def main():
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
zipped_data = zip(train_data, train_label)
batches = data_helper.gen_batch(list(zipped_data), batch_size,
max_steps)
for i in range(max_steps):
batch = next(batches)
images_batch, labels_batch = zip(*batch)
feed_dict = {
images_placeholder: images_batch,
labels_placeholder: labels_batch
}
# Periodically print out the model's current accuracy
if i % 100 == 0:
train_accuracy = sess.run(accuracy, feed_dict=feed_dict)
print('Step {:d}, training accuracy {:g}'.format(
i, train_accuracy))
sess.run([train_step, loss], feed_dict=feed_dict)
test_accuracy = sess.run(accuracy,
feed_dict={
images_placeholder: test_data,
labels_placeholder: test_label
})
print('Test accuracy {:g}'.format(test_accuracy))
def main():
labelholder = tf.placeholder(tf.int64, shape=[None], name='labels')
imageholder = tf.placeholder(tf.float32, shape=[None, 3072], name='images')
layers = construct(imageholder, 3072, 10, 100)
with tf.name_scope('En'):
L = tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(
logits=layers, labels=labelholder, name='cross_entropy'))
tf.summary.scalar('en', L)
currstep = tf.train.GradientDescentOptimizer(0.001).minimize(
L, global_step=tf.Variable(0, name='global_step', trainable=False))
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
zdata = zip(data_helper.load_data()['images_train'],
data_helper.load_data()['labels_train'])
batches = data_helper.gen_batch(list(zdata), 400, totalstep)
for i in range(totalstep):
batch = next(batches)
images_batch, labels_batch = zip(*batch)
if i % 100 == 0:
with tf.name_scope('a'):
accuracy = tf.reduce_mean(
tf.cast(tf.equal(tf.argmax(layers, 1), labelholder),
tf.float32))
tf.summary.scalar('ta', accuracy)
print('training accuracy for {:d} step is {:g}'.format(
i,
sess.run(accuracy,
feed_dict={
imageholder: images_batch,
labelholder: labels_batch
})))
sess.run([currstep, L],
feed_dict={
imageholder: images_batch,
labelholder: labels_batch
})
print('Accuracy is ' + format(
sess.run(accuracy,
feed_dict={
imageholder: data_helper.load_data()['images_test'],
labelholder: data_helper.load_data()['labels_test']
})))
def train(self, x, y):
if not os.path.exists(FLAGS.model_path):
os.makedirs(FLAGS.model_path)
with tf.device(FLAGS.device_name):
train_op = tf.train.AdamOptimizer(self.learning_rate).minimize(
loss=self.loss, name='optimaize')
saver = tf.train.Saver()
init_op = tf.global_variables_initializer()
with tf.Session() as sess:
print("模型初始化中……")
sess.run(init_op)
start_epoch = 0
checkpoint = tf.train.latest_checkpoint(FLAGS.model_path)
if checkpoint:
saver.restore(sess, checkpoint)
print("成功载入模型{0}".format(checkpoint))
start_epoch += int(checkpoint.split('-')[-1])
print("开始训练……")
total_loss = 0
try:
for epoch in range(start_epoch, self.epoches):
n_chunk = len(x) // self.batch_size
# print(n_chunk)
ave_loss = total_loss / n_chunk
total_loss = 0
for batch in range(n_chunk):
x_batch, y_batch = gen_batch(x, y, self.batch_size,
batch)
feed = {self.inputs: x_batch, self.targets: y_batch}
l, acc, _ = sess.run(
[self.loss, self.accuracy, train_op],
feed_dict=feed)
total_loss += l
if batch % self.every_epoch_to_print == 0:
print(
'Epoch:%d, last epoch loss ave:%.5f batch:%d, current epoch loss:%.5f, acc:%.3f'
% (epoch, ave_loss, batch, l, acc))
if epoch % self.every_epoch_to_save == 0:
print("保存模型……")
saver.save(sess,
os.path.join(FLAGS.model_path,
FLAGS.model_prefix),
global_step=epoch)
except KeyboardInterrupt:
print("手动终止训练，保存模型……")
saver.save(sess,
os.path.join(FLAGS.model_path, FLAGS.model_prefix),
global_step=epoch)
# Define saver to save model state at checkpoints
saver = tf.train.Saver()
# -----------------------------------------------------------------------------
# Run the TensorFlow graph
# -----------------------------------------------------------------------------
with tf.Session() as sess:
# Initialize variables and create summary-writer
sess.run(tf.global_variables_initializer())
summary_writer = tf.summary.FileWriter(logdir, sess.graph)
# Generate input data batches
zipped_data = zip(data_sets['images_train'], data_sets['labels_train'])
batches = data_helper.gen_batch(list(zipped_data), FLAGS.batch_size,
FLAGS.max_steps)
for i in range(FLAGS.max_steps):
# Get next input data batch
batch = next(batches)
images_batch, labels_batch = zip(*batch)
feed_dict = {
images_placeholder: images_batch,
labels_placeholder: labels_batch
}
# Periodically print out the model's current accuracy
if i % 100 == 0:
train_accuracy = sess.run(accuracy, feed_dict=feed_dict)
print('Step {:d}, training accuracy {:g}'.format(
l2_reg_lambda=FLAGS.L2_reg_lambda)
with tf.device('/gpu:0'):
train_step = tf.train.GradientDescentOptimizer(
learning_rate).minimize(loss=cnn.loss, global_step=global_step)
# train_step = tf.train.AdamOptimizer(1e-3).minimize(loss=cnn.loss, global_step=global_step)
saver = tf.train.Saver()
if os.path.exists(FLAGS.model_save_path + 'checkpoint'):
saver.restore(sess, tf.train.latest_checkpoint(FLAGS.model_save_path))
learning_rate = tf.constant(0.001, dtype=tf.float32)
else:
sess.run(tf.global_variables_initializer())
last = datetime.datetime.now()
for i in range(FLAGS.training_ite):
x_batch, y_batch = data_helper.gen_batch(x_train, y_train, i,
FLAGS.batch_size)
feed_dic = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
cnn.dropout_keep_prob: FLAGS.dropout
}
_, loss, acc = sess.run([train_step, cnn.loss, cnn.accuracy],
feed_dict=feed_dic)
if (i % 100) == 0:
now = datetime.datetime.now()
print('loss:{},acc:{}---time:{}'.format(loss, acc, now - last))
last = now
if (i % 1000) == 0:
feed_dic = {
cnn.input_x: x_dev,
# Weight decay Weight_decay = 0.0001 l2 = tf.add_n([tf.nn.l2_loss(var) for var in tf.trainable_variables()]) train_step = tf.train.RMSPropOptimizer(learning_rate=1e-4).minimize(cross_entropy + l2 * Weight_decay, global_step=global_step) #prediction correct_prediction = tf.equal(tf.argmax(output,1),tf.argmax(y_,1) ) accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) sess = tf.InteractiveSession() sess.run(tf.global_variables_initializer()) #start session zipped_data = zip(data_sets['images_train'], data_sets['labels_train']) batches = data_helper.gen_batch(list(zipped_data), 128, 64124) zipped_data_test = zip(data_sets['images_test'], data_sets['labels_test']) batches_test = data_helper.gen_batch(list(zipped_data_test), 128, 64124) #pp.pprint(zip(*next(batches_test))[1]) epoch_num = 0 max_train_accuracy = 0 max_test_accuracy = 0 min_loss = 99999 for i in range(64124): #391 interation_num * 164 epoch_num batch = next(batches) images_batch, labels_batch = zip(*batch) batch_test = next(batches_test)
