(sess, checkpoint.model_checkpoint_path)

print('Load last model params successfully.')

# initialize coord

coord = tf.train.Coordinator()

threads = tf.train.start_queue_runners(sess, coord)

# start training

try:

print('Start training...')

max_iterator = int(train_size * EPOCH / BATCH_SIZE)

for step in range(1, max_iterator + 1):

_, loss, train_acc, summary_str = sess.run([train_step, train_loss, train_accuracy, summary_merge])

# _, loss, accuracy, summary_str, a, b = sess.run([train_step, train_loss, train_accuracy, summary_merge, train_image, train_label])

if step % 10 == 0 or step == max_iterator:

time = str(datetime.datetime.now())

epoch = int(step * BATCH_SIZE / train_size)

print('Time %s, Epoch %d, Step: %d, Loss %s, Accuracy %s' % (time, epoch, step, loss, train_acc))

if step % 50 == 0 or step == max_iterator:

train_writer.add_summary(summary_str, step)

train_writer.flush()

pass

if step % 100 == 0 or step == max_iterator:

saver.save(sess, os.path.join(MODEL_DIR, 'model'), step)

except tf.errors.OutOfRangeError as e:

print('Error %s' % e)

finally:

coord.request_stop()

coord.join(threads)