更新lessonOne和lessonThree的代码

lessonOne/imgClassifierWeb/cnnModel.py

@@ -45,52 +45,85 @@ def create_conv_layer(input_data, filter_size, num_filters):
             conv_layer = tf.nn.conv2d(input=input_data,
                                               filter=filters,
                                               strides=[1, 1, 1, 1],
-                                              padding="VALID")
+                                              padding="SAME")
             print("Size of conv result : ", conv_layer.shape)
 
             return filters, conv_layer
 
         def create_CNN(input_data, num_classes, keep_prop):
-            filters1, conv_layer1 = create_conv_layer(input_data=input_data, filter_size=5, num_filters=4)
+            filters1, conv_layer1 = create_conv_layer(input_data=input_data, filter_size=3, num_filters=64)
             relu_layer1 = tf.nn.relu(conv_layer1)
             print("Size of relu1 result : ", relu_layer1.shape)
             max_pooling_layer1 = tf.nn.max_pool(value=relu_layer1,
                                                         ksize=[1, 2, 2, 1],
                                                         strides=[1, 1, 1, 1],
-                                                        padding="VALID")
+                                                        padding="SAME")
             print("Size of maxpool1 result : ", max_pooling_layer1.shape)
 
-            filters2, conv_layer2 = create_conv_layer(input_data=max_pooling_layer1, filter_size=7, num_filters=3)
+            filters2, conv_layer2 = create_conv_layer(input_data=max_pooling_layer1, filter_size=3, num_filters=64)
             relu_layer2 = tf.nn.relu(conv_layer2)
             print("Size of relu2 result : ", relu_layer2.shape)
             max_pooling_layer2 = tf.nn.max_pool(value=relu_layer2,
                                                         ksize=[1, 2, 2, 1],
-                                                        strides=[1, 1, 1, 1],
-                                                        padding="VALID")
+                                                        strides=[1, 2, 2, 1],
+                                                        padding="SAME")
             print("Size of maxpool2 result : ", max_pooling_layer2.shape)
 
-            # Conv layer with 2 filters and a filter sisze of 5x5.
-            filters3, conv_layer3 = create_conv_layer(input_data=max_pooling_layer2, filter_size=5, num_filters=2)
+
+            filters3, conv_layer3 = create_conv_layer(input_data=max_pooling_layer2, filter_size=3, num_filters=64)
             relu_layer3 = tf.nn.relu(conv_layer3)
             print("Size of relu3 result : ", relu_layer3.shape)
             max_pooling_layer3 = tf.nn.max_pool(value=relu_layer3,
                                                         ksize=[1, 2, 2, 1],
-                                                        strides=[1, 1, 1, 1],
+                                                        strides=[1, 2, 2, 1],
                                                         padding="VALID")
             print("Size of maxpool3 result : ", max_pooling_layer3.shape)
 
-            # Adding dropout layer before the fully connected layers to avoid overfitting.
-            flattened_layer = dropout_flatten_layer(previous_layer=max_pooling_layer3, keep_prop=keep_prop)
 
-            # First fully connected (FC) layer. It accepts the result of the dropout layer after being flattened (1D).
+
+            filters4, conv_layer4 = create_conv_layer(input_data=max_pooling_layer3, filter_size=3, num_filters=128)
+            relu_layer4 = tf.nn.relu(conv_layer4)
+            print("Size of relu4 result : ", relu_layer4.shape)
+            max_pooling_layer4 = tf.nn.max_pool(value=relu_layer4,
+                                                        ksize=[1, 2, 2, 1],
+                                                        strides=[1, 2, 2, 1],
+                                                        padding="SAME")
+            print("Size of maxpool4 result : ", max_pooling_layer4.shape)
+
+
+
+            filters5, conv_layer5 = create_conv_layer(input_data=max_pooling_layer4, filter_size=3, num_filters=128)
+            relu_layer5 = tf.nn.relu(conv_layer5)
+            print("Size of relu5 result : ", relu_layer5.shape)
+            max_pooling_layer5 = tf.nn.max_pool(value=relu_layer5,
+                                                        ksize=[1, 2, 2, 1],
+                                                        strides=[1, 2, 2, 1],
+                                                        padding="SAME")
+            print("Size of maxpool5 result : ", max_pooling_layer5.shape)
+
+
+
+            filters6, conv_layer6 = create_conv_layer(input_data=max_pooling_layer5, filter_size=3, num_filters=128)
+            relu_layer6 = tf.nn.relu(conv_layer6)
+            print("Size of relu6 result : ", relu_layer6.shape)
+            max_pooling_layer6 = tf.nn.max_pool(value=relu_layer6,
+                                                        ksize=[1, 2, 2, 1],
+                                                        strides=[1, 2, 2, 1],
+                                                        padding="SAME")
+            print("Size of maxpool6 result : ", max_pooling_layer6.shape)
+
+            # 将输出维度降为一维，并增加dropout防止过拟合
+            flattened_layer = dropout_flatten_layer(previous_layer=max_pooling_layer6, keep_prop=keep_prop)
+
+            # 全连接网络+dropout
             fc_resultl = fc_layer(flattened_layer=flattened_layer,
                                   num_inputs=flattened_layer.get_shape()[1:].num_elements(),
                                   num_outputs=200)
-            # Second fully connected layer accepting the output of the previous fully connected layer. Number of outputs is equal to the number of dataset classes.
+            # 全连接网络+dropout，这里的网络输出的数量要和标注的数量一致
             fc_result2 = fc_layer(flattened_layer=fc_resultl, num_inputs=fc_resultl.get_shape()[1:].num_elements(),
                                   num_outputs=num_classes)
             print("Fully connected layer results : ", fc_result2)
-            return fc_result2  # Returning the result of the last FC layer.
+            return fc_result2  
 
