下面程式碼的功能是先訓練一個簡單的模型，然後儲存模型，同時儲存到一個pb檔案當中，後續可以從pd檔案裡讀取權重值。

import tensorflow as tf
import numpy as np
import os
import h5py
import pickle
from tensorflow.python.framework import graph_util
from tensorflow.python.platform import gfile
#設定使用指定GPU
os.environ['CUDA_VISIBLE_DEVICES'] = '1'
#下面這段程式碼是在訓練好之後將所有的權重名字和權重值羅列出來，訓練的時候需要註釋掉
 

reader = tf.train.NewCheckpointReader('./model.ckpt-100')
variables = reader.get_variable_to_shape_map()
for ele in variables:
    print(ele)
    print(reader.get_tensor(ele))


x = tf.placeholder(tf.float32, shape=[None, 1])
y = 4 * x + 4

w = tf.Variable(tf.random_normal([1], -1, 1))
b = tf.Variable(tf.zeros([1
 
]))
y_predict = w * x + b


loss = tf.reduce_mean(tf.square(y - y_predict))
optimizer = tf.train.GradientDescentOptimizer(0.5)
train = optimizer.minimize(loss)

isTrain = False#設成True去訓練模型
train_steps = 100
checkpoint_steps = 50
checkpoint_dir = ''


saver = tf.train.Saver()  # defaults to saving all variables - in this case w and b
 

x_data = np.reshape(np.random.rand(10).astype(np.float32), (10, 1))

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    if isTrain:
        for i in xrange(train_steps):
            sess.run(train, feed_dict={x: x_data})
            if (i + 1) % checkpoint_steps == 0:
                saver.save(sess, checkpoint_dir + 'model.ckpt', global_step=i+1)
    else:
        ckpt = tf.train.get_checkpoint_state(checkpoint_dir)
        if ckpt and ckpt.model_checkpoint_path:
            saver.restore(sess, ckpt.model_checkpoint_path)
        else:
            pass     
        print(sess.run(w))
        print(sess.run(b))
        graph_def = tf.get_default_graph().as_graph_def()
        #通過修改下面的函式，個人覺得理論上能夠實現修改權重，但是很複雜，如果哪位有好辦法，歡迎指教
        output_graph_def = graph_util.convert_variables_to_constants(sess, graph_def, ['Variable'])
        with tf.gfile.FastGFile('./test.pb', 'wb') as f:
            f.write(output_graph_def.SerializeToString())


with tf.Session() as sess:
#對應最後一部分的寫，這裡能夠將對應的變數取出來
    with gfile.FastGFile('./test.pb', 'rb') as f:
        graph_def = tf.GraphDef()
        graph_def.ParseFromString(f.read())
    res = tf.import_graph_def(graph_def, return_elements=['Variable:0'])
    print(sess.run(res))
    print(sess.run(graph_def))