[action] deep learning 深度学习 tensorflow 实战(2) 实现简单神经网络以及随机梯度下降算法S.G.D

在之前的实战(1) 中，我们将数据清洗整理后，得到了'notMNIST.pickle'数据。

本文将阐述利用tensorflow创建一个简单的神经网络以及随机梯度下降算法。

# These are all the modules we'll be using later. Make sure you can import them# before proceeding further.from __future__ import print_functionimport numpy as npimport tensorflow as tffrom six.moves import cPickle as picklefrom six.moves import range

首先，载入之前整理好的数据'notMNIST.pickle'。（在实战(1)中得到的）

pickle_file = 'notMNIST.pickle'with open(pickle_file, 'rb') as f:    save = pickle.load(f)    train_dataset = save['train_dataset']    train_labels = save['train_labels']    valid_dataset = save['valid_dataset']    valid_labels = save['valid_labels']    test_dataset = save['test_dataset']    test_labels = save['test_labels']    del save  # hint to help gc free up memory 帮助回收内存        print('Training set', train_dataset.shape, train_labels.shape)    print('Validation set', valid_dataset.shape, valid_labels.shape)    print('Test set', test_dataset.shape, test_labels.shape)

运行结果为：

Training set (200000, 28, 28) (200000,)Validation set (10000, 28, 28) (10000,)Test set (10000, 28, 28) (10000,)

下一步转换数据格式。

将图像拉成一维数组。


dataset成为二维数组。


label也成为二位数组。


0 对应[1.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0]


1 对应[0.0,1.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0]

image_size = 28num_labels = 10def reformat(dataset, labels):    dataset = dataset.reshape((-1, image_size * image_size)).astype(np.float32) # -1 means unspecified value adaptive     # Map 0 to [1.0, 0.0, 0.0 ...], 1 to [0.0, 1.0, 0.0 ...]    labels = (np.arange(num_labels) == labels[:,None]).astype(np.float32)    return dataset, labelstrain_dataset, train_labels = reformat(train_dataset, train_labels)valid_dataset, valid_labels = reformat(valid_dataset, valid_labels)test_dataset, test_labels = reformat(test_dataset, test_labels)print('Training set', train_dataset.shape, train_labels.shape)print('Validation set', valid_dataset.shape, valid_labels.shape)print('Test set', test_dataset.shape, test_labels.shape)

运行结果为：

Training set (200000, 784) (200000, 10)Validation set (10000, 784) (10000, 10)Test set (10000, 784) (10000, 10)

准备好数据后，首先使用简单梯度下降法的训练数据。

tensorflow 这样工作: 首先描述你的输入，变量，以及操作。这些组成了计算图。 之后的操作要在这个block下面进行。

比如：

with graph.as_default():    ...

然后可以用命令session.run()运行你定义的操作。 上下文管理器用来定义session.你所定义的操作也一定要在session的block下面。

with tf.Session(graph=graph) as session:    ...

