Keras 比 TensorFlow 慢得不合理

Question

我正在尝试实现this TensorFlow example 的神经网络，但使用的是 Keras。

您可以在文章底部找到这两种实现的代码。

我的问题是使用 TensorFlow 的代码大约需要 1 分 30 分钟，而使用 Keras 需要 18 分钟！

我的问题是：

• 我在将 TensorFlow 代码转换为 Keras 代码时是否犯了一个新手错误？
• 或者 Keras 的速度非常慢？如果是这样，可以修复吗？

---

张量流代码：

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

import tensorflow as tf

def weight_variable(shape):
  initial = tf.truncated_normal(shape, stddev=0.1)
  return tf.Variable(initial)

def bias_variable(shape):
  initial = tf.constant(0.1, shape=shape)
  return tf.Variable(initial)

def conv2d(x, W):
  return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')

def max_pool_2x2(x):
  return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],
                        strides=[1, 2, 2, 1], padding='SAME')

x = tf.placeholder(tf.float32, [None, 784]) 
x_image = tf.reshape(x, [-1, 28, 28, 1]) 

y_ = tf.placeholder(tf.float32, [None, 10]) 

neurons_nb_layer_1 = 32
neurons_nb_layer_2 = 64
neurons_nb_layer_3 = 1024

W_conv1 = weight_variable([5, 5, 1, neurons_nb_layer_1]) 
b_conv1 = bias_variable([neurons_nb_layer_1]) 

h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1) 
h_pool1 = max_pool_2x2(h_conv1) 

W_conv2 = weight_variable([5, 5, neurons_nb_layer_1, neurons_nb_layer_2])
b_conv2 = bias_variable([neurons_nb_layer_2])

h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2) 

W_fc1 = weight_variable([7 * 7 * neurons_nb_layer_2, neurons_nb_layer_3])
b_fc1 = bias_variable([neurons_nb_layer_3])

h_pool2_flat = tf.reshape(h_pool2, [-1, 7 * 7 * neurons_nb_layer_2]) 
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1) 


keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

W_fc2 = weight_variable([neurons_nb_layer_3, 10])
b_fc2 = bias_variable([10])

y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2

cross_entropy = tf.reduce_mean(
    tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))

train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)

correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1)) 
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) 

import datetime
start = datetime.datetime.now()
with tf.Session() as sess:
  sess.run(tf.global_variables_initializer())
  for i in range(600):
    batch = mnist.train.next_batch(50)
    if i % 100 == 0:
      train_accuracy = accuracy.eval(feed_dict={
          x: batch[0], y_: batch[1], keep_prob: 1.0})
      print('step %d, training accuracy %g' % (i, train_accuracy))
    train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})

  print('test accuracy %g' % accuracy.eval(feed_dict={
      x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))
end = datetime.datetime.now()
time = (end - start).seconds
print(time//60, "min", time%60,"s")

---

Keras 代码：

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

import keras
from keras.models import Sequential

model = Sequential()

bias_initializer = keras.initializers.Constant(value = 0.1)

neurons_nb_layer_1 = 32
neurons_nb_layer_2 = 64
neurons_nb_layer_3 = 1024

from keras.layers import Reshape, Conv2D, MaxPooling2D, Dropout, Flatten, Dense
model.add(Reshape((28, 28, 1), input_shape=(784,)))
model.add(Conv2D(filters = neurons_nb_layer_1, kernel_size = 5*5, padding = 'same', activation = "relu", bias_initializer = bias_initializer))
model.add(MaxPooling2D(padding='same'))
model.add(Conv2D(filters = neurons_nb_layer_2, kernel_size = 5*5, padding = 'same', activation = "relu", bias_initializer = bias_initializer))
model.add(MaxPooling2D(padding='same'))
model.add(Reshape((1,7*7*neurons_nb_layer_2)))
model.add(Dense(units = neurons_nb_layer_3, activation = "relu", bias_initializer = bias_initializer))
model.add(Dropout(rate = 0.5))
model.add(Flatten())
model.add(Dense(units = 10, activation = "relu"))

model.summary()

model.compile(loss = keras.losses.categorical_crossentropy,
              optimizer = 'adam',
              metrics=['accuracy']
              )


import datetime
start2 = datetime.datetime.now()
for i in range(600):
    batch = mnist.train.next_batch(50)
    if i % 100 == 0:
        train_accuracy = model.evaluate(batch[0], batch[1])
        print("step", i, ":", train_accuracy)
    model.train_on_batch(batch[0], batch[1])
end2 = datetime.datetime.now()
time2 = (end2 - start2).seconds
print(time2//60, "min", time2%60,"s")

Answer 1

根据keras documentationkernel_size = 5*5 是25x25 卷积核，而不是5x5 像您的tensorflow 示例
你可能想使用kernel_size=(5,5) 或kernel_size=5