CIFAR-10 TensorFlow模型架构（二）

二、模型架构

3.cifar10_train.py:在CPU或GPU上训练CIFAR-10 模型

# -*- coding: utf-8 -*-
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================

"""A binary to train CIFAR-10 using a single GPU.
Accuracy:
cifar10_train.py achieves ~86% accuracy after 100K steps (256 epochs of
data) as judged by cifar10_eval.py.
Speed: With batch_size 128.
System        | Step Time (sec/batch)  |     Accuracy
------------------------------------------------------------------
1 Tesla K20m  | 0.35-0.60              | ~86% at 60K steps  (5 hours)
1 Tesla K40m  | 0.25-0.35              | ~86% at 100K steps (4 hours)
Usage:
Please see the tutorial and website for how to download the CIFAR-10
data set, compile the program and train the model.
http://tensorflow.org/tutorials/deep_cnn/
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

from datetime import datetime
import os.path
import time

import numpy as np
from six.moves import xrange  # pylint: disable=redefined-builtin
import tensorflow as tf

#from tensorflow.models.image.cifar10 import cifar10
import cifar10

FLAGS = tf.app.flags.FLAGS

tf.app.flags.DEFINE_string('train_dir', '/tmp/cifar10_train',
                           """Directory where to write event logs """
                           """and checkpoint.""")
tf.app.flags.DEFINE_integer('max_steps', 10000,
                            """Number of batches to run.""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
                            """Whether to log device placement.""")

def train():
  """训练 CIFAR-10 for a number of steps."""
  with tf.Graph().as_default():
    global_step = tf.Variable(0, trainable=False)

    # 获得CIFAR-10的图片核标签
    images, labels = cifar10.distorted_inputs()

    # 建立一个图表，从推理模型计算逻辑预测
    logits = cifar10.inference(images)

    # 计算损失.
    loss = cifar10.loss(logits, labels)

    # 构建一个用一批示例数据训练模型并更新模型参数的图.
    train_op = cifar10.train(loss, global_step)

    # 创建一个saver.
    saver = tf.train.Saver(tf.all_variables())

    # Build the summary operation based on the TF collection of Summaries.
    # 基于TF summaries的集合，建造summary操作
    #summary_op = tf.merge_all_summaries()
    summary_op = tf.summary.merge_all()

    # 构建初始化操作
    init = tf.initialize_all_variables()

    # 开始运行计算图的操作
    sess = tf.Session(config=tf.ConfigProto(
                      log_device_placement=FLAGS.log_device_placement))
    sess.run(init)

    # 开始队列的运行.
    tf.train.start_queue_runners(sess=sess)

    #summary_writer = tf.train.SummaryWriter(FLAGS.train_dir, sess.graph)
    summary_writer = tf.summary.FileWriter(FLAGS.train_dir,sess.graph)

    for step in xrange(FLAGS.max_steps):
      start_time = time.time()
      _, loss_value = sess.run([train_op, loss])
      duration = time.time() - start_time

      assert not np.isnan(loss_value), 'Model diverged with loss = NaN'

      if step % 10 == 0:
        num_examples_per_step = FLAGS.batch_size
        examples_per_sec = num_examples_per_step / duration
        sec_per_batch = float(duration)

        format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
                      'sec/batch)')
        print (format_str % (datetime.now(), step, loss_value,
                             examples_per_sec, sec_per_batch))

      if step % 100 == 0:
        summary_str = sess.run(summary_op)
        summary_writer.add_summary(summary_str, step)

      # Save the model checkpoint periodically.
      if step % 1000 == 0 or (step + 1) == FLAGS.max_steps:
        checkpoint_path = os.path.join(FLAGS.train_dir, 'model.ckpt')
        saver.save(sess, checkpoint_path, global_step=step)

def main(argv=None):  # pylint: disable=unused-argument
  cifar10.maybe_download_and_extract()
  if tf.gfile.Exists(FLAGS.train_dir):
    tf.gfile.DeleteRecursively(FLAGS.train_dir)
  tf.gfile.MakeDirs(FLAGS.train_dir)
  train()

if __name__ == '__main__':
  tf.app.run()
 

4.cifar10_multi_gpu_train.py：在多GPU上训练CIFAR-10 模型.

5.cifar10_eval.py ：评估一个CIFAR模型的预测性能