将CNN中的dropout替换为L2正则化

Author: qyz777

.gitignore

@@ -1,4 +1,6 @@
 .DS_Store
+pb/
+tflite/
 data/cnews
 data/thucnews
 __pycache__

cnn_model.py

@@ -1,4 +1,5 @@
 # coding: utf-8
+from functools import partial
 
 import tensorflow as tf
 
@@ -24,6 +25,8 @@ class TCNNConfig(object):
     print_per_batch = 100  # 每多少轮输出一次结果
     save_per_batch = 10  # 每多少轮存入tensorboard
 
+    scale = 0.01
+
 
 class TextCNN(object):
     """文本分类，CNN模型"""
@@ -40,6 +43,11 @@ def __init__(self, config):
 
     def cnn(self):
         """CNN模型"""
+        my_dense_layer = partial(
+            tf.layers.dense, activation=tf.nn.relu,
+            # 在这里传入了L2正则化函数，并在函数中传入正则化系数。
+            kernel_regularizer=tf.contrib.layers.l2_regularizer(self.config.scale)
+        )
         # 词向量映射
         with tf.device('/cpu:0'):
             embedding = tf.get_variable('embedding', [self.config.vocab_size, self.config.embedding_dim])
@@ -52,10 +60,11 @@ def cnn(self):
             gmp = tf.reduce_max(conv, reduction_indices=[1], name='gmp')
 
         with tf.name_scope("score"):
-            # 全连接层，后面接dropout以及relu激活
-            fc = tf.layers.dense(gmp, self.config.hidden_dim, name='fc1')
-            fc = tf.contrib.layers.dropout(fc, self.keep_prob)
-            fc = tf.nn.relu(fc)
+            # 全连接层
+            fc = my_dense_layer(gmp, self.config.hidden_dim, name='fc1')
+            # fc = tf.layers.dense(gmp, self.config.hidden_dim, name='fc1')
+            # fc = tf.contrib.layers.dropout(fc, self.keep_prob)
+            # fc = tf.nn.relu(fc)
 
             # 分类器
             self.logits = tf.layers.dense(fc, self.config.num_classes, name='fc2')
@@ -64,7 +73,9 @@ def cnn(self):
         with tf.name_scope("optimize"):
             # 损失函数，交叉熵
             cross_entropy = tf.nn.softmax_cross_entropy_with_logits(logits=self.logits, labels=self.input_y)
-            self.loss = tf.reduce_mean(cross_entropy)
+            reg_losses = tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)
+            self.loss = tf.add_n([tf.reduce_mean(cross_entropy)] + reg_losses)
+            # self.loss = tf.reduce_mean(cross_entropy)
             # 优化器
             self.optim = tf.train.AdamOptimizer(learning_rate=self.config.learning_rate).minimize(self.loss)
 

freeze.py

@@ -0,0 +1,52 @@
+import tensorflow as tf
+
+version = 1.0
+output_graph = "./pb/model_{}.pb".format(version)
+output_tflite_model = "./tflite/model_{}.tflite".format(version)
+
+
+def freeze_graph(input_checkpoint):
+    """
+    :param input_checkpoint:
+    :return:
+    """
+    # checkpoint = tf.train.get_checkpoint_state(model_folder) #检查目录下ckpt文件状态是否可用
+    # input_checkpoint = checkpoint.model_checkpoint_path #得ckpt文件路径
+
+    # 指定输出的节点名称,该节点名称必须是原模型中存在的节点
+    output_node_names = "score/fc2/BiasAdd"
+    saver = tf.train.import_meta_graph(input_checkpoint + '.meta', clear_devices=True)
+
+    with tf.Session() as sess:
+        saver.restore(sess, input_checkpoint)  # 恢复图并得到数据
+        output_graph_def = tf.graph_util.convert_variables_to_constants(  # 模型持久化，将变量值固定
+            sess=sess,
+            input_graph_def=sess.graph_def,  # 等于:sess.graph_def
+            output_node_names=output_node_names.split(",")
+        )  # 如果有多个输出节点，以逗号隔开
+
+        with tf.gfile.GFile(output_graph, "wb") as f:  # 保存模型
+            f.write(output_graph_def.SerializeToString())  # 序列化输出
+
+
+def convert_to_tflite():
+    input_tensors = [
+        "input_x"
+    ]
+    output_tensors = [
+        "score/fc2/BiasAdd"
+    ]
+    converter = tf.lite.TFLiteConverter.from_frozen_graph(
+        output_graph,
+        input_tensors,
+        output_tensors)
+    converter.target_ops = [tf.lite.OpsSet.TFLITE_BUILTINS,
+                            tf.lite.OpsSet.SELECT_TF_OPS]
+    tflite_model = converter.convert()
+    open(output_tflite_model, "wb").write(tflite_model)
+
+
+if __name__ == "__main__":
+    # freeze_graph("./checkpoints/textcnn/best_validation")
+    convert_to_tflite()
+