这时我们可以载入数据进行训练啦。

# With gradient descent training, even this much data is prohibitive.# Subset the training data for faster turnaround.train_subset = 10000graph = tf.Graph()with graph.as_default():    # Input data. 定义输入数据并载入                            -----------------------------------------1    # Load the training, validation and test data into constants that are    # attached to the graph.    tf_train_dataset = tf.constant(train_dataset[:train_subset, :])    tf_train_labels = tf.constant(train_labels[:train_subset])        tf_valid_dataset = tf.constant(valid_dataset)    tf_test_dataset = tf.constant(test_dataset)      # Variables.定义变量 要训练得到的参数weight, bias  ----------------------------------------2    # These are the parameters that we are going to be training. The weight    # matrix will be initialized using random values following a (truncated)    # normal distribution. The biases get initialized to zero.    weights = tf.Variable(tf.truncated_normal([image_size * image_size, num_labels])) # changing when training     biases = tf.Variable(tf.zeros([num_labels])) # changing when training         #   tf.truncated_normal    #   tf.truncated_normal(shape, mean=0.0, stddev=1.0, dtype=tf.float32, seed=None, name=None)    #   Outputs random values from a truncated normal distribution.    #  The generated values follow a normal distribution with specified mean and    #  standard deviation, except that values whose magnitude is more than 2 standard    #  deviations from the mean are dropped and re-picked.        # tf.zeros    #  tf.zeros([10])      <tf.Tensor 'zeros:0' shape=(10,) dtype=float32>      # Training computation. 训练数据                                ----------------------------------------3    # We multiply the inputs with the weight matrix, and add biases. We compute    # the softmax and cross-entropy (it's one operation in TensorFlow, because    # it's very common, and it can be optimized). We take the average of this    # cross-entropy across all training examples: that's our loss.    logits = tf.matmul(tf_train_dataset, weights) + biases             # tf.matmul          matrix multiply             loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, tf_train_labels))  # compute average cross entropy loss    #  softmax_cross_entropy_with_logits        # The activation ops provide different types of nonlinearities for use in neural    # networks.  These include smooth nonlinearities (`sigmoid`, `tanh`, `elu`,    #   `softplus`, and `softsign`), continuous but not everywhere differentiable    # functions (`relu`, `relu6`, and `relu_x`), and random regularization (`dropout`).            #  tf.reduce_mean    #    tf.reduce_mean(input_tensor, reduction_indices=None, keep_dims=False, name=None)    #   Computes the mean of elements across dimensions of a tensor.      # Optimizer.                                                                    -----------------------------------------4    # We are going to find the minimum of this loss using gradient descent.    optimizer = tf.train.GradientDescentOptimizer(0.5).minimize(loss)     # 0.5 means learning rate    #  tf.train.GradientDescentOptimizer(    #  tf.train.GradientDescentOptimizer(self, learning_rate, use_locking=False, name='GradientDescent')                  # Predictions for the training, validation, and test data.---------------------------------------5    # These are not part of training, but merely here so that we can report    # accuracy figures as we train.        train_prediction = tf.nn.softmax(logits) # weights  and bias have been changed    valid_prediction = tf.nn.softmax(tf.matmul(tf_valid_dataset, weights) + biases)    test_prediction = tf.nn.softmax(tf.matmul(tf_test_dataset, weights) + biases)        # tf.nn.softmax    #  Returns: A `Tensor`. Has the same type as `logits`. Same shape as `logits`.(num, 784) *(784,10)  + = (num, 10)

下面进行简单的梯度下降，开始迭代。

num_steps = 801def accuracy(predictions, labels):    ''' predictions = [0.8,0,0,0,0.1,0,0,0.1,0,0]        labels = [1,0,0,0,0,0,0,0,0,0]    '''    return (100.0 * np.sum(np.argmax(predictions, 1) == np.argmax(labels, 1)) / predictions.shape[0])with tf.Session(graph=graph) as session:    # This is a one-time operation which ensures the parameters get initialized as    # we described in the graph:     #  random weights for the matrix, zeros for the biases.     tf.initialize_all_variables().run() # initialize    print('Initialized')    for step in xrange(num_steps):        # Run the computations. We tell .run() that we want to run the optimizer,        # and get the loss value and the training predictions returned as numpy        # arrays.         _, l, predictions = session.run([optimizer, loss, train_prediction]) # using train_prediction to train and return prediction in train data set        if (step % 100 == 0):            print('Loss at step %d: %f' % (step, l))            print('Training accuracy: %.1f%%' % accuracy(            predictions, train_labels[:train_subset, :]))            # Calling .eval() on valid_prediction is basically like calling run(), but            # just to get that one numpy array. Note that it recomputes all its graph            # dependencies.            print('Validation accuracy: %.1f%%' % accuracy(valid_prediction.eval(), valid_labels))            print('Test accuracy: %.1f%%' % accuracy(test_prediction.eval(), test_labels))

