def train(self):
mnist = logistic_regression.load_data_mnist()
# negative loss likelihood
cross_entropy = -tf.reduce_sum(self.y_ * tf.log(self.y_conv))
# 最適化関数にコストを渡す
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
# 正しいかどうか
correct_prediction = tf.equal(tf.argmax(self.y_conv, 1), tf.argmax(self.y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
# 全変数の初期化
self.sess.run(tf.initialize_all_variables())
for i in range(1000):
batch = mnist.train.next_batch(50)
if i % 100 == 0:
# 100回ごとにテスト
train_accuracy = accuracy.eval(feed_dict={self.x: batch[0], self.y_: batch[1], self.keep_prob: 1.0})
print "setp %d, training accuracy %g" % (i, train_accuracy)
train_step.run(feed_dict={self.x: batch[0], self.y_: batch[1], self.keep_prob: 1.0})
print "test accuracy %g" % accuracy.eval(
feed_dict={self.x: mnist.test.images, self.y_: mnist.test.labels, self.keep_prob: 1.0}
)
def test():
print("load settings:")
print(FLAGS.learning_rate)
print("load dataset:")
mnist = logistic_regression.load_data_mnist()
print("place holder")
images_placeholder, labels_placeholder = ph.placeholder_inputs(FLAGS.batch_size, model.IMAGE_PIXELS, model.NUM_CLASSES)
print("fill feed dict")
fd.fill_feed_dict(mnist.train, images_placeholder, labels_placeholder, FLAGS.batch_size, FLAGS.fake_data)
def test():
print("load settings:")
print(FLAGS.learning_rate)
print("load dataset:")
mnist = logistic_regression.load_data_mnist()
print("place holder")
images_placeholder, labels_placeholder = ph.placeholder_inputs(
FLAGS.batch_size, model.IMAGE_PIXELS, model.NUM_CLASSES)
print("fill feed dict")
fd.fill_feed_dict(mnist.train, images_placeholder, labels_placeholder,
FLAGS.batch_size, FLAGS.fake_data)
def train(self):
mnist = logistic_regression.load_data_mnist()
# negative loss likelihood
cross_entropy = -tf.reduce_sum(self.y_ * tf.log(self.y_conv))
# 最適化関数にコストを渡す
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
# 正しいかどうか
correct_prediction = tf.equal(tf.argmax(self.y_conv, 1),
tf.argmax(self.y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
# 全変数の初期化
self.sess.run(tf.initialize_all_variables())
for i in range(1000):
batch = mnist.train.next_batch(50)
if i % 100 == 0:
# 100回ごとにテスト
train_accuracy = accuracy.eval(feed_dict={
self.x: batch[0],
self.y_: batch[1],
self.keep_prob: 1.0
})
print "setp %d, training accuracy %g" % (i, train_accuracy)
# input data, target data, placeholder
train_step.run(feed_dict={
self.x: batch[0],
self.y_: batch[1],
self.keep_prob: 1.0
})
print "test accuracy %g" % accuracy.eval(
feed_dict={
self.x: mnist.test.images,
self.y_: mnist.test.labels,
self.keep_prob: 1.0
})
def run_training():
data_sets = logistic_regression.load_data_mnist(one_hot=False)
hiddens = [FLAGS.hidden1, FLAGS.hidden2]
# Graph実行用のSession
with tf.Graph().as_default():
# placeholderの生成
images_placeholder, labels_placeholder = ph.placeholder_inputs(FLAGS.batch_size, model.IMAGE_PIXELS, model.NUM_CLASSES)
# logitsの生成
logits = model.inference(images_placeholder, hiddens)
# lossの生成
loss = model.loss(logits, labels_placeholder)
# trainオペレーションの生成
train_op = model.train(loss, FLAGS.learning_rate)
# evaluationオペレーション
eval_correct = model.evaluation(logits, labels_placeholder)
# 可視化用サマリ(グラフのビルド時に初期化)
summary_op = tf.merge_all_summaries()
# modelの保存用オブジェクト
saver = tf.train.Saver()
# sessionオブジェクト
sess = tf.Session()
