modify

Author: jianan liu

__init__.py

@@ -0,0 +1,3 @@
+from chapter_18 import env
+
+__all__ = ["env"]

chapter_1/README.md

@@ -60,7 +60,7 @@ step 700, training accuracy 0.9
 step 800, training accuracy 0.91  
 step 900, training accuracy 0.88  
 
--tf.reduce_sum(y_ * tf.nn.softmax(y_conv)  
+-tf.reduce_sum(y_ * tf.log(tf.nn.softmax(y_conv))  
 step 0, training accuracy 0.08  
 step 100, training accuracy 0.52  
 step 200, training accuracy 0.62  

chapter_1/convolutional.py

@@ -60,10 +60,10 @@ def max_pool_2x2(x):
 
     # todo 这个调节参数能力强
     # 我们不采用先Softmax再计算交叉熵的方法，而是直接用tf.nn.softmax_cross_entropy_with_logits直接计算
-    cross_entropy = tf.reduce_mean(
-        tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))
+    # cross_entropy = tf.reduce_mean(
+    #     tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))
     # 自定义交叉熵
-    # cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.nn.softmax(y_conv)))
+    cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(tf.nn.softmax(y_conv))))
     # 同样定义train_step
     train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
 

chapter_12/sample.py

@@ -16,7 +16,7 @@
 
 
 def main(_):
-    FLAGS.start_string = FLAGS.start_string.decode('utf-8')
+    FLAGS.start_string = FLAGS.start_string.encode('utf-8')
     converter = TextConverter(filename=FLAGS.converter_path)
     if os.path.isdir(FLAGS.checkpoint_path):
         FLAGS.checkpoint_path =\

chapter_18/q_learning.py

@@ -1,7 +1,7 @@
 from __future__ import print_function
 import numpy as np
 import time
-from env import Env
+from chapter_18.env import Env
 
 
 EPSILON = 0.1

chapter_18/q_learning_reprint.py

@@ -1,7 +1,7 @@
 from __future__ import print_function
 import numpy as np
 import time
-from env import Env
+from chapter_18.env import Env
 from reprint import output
 
 
@@ -25,7 +25,7 @@ def epsilon_greedy(Q, state):
 Q = np.zeros((e.state_num, 4))
 
 with output(output_type="list", initial_len=len(e.map), interval=0) as output_list:
-    for i in range(100):
+    for i in range(1):
         e = Env()
         while (e.is_end is False) and (e.step < MAX_STEP):
             action = epsilon_greedy(Q, e.present_state)

chapter_19/sarsa.py

@@ -1,7 +1,7 @@
 from __future__ import print_function
 import numpy as np
 import time
-from env import Env
+from chapter_19.env import Env
 
 
 EPSILON = 0.1

chapter_19/sarsa_reprint.py

@@ -1,7 +1,7 @@
 from __future__ import print_function
 import numpy as np
 import time
-from env import Env
+from chapter_19.env import Env
 from reprint import output
 
 

chapter_2/__init__.py

@@ -43,7 +43,7 @@
 from six.moves import urllib
 import tensorflow as tf
 
-from chapter_2 import *
+import cifar10_input
 
 FLAGS = tf.app.flags.FLAGS
 
@@ -54,6 +54,8 @@
                            """Path to the CIFAR-10 data directory.""")
 tf.app.flags.DEFINE_boolean('use_fp16', False,
                             """Train the model using fp16.""")
+tf.app.flags.DEFINE_boolean('use_raw_img', True,
+                            """Train the model using raw img""")
 
 # Global constants describing the CIFAR-10 data set.
 IMAGE_SIZE = cifar10_input.IMAGE_SIZE
@@ -151,9 +153,11 @@ def distorted_inputs():
   """
   if not FLAGS.data_dir:
     raise ValueError('Please supply a data_dir')
-  data_dir = os.path.join(FLAGS.data_dir, 'cifar-10-batches-bin')
+  default_data_dir = 'cifar-10-raw-pic' if FLAGS.use_raw_img else 'cifar-10-batches-bin'
+  data_dir = os.path.join(FLAGS.data_dir, default_data_dir)
   images, labels = cifar10_input.distorted_inputs(data_dir=data_dir,
-                                                  batch_size=FLAGS.batch_size)
+                                                  batch_size=FLAGS.batch_size,
+                                                  use_raw_img=FLAGS.use_raw_img)
   if FLAGS.use_fp16:
     images = tf.cast(images, tf.float16)
     labels = tf.cast(labels, tf.float16)