         def dropout_flatten_layer(previous_layer, keep_prop):
 

lessonOne/imgClassifierWeb/config.ini

@@ -14,11 +14,11 @@ num_channels = 3
 num_files=5
 images_per_file=10000
 max_gradient_norm=5
+percent=1
 
 [floats]
 learning_rate = 0.01
 keeps=0.5
-learning_rate_decay_factor = 0.9
-percent=1
+learning_rate_decay_factor = 0.5
 end_learning_rate=0.0
 

lessonOne/imgClassifierWeb/execute.py

@@ -57,11 +57,8 @@ def create_model(session,forward_only):
     if ckpt and ckpt.model_checkpoint_path:
         print("Reading model parameters from %s" % ckpt.model_checkpoint_path)
         model.saver.restore(session, ckpt.model_checkpoint_path)
-        session.run(tf.global_variables_initializer())
         graph = tf.get_default_graph()
-
         return model,graph
-
     else:
         print("Created model with fresh parameters.")
         session.run(tf.global_variables_initializer())
@@ -130,8 +127,8 @@ def train():
 
             # 达到一个训练模型保存点后，将模型保存下来，并打印出这个保存点的平均准确率
             if current_step % gConfig['steps_per_checkpoint'] == 0:
-                #如果超过三次预测正确率没有升高则改变学习率
-                if len(previous_correct) > 2 and accuracy < min(previous_correct[-3:]):
+                #如果超过5次预测正确率没有升高则改变学习率
+                if len(previous_correct) > 2 and accuracy < min(previous_correct[-5:]):
                     sess.run(model.learning_rate_decay_op)
                 previous_correct.append(accuracy)
                 checkpoint_path = os.path.join(gConfig['working_directory'], "cnn.ckpt")

lessonThree/Anti-Fraud-App/VAE.py

@@ -4,17 +4,14 @@
 import matplotlib.pyplot as plt
 from tensorflow.examples.tutorials.mnist import input_data
 import getConfig
+import sys
+
+import os
 gConfig={}
 gConfig=getConfig.get_config(config_file='config.ini')
 #定义训练数据的维度
 seq_len=gConfig['seqlen']
 
-#class vaeModel(object):
-
-   # def __init__(self, learning_rate,learning_rate_decay_factor):
-   #     self.learning_rate=tf.Variable(float(learning_rate), trainable=False)
-   #     self.learning_rate_decay_op = self.learning_rate.assign(self.learning_rate * learning_rate_decay_factor)
-    #    self.global_step = tf.Variable(0, trainable=False)
 
 #读取数据
 def read_data(source_file):
@@ -76,27 +73,63 @@ def decoder(sampled_z, keep_prob):
           decoder_set = tf.reshape(y_out, shape=[-1, seq_len, 1])
           return decoder_set
 #定义计算loss以及进行优化器优化的一系列tensor
+global_step = tf.Variable(0, trainable=False)
 sampled, mn, sd = encoder(X_in, keep_prob)
 dec = decoder(sampled, keep_prob)
 unreshaped = tf.reshape(dec, [-1, seq_len*1])
 img_loss = tf.reduce_sum(tf.squared_difference(unreshaped, Y_flat), 1)
 latent_loss = -0.5 * tf.reduce_sum(1.0 + 2.0 * sd - tf.square(mn) - tf.exp(2.0 * sd), 1)
 dst_loss=img_loss+latent_loss
 loss = tf.reduce_mean(img_loss + latent_loss)
-optimizer = tf.train.AdamOptimizer(gConfig['learning_rate']).minimize(loss)
+optimizer = tf.train.AdamOptimizer(gConfig['learning_rate']).minimize(loss,global_step=global_step)
 sess = tf.Session()
 sess.run(tf.global_variables_initializer())
-
+saver = tf.train.Saver(tf.all_variables())
+
+
+def vae_train():       
 
