增加mnist数字rnn识别

mnistRnn/mnist_rnn.py

@@ -0,0 +1,38 @@
+import tensorflow as tf
+import numpy as np
+from mnistRnn.rnn_train import rnn_graph
+from mnistRnn.settings import mnist, width, height, rnn_size, out_size
+
+
+def mnist2text(image_list, height, width, rnn_size, out_size):
+    '''
+    mnist数字向量转为文本
+    :param image_list:
+    :param height:
+    :param width:
+    :param rnn_size:
+    :param out_size:
+    :return:
+    '''
+    x = tf.placeholder(tf.float32, [None, height, width])
+    y_conv = rnn_graph(x, rnn_size, out_size, width, height)
+    saver = tf.train.Saver()
+    with tf.Session() as sess:
+        saver.restore(sess, tf.train.latest_checkpoint('.'))
+        predict = tf.argmax(y_conv, 1)
+        vector_list = sess.run(predict, feed_dict={x: image_list})
+        vector_list = vector_list.tolist()
+        return vector_list
+
+
+if __name__ == '__main__':
+    batch_x_test = mnist.test.images
+    batch_x_test = batch_x_test.reshape([-1, height, width])
+
+    batch_y_test = mnist.test.labels
+    batch_y_test = list(np.argmax(batch_y_test, 1))
+
+    pre_y = list(mnist2text(batch_x_test, height, width, rnn_size, out_size))
+    for text in batch_y_test:
+        print('Label:', text, ' Predict:', pre_y[batch_y_test.index(text)])
+

mnistRnn/rnn_train.py

@@ -0,0 +1,124 @@
+import os
+import tensorflow as tf
+from datetime import datetime
+from mnistRnn.settings import mnist, batch_size, width, height, rnn_size, out_size
+
+
+def weight_variable(shape, w_alpha=0.01):
+    '''
+    增加噪音，随机生成权重
+    :param shape:
+    :param w_alpha:
+    :return:
+    '''
+    initial = w_alpha * tf.random_normal(shape)
+    return tf.Variable(initial)
+
+
+def bias_variable(shape, b_alpha=0.1):
+    '''
+    增加噪音，随机生成偏置项
+    :param shape:
+    :param b_alpha:
+    :return:
+    '''
+    initial = b_alpha * tf.random_normal(shape)
+    return tf.Variable(initial)
+
+
+def rnn_graph(x, rnn_size, out_size, width, height):
+    '''
+    循环神经网络计算图
+    :param x:
+    :param rnn_size:
+    :param out_size:
+    :param width:
+    :param height:
+    :return:
+    '''
+    # 权重及偏置
+    w = weight_variable([rnn_size, out_size])
+    b = bias_variable([out_size])
+    # LSTM
+    lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(rnn_size)
+    # 原排列[0,1,2]transpose为[1,0,2]代表前两维装置，如shape=(1,2,3)转为shape=(2,1,3)
+    # 这里的实际意义是把所有图像向量的相同行号向量转到一起，如x1的第一行与x2的第一行
+    x = tf.transpose(x, [1, 0, 2])
+    # reshape -1 代表自适应，这里按照图像每一列的长度为reshape后的列长度
+    x = tf.reshape(x, [-1, width])
+    # split默任在第一维即0 dimension进行分割，分割成height份，这里实际指把所有图片向量按对应行号进行重组
+    x = tf.split(x, height)
+    # 这里RNN会有与输入层相同数量的输出层，我们只需要最后一个输出
+    outputs, status = tf.nn.static_rnn(lstm_cell, x, dtype=tf.float32)
+    y_conv = tf.add(tf.matmul(outputs[-1], w), b)
+
+    return y_conv
+
+
+def optimize_graph(y, y_conv):
+    '''
+    优化计算图
+    :param y:
+    :param y_conv:
+    :return:
+    '''
+    loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=y_conv, labels=y))
+    optimizer = tf.train.AdamOptimizer().minimize(loss)
+    return optimizer
+
+
+def accuracy_graph(y, y_conv):
+    '''
+    偏差计算图
+    :param y:
+    :param y_conv:
+    :return:
+    '''
+    correct = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y, 1))
+    accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))
+    return accuracy
+
+
+def train():
+    '''
+    rnn训练
+    :return:
+    '''
+    # 按照图片大小申请占位符
+    x = tf.placeholder(tf.float32, [None, height, width])
+    y = tf.placeholder(tf.float32)
+    # rnn模型
+    y_conv = rnn_graph(x, rnn_size, out_size, width, height)
+    # 最优化
+    optimizer = optimize_graph(y, y_conv)
+    # 偏差
+    accuracy = accuracy_graph(y, y_conv)
+
+    # 启动会话.开始训练
+    saver = tf.train.Saver()
+    session = tf.Session()
+    session.run(tf.global_variables_initializer())
+    step = 0
+    acc_rate = 0.98
+    while 1:
+        batch_x, batch_y = mnist.train.next_batch(batch_size)
+        batch_x = batch_x.reshape([batch_size, height, width])
+        session.run(optimizer, feed_dict={x: batch_x, y: batch_y})
+        # 每训练10次测试一次
+        if step % 10 == 0:
+            batch_x_test = mnist.test.images
+            batch_y_test = mnist.test.labels
+            batch_x_test = batch_x_test.reshape([-1, height, width])
+            acc = session.run(accuracy, feed_dict={x: batch_x_test, y: batch_y_test})
+            print(datetime.now().strftime('%c'), ' step:', step, ' accuracy:', acc)
+            # 偏差满足要求，保存模型
+            if acc >= acc_rate:
+                model_path = os.getcwd() + os.sep + str(acc_rate) + "mnist.model"
+                saver.save(session, model_path, global_step=step)
+                break
+        step += 1
+    session.close()
+
+
+if __name__ == '__main__':
+    train()

mnistRnn/settings.py

@@ -0,0 +1,13 @@
+from tensorflow.examples.tutorials.mnist import input_data
+
+# minst测试集
+mnist = input_data.read_data_sets('mnist/', one_hot=True)
+# 每次使用100条数据进行训练
+batch_size = 100
+# 图像向量
+width = 28
+height = 28
+# LSTM隐藏神经元数量
+rnn_size = 256
+# 输出层one-hot向量长度的
+out_size = 10