运行结果如下：

Initialized
Loss at step 0: 17.639723
Training accuracy: 8.9%
Validation accuracy: 11.4%
Loss at step 100: 2.268863
Training accuracy: 71.8%
Validation accuracy: 70.8%
Loss at step 200: 1.818829
Training accuracy: 74.9%
Validation accuracy: 73.6%
Loss at step 300: 1.580101
Training accuracy: 76.5%
Validation accuracy: 74.5%
Loss at step 400: 1.419103
Training accuracy: 77.1%
Validation accuracy: 75.1%
Loss at step 500: 1.299344
Training accuracy: 77.7%
Validation accuracy: 75.3%
Loss at step 600: 1.205005
Training accuracy: 78.3%
Validation accuracy: 75.3%
Loss at step 700: 1.127984
Training accuracy: 78.8%
Validation accuracy: 75.5%
Loss at step 800: 1.063572
Training accuracy: 79.3%
Validation accuracy: 75.7%
Test accuracy: 82.6%

之后，我们可以用更快的优化算法，随机梯度算法进行训练。

graph的定义与之前类似，不同的是我们的训练数据是一小批一小批的。

所以要在运行session.run()时并导入小批量数据之前定义占位量(placeholder).。

batch_size = 128graph = tf.Graph()with graph.as_default():    # Input data. For the training data, we use a placeholder that will be fed ----------------------------------------1    # at run time with a training minibatch.    #  相当于开辟空间    tf_train_dataset = tf.placeholder(tf.float32, shape=(batch_size, image_size * image_size))    tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))        tf_valid_dataset = tf.constant(valid_dataset)    tf_test_dataset = tf.constant(test_dataset)      # Variables.                                                                                                       ------------------------------------------2    weights = tf.Variable(tf.truncated_normal([image_size * image_size, num_labels]))    biases = tf.Variable(tf.zeros([num_labels]))      # Training computation.                                                                                  ------------------------------------------3    logits = tf.matmul(tf_train_dataset, weights) + biases    loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, tf_train_labels))    # Optimizer.                                                                                                       -------------------------------------------4    optimizer = tf.train.GradientDescentOptimizer(0.5).minimize(loss)      # Predictions for the training, validation, and test data.                             --------------------------------------------5    train_prediction = tf.nn.softmax(logits)    valid_prediction = tf.nn.softmax(tf.matmul(tf_valid_dataset, weights) + biases)    test_prediction = tf.nn.softmax(tf.matmul(tf_test_dataset, weights) + biases)

下面是对应的训练操作代码：

num_steps = 3001with tf.Session(graph=graph) as session:    tf.initialize_all_variables().run()    print("Initialized")    for step in range(num_steps):    # Pick an offset within the training data, which has been randomized.    # Note: we could use better randomization across epochs.        offset = (step * batch_size) % (train_labels.shape[0] - batch_size)    # Generate a minibatch.        batch_data = train_dataset[offset:(offset + batch_size), :]        batch_labels = train_labels[offset:(offset + batch_size), :]    # Prepare a dictionary telling the session where to feed the minibatch.    # The key of the dictionary is the placeholder node of the graph to be fed,    # and the value is the numpy array to feed to it.        #  传递值到tf的命名空间        feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}        _, l, predictions = session.run([optimizer, loss, train_prediction], feed_dict=feed_dict)        if (step % 500 == 0):            print("Minibatch loss at step %d: %f" % (step, l))            print("Minibatch accuracy: %.1f%%" % accuracy(predictions, batch_labels))            print("Validation accuracy: %.1f%%" % accuracy(valid_prediction.eval(), valid_labels))    print("Test accuracy: %.1f%%" % accuracy(test_prediction.eval(), test_labels))

运行结果如下：

InitializedMinibatch loss at step 0: 16.076256Minibatch accuracy: 14.1%Validation accuracy: 17.9%Minibatch loss at step 500: 1.690020Minibatch accuracy: 72.7%Validation accuracy: 75.1%Minibatch loss at step 1000: 1.430756Minibatch accuracy: 77.3%Validation accuracy: 76.1%Minibatch loss at step 1500: 1.065795Minibatch accuracy: 81.2%Validation accuracy: 77.0%Minibatch loss at step 2000: 1.248749Minibatch accuracy: 75.0%Validation accuracy: 77.3%Minibatch loss at step 2500: 0.934266Minibatch accuracy: 81.2%Validation accuracy: 78.1%Minibatch loss at step 3000: 1.047278Minibatch accuracy: 76.6%Validation accuracy: 78.4%Test accuracy: 85.4%

