1. 添加name

   
   with tf.name_scope("output"):
               self.out = tf.add(tf.matmul(concat_input, self.weights["concat_projection"]), self.weights["concat_bias"])
               if self.loss_type == "logloss":
                   self.out = tf.nn.sigmoid(self.out, name="predictlabel")

2. 训练模型，得到模型文件

3. 导出pd，新建model.py（跟模型在同一文件夹下）

from tensorflow.python import pywrap_tensorflow
import tensorflow as tf
from tensorflow.python.framework import graph_util

def getAllNodes(checkpoint_path):
    reader = pywrap_tensorflow.NewCheckpointReader(checkpoint_path)
    var_to_shape_map = reader.get_variable_to_shape_map()
    # Print tensor name and values
    for key in var_to_shape_map:
        print("tensor_name: ", key)
        #print(reader.get_tensor(key))

def freeze_graph(ckpt, output_graph):
    output_node_names = 'output/predictlabel'

    # saver = tf.train.import_meta_graph(ckpt+'.meta', clear_devices=True)
    saver = tf.compat.v1.train.import_meta_graph(ckpt+".meta", clear_devices=True)
    graph = tf.get_default_graph()
    input_graph_def = graph.as_graph_def()

    with tf.Session() as sess:
        saver.restore(sess, ckpt)
        output_graph_def = graph_util.convert_variables_to_constants(
            sess=sess,
            input_graph_def=input_graph_def,
            output_node_names=output_node_names.split(',')
        )
        with tf.gfile.GFile(output_graph, 'wb') as fw:
            fw.write(output_graph_def.SerializeToString())
        print('{} ops in the final graph.'.format(len(output_graph_def.node)))

if __name__ == '__main__':
    ckpt_path = 'model'

    getAllNodes(ckpt_path)

    output_graph_path = 'res.pb'
    freeze_graph(ckpt_path, output_graph_path)

有两个地方注意：
a： ckpt_path=“model” 是前缀，见图片。 b： output_node_names = 'output/predictlabel' 跟第一步设置的一样。

4. 运行此python文件，得到pd文件。