# 変数の初期化
init = tf.initialize_all_variables()
sess.run(init)
# 可視化用の変数初期化
summary_writer = tf.train.SummaryWriter(FLAGS.model_dir, graph_def=sess.graph_def)
# 学習ステップの実行
for step in xrange(FLAGS.max_steps):
start_time = time.time()
# 現在のステップ用のデータを取得
feed_dict = fd.fill_feed_dict(data_sets.train, images_placeholder, labels_placeholder, FLAGS.batch_size, FLAGS.fake_data)
# train_opを実行
_, loss_value = sess.run([train_op, loss], feed_dict=feed_dict)
duration = time.time() - start_time
# 100ステップごとにサマリーを出力
if step%FLAGS.summary_step == 0:
# stdout
print('Step %d: loss = %.2f , %.3f sec' % (step, loss_value, duration))
# 可視化用の変数アップデート
summary_str = sess.run(summary_op, feed_dict=feed_dict)
summary_writer.add_summary(summary_str, step)
# 1000ステップごとにcheckpointを保存(最大ステップに達した際も保存)
if (step+1)%FLAGS.checkpoint_step == 0 or (step+1) == FLAGS.max_steps:
saver.save(sess, FLAGS.model_dir, global_step =step)
# Evaluate model
print('CheckPoint Train Evalation:')
do_eval(sess, eval_correct, images_placeholder, labels_placeholder, data_sets.train)
print('CheckPoint Valid Evalation:')
do_eval(sess, eval_correct, images_placeholder, labels_placeholder, data_sets.validation)
print('CheckPoint Test Evalation:')
do_eval(sess, eval_correct, images_placeholder, labels_placeholder, data_sets.test)
def run_training():
data_sets = logistic_regression.load_data_mnist(one_hot=False)
hiddens = [FLAGS.hidden1, FLAGS.hidden2]
# Graph実行用のSession
with tf.Graph().as_default():
# placeholderの生成
images_placeholder, labels_placeholder = ph.placeholder_inputs(
FLAGS.batch_size, model.IMAGE_PIXELS, model.NUM_CLASSES)
# logitsの生成
logits = model.inference(images_placeholder, hiddens)
# lossの生成
loss = model.loss(logits, labels_placeholder)
# trainオペレーションの生成
train_op = model.train(loss, FLAGS.learning_rate)
# evaluationオペレーション
eval_correct = model.evaluation(logits, labels_placeholder)
# 可視化用サマリ(グラフのビルド時に初期化)
summary_op = tf.merge_all_summaries()
# modelの保存用オブジェクト
saver = tf.train.Saver()
# sessionオブジェクト
sess = tf.Session()
# 変数の初期化
init = tf.initialize_all_variables()
sess.run(init)
# 可視化用の変数初期化
summary_writer = tf.train.SummaryWriter(FLAGS.model_dir,
graph_def=sess.graph_def)
# 学習ステップの実行
for step in xrange(FLAGS.max_steps):
start_time = time.time()
# 現在のステップ用のデータを取得
feed_dict = fd.fill_feed_dict(data_sets.train, images_placeholder,
labels_placeholder, FLAGS.batch_size,
FLAGS.fake_data)
# train_opを実行
_, loss_value = sess.run([train_op, loss], feed_dict=feed_dict)
duration = time.time() - start_time
# 100ステップごとにサマリーを出力
if step % FLAGS.summary_step == 0:
# stdout
print('Step %d: loss = %.2f , %.3f sec' %
(step, loss_value, duration))
# 可視化用の変数アップデート
summary_str = sess.run(summary_op, feed_dict=feed_dict)
summary_writer.add_summary(summary_str, step)
# 1000ステップごとにcheckpointを保存(最大ステップに達した際も保存)
if (step + 1) % FLAGS.checkpoint_step == 0 or (
step + 1) == FLAGS.max_steps:
saver.save(sess, FLAGS.model_dir, global_step=step)
# Evaluate model
print('CheckPoint Train Evalation:')
do_eval(sess, eval_correct, images_placeholder,
labels_placeholder, data_sets.train)
print('CheckPoint Valid Evalation:')
do_eval(sess, eval_correct, images_placeholder,
labels_placeholder, data_sets.validation)
print('CheckPoint Test Evalation:')
do_eval(sess, eval_correct, images_placeholder,
labels_placeholder, data_sets.test)