def check_accuracy(mnist):
with tf.Graph().as_default() as grh:
x = tf.placeholder(tf.float32, [None, mnsit_forward.INPUT_NODE])
y_ = tf.placeholder(tf.float32, [None, mnsit_forward.OUTPUT_NODE])
y = mnsit_forward.forward(x, None)
ema = tf.train.ExponentialMovingAverage(
mnsit_backward.MOVING_AVG_DECAY)
ema_restore = ema.variables_to_restore()
saver = tf.train.Saver(ema_restore)
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
while 1:
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state(
mnsit_backward.MODEL_SAVE_PATH)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = ckpt.model_checkpoint_path.split(
'/')[-1].split('-')[-1]
accuracy_rate = sess.run(accuracy,
feed_dict={
x: mnist.test.images,
y_: mnist.test.labels
})
fmt = 'After {:s} steps, test accuracy rate is {:.09f}'
print(fmt.format(global_step, accuracy_rate))
time.sleep(TEST_INTERVAL_SECS)
def check_accuracy():
with tf.Graph().as_default() as grh:
x = tf.placeholder(tf.float32, [None, mnsit_forward.INPUT_NODE])
y_ = tf.placeholder(tf.float32, [None, mnsit_forward.OUTPUT_NODE])
y = mnsit_forward.forward(x, None)
ema = tf.train.ExponentialMovingAverage(
mnsit_backward.MOVING_AVG_DECAY)
ema_restore = ema.variables_to_restore()
saver = tf.train.Saver(ema_restore)
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
img_batch, label_batch = mnist_generate_recode.get_tf_record(
TEST_NUM, is_train=False)
while 1:
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state(
mnsit_backward.MODEL_SAVE_PATH)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = ckpt.model_checkpoint_path.split(
'/')[-1].split('-')[-1]
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess,
coord=coord)
xs, ys = sess.run([img_batch, label_batch])
accuracy_rate = sess.run(accuracy,
feed_dict={
x: xs,
y_: ys
})
fmt = 'After {:s} steps, test accuracy rate is {:.09f}'
print(fmt.format(global_step, accuracy_rate))
coord.request_stop()
coord.join(threads)
time.sleep(TEST_INTERVAL_SECS)
def restore_model(pic_array):
# 重现计算图
with tf.Graph().as_default() as gph:
# 只需要对输入占位
x = tf.placeholder(tf.float32, [None, mnsit_forward.INPUT_NODE])
y = mnsit_forward.forward(x, None)
# y的最大值对应的索引号，就是预测的数字的值
pre_value = tf.argmax(y, 1)
variable_avg = tf.train.ExponentialMovingAverage(
mnsit_backward.MOVING_AVG_DECAY)
variable_to_restore = variable_avg.variables_to_restore()
saver = tf.train.Saver(variable_to_restore)
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state(
mnsit_backward.MODEL_SAVE_PATH)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
# 预测操作
pre_value = sess.run(pre_value, feed_dict={x: pic_array})
return pre_value
def backward(mnsit):
x = tf.placeholder(tf.float32, [None, mnsit_forward.INPUT_NODE])
y_ = tf.placeholder(tf.float32, [None, mnsit_forward.OUTPUT_NODE])
y = mnsit_forward.forward(x, FLAGS.regularizer)
global_step = tf.Variable(0, trainable=False)
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
logits=y, labels=tf.argmax(y_, 1))
cross_entropy_mean = tf.reduce_mean(cross_entropy)
loss = cross_entropy_mean + tf.add_n(tf.get_collection('losses'))
learning_rate = tf.train.exponential_decay(FLAGS.learning_rate_base,
global_step,
mnsit.train.num_examples /
FLAGS.batch_size,
FLAGS.learning_rate_decay,
staircase=True)
train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(
loss, global_step=global_step)
ema = tf.train.ExponentialMovingAverage(FLAGS.moving_avg_decay,
global_step)
ema_op = ema.apply(tf.trainable_variables())
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
with tf.control_dependencies([train_step, ema_op]):
train_op = tf.no_op(name='train')
saver = tf.train.Saver()
with tf.Session() as sess:
init_op = tf.global_variables_initializer()
sess.run(init_op)
ckpt = tf.train.get_checkpoint_state(FLAGS.model_save_path)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
for i in range(FLAGS.steps):
xs, ys = mnsit.train.next_batch(FLAGS.batch_size)
feed_dict = {x: xs, y_: ys}
_, loss_value, learning_rate_val, step = sess.run(
[train_op, loss, learning_rate, global_step],
feed_dict=feed_dict)
if 0 == i % 100:
fmt = 'After {:05d} steps, loss is {:.09f}, learning rate is {:.09f}'
print(fmt.format(step, loss_value, learning_rate_val))
saver.save(sess,
os.path.join(FLAGS.model_save_path,
FLAGS.model_name),
global_step=global_step)
accuracy_rate = sess.run(accuracy,
feed_dict={
x: mnsit.test.images,
y_: mnsit.test.labels
})
fmt = 'After {:d} steps, test accuracy rate is {:.09f}'
print(fmt.format(step, accuracy_rate))