chapter_2/cifar10.py

@@ -41,7 +41,7 @@
 import numpy as np
 import tensorflow as tf
 
-from chapter_2 import cifar10
+import cifar10
 
 FLAGS = tf.app.flags.FLAGS
 

chapter_2/cifar10_eval.py

@@ -1,6 +1,6 @@
 #coding: utf-8
 # 导入当前目录的cifar10_input，这个模块负责读入cifar10数据
-from chapter_2 import cifar10_input
+import cifar10_input
 # 导入TensorFlow和其他一些可能用到的模块。
 import tensorflow as tf
 import os

chapter_2/cifar10_extract.py

@@ -97,6 +97,43 @@ class CIFAR10Record(object):
 
   return result
 
+def read_and_decode(filename_queue):
+
+    class CIFAR10Record(object):
+        pass
+    result = CIFAR10Record()
+
+    # Dimensions of the images in the CIFAR-10 dataset.
+    # See http://www.cs.toronto.edu/~kriz/cifar.html for a description of the
+    # input format.
+    label_bytes = 1  # 2 for CIFAR-100
+    result.height = 32
+    result.width = 32
+    result.depth = 3
+
+    # 创建文件队列,不限读取的数量
+    # filename_queue = tf.train.string_input_producer([filename])
+    # create a reader from file queue
+    reader = tf.TFRecordReader()
+    # reader从文件队列中读入一个序列化的样本
+    _, serialized_example = reader.read(filename_queue)
+    # get feature from serialized example
+    # 解析符号化的样本
+    features = tf.parse_single_example(
+        serialized_example,
+        features={
+            'label': tf.FixedLenFeature([], tf.int64),
+            'img_raw': tf.FixedLenFeature([], tf.string)
+        })
+    label = features['label']
+    label = tf.cast(label, tf.int32)
+    result.label = tf.reshape(label, [1])
+    img = features['img_raw']
+    img = tf.decode_raw(img, tf.uint8)
+    img = tf.reshape(img, [result.height, result.width, result.depth])
+    # img = tf.cast(img, tf.float32) * (1. / 255) - 0.5
+    result.uint8image = img
+    return result
 
 def _generate_image_and_label_batch(image, label, min_queue_examples,
                                     batch_size, shuffle):
@@ -137,7 +174,7 @@ def _generate_image_and_label_batch(image, label, min_queue_examples,
   return images, tf.reshape(label_batch, [batch_size])
 
 
-def distorted_inputs(data_dir, batch_size):
+def distorted_inputs(data_dir, batch_size, use_raw_img):
   """Construct distorted input for CIFAR training using the Reader ops.
 
   Args:
@@ -148,8 +185,7 @@ def distorted_inputs(data_dir, batch_size):
     images: Images. 4D tensor of [batch_size, IMAGE_SIZE, IMAGE_SIZE, 3] size.
     labels: Labels. 1D tensor of [batch_size] size.
   """
-  filenames = [os.path.join(data_dir, 'data_batch_%d.bin' % i)
-               for i in xrange(1, 6)]
+  filenames = [os.path.join(data_dir, 'data_batch_%d.bin' % i) for i in xrange(1, 6)] if not use_raw_img else [os.path.join(data_dir, 'cifar10_train.tfrecords')]
   for f in filenames:
     if not tf.gfile.Exists(f):
       raise ValueError('Failed to find file: ' + f)
@@ -158,7 +194,8 @@ def distorted_inputs(data_dir, batch_size):
   filename_queue = tf.train.string_input_producer(filenames)
 
   # Read examples from files in the filename queue.
-  read_input = read_cifar10(filename_queue)
+  # read_input = read_cifar10(filename_queue)
+  read_input = read_and_decode(filename_queue)
   reshaped_image = tf.cast(read_input.uint8image, tf.float32)
 
   height = IMAGE_SIZE

chapter_2/cifar10_input.py

@@ -41,7 +41,7 @@
 
 import tensorflow as tf
 
-from chapter_2 import cifar10
+import cifar10
 
 FLAGS = tf.app.flags.FLAGS
 

chapter_2/cifar10_train.py

@@ -11,7 +11,7 @@
     # 我们要读三幅图片A.jpg, B.jpg, C.jpg
     filename = ['A.jpg', 'B.jpg', 'C.jpg']
     # string_input_producer会产生一个文件名队列
-    filename_queue = tf.train.string_input_producer(filename, shuffle=False, num_epochs=5)
+    filename_queue = tf.train.string_input_producer(filename, shuffle=False, num_epochs=3)
     # reader从文件名队列中读数据。对应的方法是reader.read
     reader = tf.WholeFileReader()
     key, value = reader.read(filename_queue)
@@ -23,7 +23,12 @@
     while True:
         i += 1
         # 获取图片数据并保存
-        image_data = sess.run(value)
-        with open('read/test_%d.jpg' % i, 'wb') as f:
-            f.write(image_data)
+        try:
+            (k, image_data) = sess.run([key, value])
+            with open('read/test_%d.jpg' % i, 'wb') as f:
+                print("key:" + str(k) + " finished")
+                f.write(image_data)
+        except tf.errors.OutOfRangeError:
+            print("string input queue is over")
+            break
 # 程序最后会抛出一个OutOfRangeError，这是epoch跑完，队列关闭的标志

