基於tensorflow的MNIST手寫識別

這個例子，是學習tensorflow的人員通常會用到的，也是基本的學習曲線中的一環。我也是！

 

這個例子很簡單，這里，就是簡單的說下，不同的tensorflow版本，相關的接口函數，可能會有不一樣喲。在TensorFlow的中文介紹文檔中的內容，有些可能與你使用的tensorflow的版本不一致了，我這里用到的tensorflow的版本就有這個問題。 另外，還給大家說下，例子中的MNIST所用到的資源圖片，在原始的官網上，估計很多人都下載不到了。我也提供一下下載地址。

 

我的tensorflow的版本信息：

>>> import tensorflow as tf
>>> print tf.VERSION    
1.0.1
>>> print tf.GIT_VERSION
v1.0.0-65-g4763edf-dirty
>>> print tf.COMPILER_VERSION
4.8.4

 

下面，就看看，我參考的中文tensorflow網站的代碼，在自己的環境里，運行的結果。

[root@bogon tensorflow]# python
Python 2.7.5 (default, Nov  6 2016, 00:28:07) 
[GCC 4.8.5 20150623 (Red Hat 4.8.5-11)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow.examples.tutorials.mnist.input_data as input_data
>>> mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py", line 211, in read_data_sets
    SOURCE_URL + TRAIN_IMAGES)
  File "/usr/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/datasets/base.py", line 208, in maybe_download
    temp_file_name, _ = urlretrieve_with_retry(source_url)
  File "/usr/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/datasets/base.py", line 165, in wrapped_fn
    return fn(*args, **kwargs)
  File "/usr/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/datasets/base.py", line 190, in urlretrieve_with_retry
    return urllib.request.urlretrieve(url, filename)
  File "/usr/lib64/python2.7/urllib.py", line 94, in urlretrieve
    return _urlopener.retrieve(url, filename, reporthook, data)
  File "/usr/lib64/python2.7/urllib.py", line 240, in retrieve
    fp = self.open(url, data)
  File "/usr/lib64/python2.7/urllib.py", line 203, in open
    return self.open_unknown_proxy(proxy, fullurl, data)
  File "/usr/lib64/python2.7/urllib.py", line 222, in open_unknown_proxy
    raise IOError, ('url error', 'invalid proxy for %s' % type, proxy)
IOError: [Errno url error] invalid proxy for http: '10.90.1.101:8080'
>>> 
>>> mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
28 Extracting MNIST_data/train-images-idx3-ubyte.gz 29 Extracting MNIST_data/train-labels-idx1-ubyte.gz 30 Extracting MNIST_data/t10k-images-idx3-ubyte.gz 31 Extracting MNIST_data/t10k-labels-idx1-ubyte.gz 32 >>> import tensorflow as tf
>>> x = tf.placeholder(tf.float32, [None, 784])
>>> W = tf.Variable(tf.zeros([784,10]))
>>> b = tf.Variable(tf.zeros([10]))
>>> y = tf.nn.softmax(tf.matmul(x,W) + b)
>>> y_ = tf.placeholder("float", [None,10])
>>> cross_entropy = -tf.reduce_sum(y_*tf.log(y))
>>> train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)
40 >>> init = tf.initialize_all_variables() 41 WARNING:tensorflow:From <stdin>:1: initialize_all_variables (from tensorflow.python.ops.variables) is deprecated and will be removed after 2017-03-02.
Instructions for updating:
Use `tf.global_variables_initializer` instead.
44 >>> init = tf.global_variables_initializer() 45 >>> sess = tf.Session()
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE3 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
>>> sess.run(init)
>>> for i in range(1000):
...   batch_xs, batch_ys = mnist.train.next_batch(100)
...   sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
... 
>>> correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
>>> accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
>>> print sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels})
0.9088
>>> 