现在我们加入一层1024节点的隐含层，并使用rectified linear units神经单元，随后利用S.G.D进行训练看看效果。

当然结果肯定会有所提升。

batch_size = 128hiden_layer_node_num = 1024graph = tf.Graph()with graph.as_default():    # input                                                                                                             -----------------------------------------1    tf_train_dataset = tf.placeholder(tf.float32, shape=(batch_size, image_size * image_size))    tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))        tf_valid_dataset = tf.constant(valid_dataset)    tf_test_dataset = tf.constant(test_dataset)      # Variables.                                                                                                       ------------------------------------------2    weights1 = tf.Variable(tf.truncated_normal([image_size * image_size, hiden_layer_node_num]))    biases1 = tf.Variable(tf.zeros([hiden_layer_node_num]))        # input layer output (batch_size, hiden_layer_node_num)    weights2 = tf.Variable(tf.truncated_normal([hiden_layer_node_num, num_labels]))    biases2 = tf.Variable(tf.zeros([num_labels]))          # Training computation.                                                                                  ------------------------------------------3    logits = tf.matmul(tf.nn.relu(tf.matmul(tf_train_dataset, weights1) + biases1), weights2) + biases2    loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, tf_train_labels))    # Optimizer.                                                                                                       -------------------------------------------4    optimizer = tf.train.GradientDescentOptimizer(0.5).minimize(loss)      # Predictions for the training, validation, and test data.                            --------------------------------------------5    train_prediction = tf.nn.softmax(logits)    valid_prediction = tf.nn.softmax(tf.matmul(tf.nn.relu(tf.matmul(tf_valid_dataset, weights1) + biases1), weights2) + biases2)    test_prediction = tf.nn.softmax(tf.matmul(tf.nn.relu(tf.matmul(tf_test_dataset, weights1) + biases1), weights2) + biases2)num_steps = 3001with tf.Session(graph=graph) as session:    tf.initialize_all_variables().run()    print("Initialized")    for step in range(num_steps):    # Pick an offset within the training data, which has been randomized.    # Note: we could use better randomization across epochs.        offset = (step * batch_size) % (train_labels.shape[0] - batch_size)    # Generate a minibatch.        batch_data = train_dataset[offset:(offset + batch_size), :]        batch_labels = train_labels[offset:(offset + batch_size), :]    # Prepare a dictionary telling the session where to feed the minibatch.    # The key of the dictionary is the placeholder node of the graph to be fed,    # and the value is the numpy array to feed to it.        #  传递值到tf的命名空间        feed_dict = {tf_train_dataset : batch_data, tf_train_labels : batch_labels}        _, l, predictions = session.run([optimizer, loss, train_prediction], feed_dict=feed_dict)        if (step % 500 == 0):            print("Minibatch loss at step %d: %f" % (step, l))            print("Minibatch accuracy: %.1f%%" % accuracy(predictions, batch_labels))            print("Validation accuracy: %.1f%%" % accuracy(valid_prediction.eval(), valid_labels))    print("Test accuracy: %.1f%%" % accuracy(test_prediction.eval(), test_labels))

运行结果如下：

InitializedMinibatch loss at step 0: 379.534973Minibatch accuracy: 8.6%Validation accuracy: 21.7%Minibatch loss at step 500: 12.951815Minibatch accuracy: 86.7%Validation accuracy: 80.8%Minibatch loss at step 1000: 9.569818Minibatch accuracy: 82.8%Validation accuracy: 80.9%Minibatch loss at step 1500: 7.165316Minibatch accuracy: 84.4%Validation accuracy: 78.8%Minibatch loss at step 2000: 10.387121Minibatch accuracy: 78.9%Validation accuracy: 80.8%Minibatch loss at step 2500: 3.324355Minibatch accuracy: 80.5%Validation accuracy: 80.8%Minibatch loss at step 3000: 4.396149Minibatch accuracy: 89.8%Validation accuracy: 81.3%Test accuracy: 88.9%

测试结果正确率达到了88.9%

这样一个简单的神经网络就搭建好了。