chapter_2/test.py

@@ -0,0 +1,90 @@
+# 将原始图片转换成需要的大小，并将其保存
+#https://blog.csdn.net/ywx1832990/article/details/78609323
+# ========================================================================================
+import os
+import tensorflow as tf
+from PIL import Image
+
+# 原始图片的存储位置
+orig_picture = '/Users/jiananliu/work/machinelearn/dl/Deep-Learning-21-Examples/chapter_2/cifar10_data/cifar-10-raw-pic/cifar10/'
+
+# 生成图片的存储位置
+gen_picture = '/Users/jiananliu/work/machinelearn/dl/Deep-Learning-21-Examples/chapter_2/cifar10_data/cifar-10-raw-pic'
+
+# 需要的识别类型
+classes = {'airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck'}
+
+# 样本总数
+num_samples = 120
+heigth = 32
+width = 32
+channel = 3
+
+
+# 制作TFRecords数据
+def create_record():
+    writer = tf.python_io.TFRecordWriter("cifar10_train.tfrecords")
+    for index, name in enumerate(classes):
+        class_path = orig_picture + "/" + name + "/"
+        for img_name in os.listdir(class_path):
+            img_path = class_path + img_name
+            img = Image.open(img_path)
+            img = img.resize((heigth, width))  # 设置需要转换的图片大小
+            img_raw = img.tobytes()  # 将图片转化为原生bytes
+            print(index, img_raw)
+            example = tf.train.Example(
+                features=tf.train.Features(feature={
+                    "label": tf.train.Feature(int64_list=tf.train.Int64List(value=[index])),
+                    'img_raw': tf.train.Feature(bytes_list=tf.train.BytesList(value=[img_raw]))
+                }))
+            writer.write(example.SerializeToString())
+    writer.close()
+
+
+# =======================================================================================
+def read_and_decode(filename):
+    # 创建文件队列,不限读取的数量
+    filename_queue = tf.train.string_input_producer([filename])
+    # create a reader from file queue
+    reader = tf.TFRecordReader()
+    # reader从文件队列中读入一个序列化的样本
+    _, serialized_example = reader.read(filename_queue)
+    # get feature from serialized example
+    # 解析符号化的样本
+    features = tf.parse_single_example(
+        serialized_example,
+        features={
+            'label': tf.FixedLenFeature([], tf.int64),
+            'img_raw': tf.FixedLenFeature([], tf.string)
+        })
+    label = features['label']
+    img = features['img_raw']
+    img = tf.decode_raw(img, tf.uint8)
+    img = tf.reshape(img, [heigth, width, channel])
+    # img = tf.cast(img, tf.float32) * (1. / 255) - 0.5
+    label = tf.cast(label, tf.int32)
+    return img, label
+
+
+# =======================================================================================
+if __name__ == '__main__':
+    create_record()
+    batch = read_and_decode('cifar10_train.tfrecords')
+    init_op = tf.group(tf.global_variables_initializer(), tf.local_variables_initializer())
+
+    with tf.Session() as sess:  # 开始一个会话
+        sess.run(init_op)
+        coord = tf.train.Coordinator()
+        threads = tf.train.start_queue_runners(coord=coord)
+
+        for i in range(num_samples):
+            example, lab = sess.run(batch)  # 在会话中取出image和label
+            img = Image.fromarray(example, 'RGB')  # 这里Image是之前提到的
+            img.save(gen_picture + '/' + str(i) + 'samples' + str(lab) + '.jpg')  # 存下图片;注意cwd后边加上‘/’
+            print(example, lab)
+        coord.request_stop()
+        coord.join(threads)
+        sess.close()
+
+# ========================================================================================
+

chapter_2/test_tfrecord.py

@@ -35,7 +35,7 @@ python main.py --network_header_type=nips --env_name=Breakout-v0 --use_gpu=True
 
 测试在CPU上训练的模型：
 ```
-python main.py --network_header_type=nips --env_name=Breakout-v0 --use_gpu=True --is_train=True
+python main.py --network_header_type=nips --env_name=Breakout-v0 --use_gpu=False --is_train=False
 ```
 
 在上述命令中加入--display=True选项，可以实时显示游戏进程。