 #开始vae的训练
-for i in range(gConfig['vae_steps']):
-    batch=dataset
-    sess.run(optimizer, feed_dict = {X_in: batch, Y: batch, keep_prob: 0.8})
-    if not i % 200:
+  for i in range(gConfig['vae_steps']):
+     batch=dataset
+#通过循环训练来通过优化器进行BP计算，进而更新参数使网络进行拟合，这里使用的是fullbatch模式
+     sess.run(optimizer, feed_dict = {X_in: batch, Y: batch, keep_prob: 0.8})
+#根据checkpoint点对网络的收敛情况进行监测并保存下来模型     
+     if not i % 200:
         ls, d, i_ls, d_ls, mu, sampled_data = sess.run([loss, dec, img_loss, dst_loss, mn, sampled], feed_dict = {X_in: batch, Y: batch, keep_prob: 1.0})
-
         print(i, ls, np.mean(i_ls), np.mean(d_ls))
+        checkpoint_path = os.path.join(gConfig['working_directory'], "vae.ckpt")
+        saver=tf.train.Saver()
+        saver.save(sess,checkpoint_path,global_step=global_step)
+#最后将训练集的所有的数据的encoder数据保存下来，用于接下来的kmeans训练
+  sampled_data = sess.run(sampled, feed_dict = {X_in: batch, Y: batch, keep_prob: 1.0})
+  sampled_data=pd.DataFrame(sampled_data)
+  sampled_data.to_csv(gConfig['sampled_path'])
+
+#定义
+def vae_encoder(sess,input_data):
+
+
+  ckpt=tf.train.get_checkpoint_state(gConfig['working_directory'])
+  if ckpt and ckpt.model_checkpoint_path:
+        print("Reading model parameters from %s" % ckpt.model_checkpoint_path)
+        saver.restore(sess, ckpt.model_checkpoint_path)
+        graph = tf.get_default_graph()
+
+  sampled_data = sess.run(sampled, feed_dict = {X_in: input_data, Y: input_data, keep_prob: 1.0})
+
+  return sampled_data
+
+
+if __name__=='__main__':
+  if len(sys.argv) - 1:
+     gConfig = getConfig(sys.argv[1])
+  else:
+        # get configuration from config.ini
+     gConfig = getConfig.get_config()
+  if gConfig['mode']=='train':
+        vae_train()
+  elif gConfig['mode']=='server':
+     print('Sever Usage:python3 app.py')
+
 
-#保存训练的encode的结果，就是要进行特征压缩后的特征
-sampled_data=pd.DataFrame(sampled_data)
-sampled_data.to_csv(gConfig['sampled_path'])

lessonThree/Anti-Fraud-App/app.py

@@ -12,8 +12,8 @@
 import hashlib
 import threading
 import execute
+import VAE
 
-#
 app = Flask(__name__) 
 #路由注解，我们这里使用的是path的形式进行传参
 #示例url:http://0.0.0.0:8088/predict/1/2/3/4/5/6/7/8/9/10/11/12 这里的1...12换成需要进行聚类的值就可以了
@@ -26,6 +26,7 @@ def predict(a,b,c,d,e,f,g,h,i,j,k,l):
 	lines=range(k)
 	lines=[int(i) for i in line]
 	lines=[lines]
+	lines=VAE.vae_encoder(VAE.sess,lines)
 	predict_result=execute.predicts(lines)
 
 	return jsonify( { 'result of cluster': str(predict_result) } )

lessonThree/Anti-Fraud-App/config.ini

@@ -9,7 +9,7 @@ steps=10
 #VAE训练步数
 vae_steps=300
 #需要聚类的序列的长度,如果要将vae和clustering串起来用这个值要和enc_out_len保持一致
-seqlen=12
+seqlen=8
 #vae encoder输出的中间向量的长度
 encoutlen=8
 
@@ -25,7 +25,7 @@ keeps=0.5
 [strings]
 mode = train
 working_directory=working_directory/
-input_file=train_data/train_data.csv
+input_file=train_data/sampled_data.csv
 #这里model的文件路径，要根据kmeansMode中的文件夹的来设置
 model_path=kmeansMode/1542258970
 sampled_path=train_data/sampled_data.csv

lessonThree/Anti-Fraud-App/readme.md

@@ -0,0 +1 @@
+最新的代码将VAE和kmeans进行了结合，如果需要使用的话，需要先执行python3 VAE.py 然后执行python3 execute.py 最后再执行python3.py