上述日志，是我的測試全過程記錄，上面反映的信息有如下幾點：

1. 紅色部分的錯誤，因為我本地機器是通過代理上網的，這個過程中，tensorflow會用urllib進行MNIST的圖片資源的下載，由於網絡問題，資源文件下載失敗。

2. 都有哪些資源文件要下載呢？追蹤日志中的文件/usr/lib/python2.7/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py第211行前后：

def read_data_sets(train_dir,
                   fake_data=False,
                   one_hot=False,
                   dtype=dtypes.float32,
                   reshape=True,
                   validation_size=5000):
  if fake_data:

    def fake():
      return DataSet([], [], fake_data=True, one_hot=one_hot, dtype=dtype)

    train = fake()
    validation = fake()
    test = fake()
    return base.Datasets(train=train, validation=validation, test=test)

 TRAIN_IMAGES = 'train-images-idx3-ubyte.gz' TRAIN_LABELS = 'train-labels-idx1-ubyte.gz' TEST_IMAGES = 't10k-images-idx3-ubyte.gz' TEST_LABELS = 't10k-labels-idx1-ubyte.gz'

  local_file = base.maybe_download(TRAIN_IMAGES, train_dir,
                                   SOURCE_URL + TRAIN_IMAGES)
  with open(local_file, 'rb') as f:
    train_images = extract_images(f)

  local_file = base.maybe_download(TRAIN_LABELS, train_dir,
                                   SOURCE_URL + TRAIN_LABELS)
  with open(local_file, 'rb') as f:
    train_labels = extract_labels(f, one_hot=one_hot)

  local_file = base.maybe_download(TEST_IMAGES, train_dir,
                                   SOURCE_URL + TEST_IMAGES)
  with open(local_file, 'rb') as f:
    test_images = extract_images(f)

  local_file = base.maybe_download(TEST_LABELS, train_dir,
                                   SOURCE_URL + TEST_LABELS)
  with open(local_file, 'rb') as f:
    test_labels = extract_labels(f, one_hot=one_hot)

  if not 0 <= validation_size <= len(train_images):
    raise ValueError(
        'Validation size should be between 0 and {}. Received: {}.'
        .format(len(train_images), validation_size))

  validation_images = train_images[:validation_size]
  validation_labels = train_labels[:validation_size]
  train_images = train_images[validation_size:]
  train_labels = train_labels[validation_size:]

  train = DataSet(train_images, train_labels, dtype=dtype, reshape=reshape)
  validation = DataSet(validation_images,
                       validation_labels,
                       dtype=dtype,
                       reshape=reshape)
  test = DataSet(test_images, test_labels, dtype=dtype, reshape=reshape)

  return base.Datasets(train=train, validation=validation, test=test)

看到上面紅色的部分，就是這里需要下載的圖片資源文件。這個，我的網絡環境是下載不了的。我通過其他途徑下載到了這里需要的資源。我將下載的圖片資源，放在了我進入python時所在的路徑下。雖然直接下載沒有成功，但是在當前路徑下還是創建了MNIST_data的目錄的。如下圖，紅色圈目錄就是程序創建的目錄。我將下載的train-images-idx3-ubyte.gz，train-labels-idx1-ubyte.gz，t10k-images-idx3-ubyte.gz，t10k-labels-idx1-ubyte.gz放在MNIST_data目錄了。

然后，再次執行mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)，就ok了，不會報錯。得到28-31行的輸出信息。

3. 執行到第40行的代碼時，爆出WARNING，提示用新的函數，按照提示信息，執行了第41行的代碼，OK。說明版本兼容性，在tensorflow中需要注意。

4. 執行后，得到結果，如60行顯示，識別率為0.9088。

 

關於MNIST的這個例子的手寫識別性能的理論，不是本博文的重點，讀者可以參照MNIST相關的文章自行學習。

最后，附上MNIST這個例子中，用到的資源圖片下載地址，點擊進行下載。（說明：需要積分才能下載的，諒解）