def train_model(model_name):
"""
train a model and save it to folder '/trained_model/model_name'
"""
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data", one_hot=True)
model = CNNModel(image_size=[28, 28], char_number=10, channel=1)
model.addLayer(Convolution2D(size=[5, 5], features=16))
model.addLayer(ReLU())
model.addLayer(MaxPool(size=[2, 2]))
model.addLayer(Convolution2D(size=[5, 5], features=16))
model.addLayer(ReLU())
model.addLayer(MaxPool(size=[2, 2]))
model.addLayer(Convolution2D(size=[5, 5], features=16))
model.addLayer(ReLU())
model.addLayer(MaxPool(size=[2, 2]))
model.addLayer(FullyConnected(features=10))
model.addOutputLayer(Softmax())
model.train(
dataset=mnist,
eval_every=5,
epochs=500,
evaluation_size=500,
batch_size=100,
optimizer=train.MomentumOptimizer(0.005, 0.9))
model.test(mnist)
model_path = "trained_model/" + model_name + "/" + model_name
model.save(model_path)
def train(self,
dataset,
eval_every=5,
epochs=500,
evaluation_size=500,
batch_size=100,
optimizer=train.MomentumOptimizer(0.005, 0.9)):
train_step = optimizer.minimize(self.loss)
prediction = argmax(self.model, 1, name="prediction")
result = equal(argmax(self.labels, 1), prediction, name="result")
self.accuracy = reduce_mean(cast(result, float32), name="accuracy")
train_loss = []
train_accuracy = []
test_accuracy = []
global_variables_initializer().run()
for i in range(epochs):
# Lay ra batch_size hinh anh tu tap train
train_batch = dataset.train.next_batch(batch_size)
train_dict = {
self.inputs: train_batch[0],
self.labels: train_batch[1]
}
if i % eval_every == 0:
# Cu sau eval_every buoc lap thi test mot lan
test_batch = dataset.test.next_batch(evaluation_size)
temp_train_loss = self.loss.eval(feed_dict=train_dict)
temp_train_accuracy = self.accuracy.eval(feed_dict=train_dict)
temp_test_accuracy = self.accuracy.eval(
feed_dict={
self.inputs: test_batch[0],
self.labels: test_batch[1]
})
print(
'Epoch # %d, Train Loss: %g. Train Accuracy (Test Accuracy): %g (%g)'
% (i, temp_train_loss, temp_train_accuracy,
temp_test_accuracy))
# Luu cac gia tri de ve bieu do
train_loss.append(temp_train_loss)
train_accuracy.append(temp_train_accuracy)
test_accuracy.append(temp_test_accuracy)
# Chay thuat toan toi uu ham mat mat
self.sess.run(train_step, feed_dict=train_dict)
# Show plots
loss_per_epoch(epochs, eval_every, train_loss)
train_test_accuracy(epochs, eval_every, train_accuracy, test_accuracy)
def train(self, cases, labels):
'''训练'''
limit = self.init_temp['max_itera']
learn_rate = self.init_temp['dg_initial'] * 0.00000001
self.loss = reduce_mean(
reduce_sum(square((self.label_layer - self.output_layer)),
reduction_indices=[1]))
learning_rate = placeholder(float32, shape=[])
#self.train_step = #tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(self.loss)
self.train_step = train.MomentumOptimizer(learning_rate=learning_rate,
momentum=0.9,
use_locking=False,
use_nesterov=True).minimize(
self.loss) #动量优化器
self.session.run(global_variables_initializer())
next_done = 0
starttime = datetime.now()
for i in range(limit):
self.session.run(self.train_step,
feed_dict={
self.input_layer: cases,
self.label_layer: labels,
learning_rate: learn_rate
})
if i % 2 == 0:
if not GlobalMap().get('running_global'):
break
next_done = 1
next_times = i + 1
train_0 = self.train_step
loss0 = self.session.run(self.loss,
feed_dict={
self.input_layer: cases,
self.label_layer: labels
})
#更新前台和处理数据
if next_done == 1 and i % next_times == 0:
loss1 = self.session.run(self.loss,
feed_dict={
self.input_layer: cases,
self.label_layer: labels
})
GlobalMap().set(loss_now=loss0,
times=i,
used_time=(datetime.now() - starttime).seconds)
if loss1 < self.init_temp['precision']:
break
if (i + 1) % 50 == 0 and self.init_temp['fix_auto'] and (
loss0 - loss1 < 0.1 or isnan(loss1)):
self.auto_times += 1
#print(self.auto_times)
GlobalMap().set(progress_num=5 + i * 100 // limit +
10) #进度条
if self.auto_times > 8:
break
if loss1 < loss0: #在动量的基础上调整学习率
learn_rate *= 1.0 + self.init_temp['dg_adaptive']
else:
learn_rate *= 1.0 - self.init_temp['dg_adaptive']
self.train_step = train_0
#print("1:", loss1)
#print("shu:",learn_rate)
next_done = 0
self.saver.Saver().save(self.session, self.ckpt_path)