test tensorflow assign

assign.py

import os
import numpy as np
import tensorflow.compat.v1 as tf
def add_batch_dimension(data,batch=1):
    shape=data.shape
    dim=len(shape)
    if dim==1:
        width=int(len(data)/batch)
        return np.asarray(data[:]).reshape((batch,width))
    elif dim==2:
        height=int(shape[0]/batch)
        width=shape[1]
        return np.asarray(data[:]).reshape((batch,height,width))
    else:
        raise Exception("dimision error")
def print_info(sess,trainable_var,log_v=True):
    for param in trainable_var:
        print(param.name)
        if log_v:
            result=sess.run(param)
            print(result)
            #print(param.numpy()) #v2
class Model():
    def __init__(self,input_size,output_size,size,name):
        self.input= tf.placeholder(dtype=tf.float32, shape=(None, input_size))
        with tf.variable_scope(name_or_scope=name):
            self.W1 = tf.get_variable('W1', shape=[input_size, size], initializer=tf.glorot_uniform_initializer())
            self.W2 = tf.get_variable('W2', shape=[size, size], initializer=tf.glorot_uniform_initializer())
            self.W3 = tf.get_variable('W3', shape=[size, output_size], initializer=tf.glorot_uniform_initializer())
        
            self.b1 = tf.Variable(tf.zeros([1], dtype=tf.float32))
            self.b2 = tf.Variable(tf.zeros([1], dtype=tf.float32))
        
            self.L1 = tf.nn.relu(tf.matmul(self.input, self.W1) + self.b1)
            self.L2 = tf.nn.relu(tf.matmul(self.L1, self.W2) + self.b2)
            self.Q= tf.matmul(self.L2, self.W3)
        self.trainable_var = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, name)
class Model2():
    def __init__(self,input_size,output_size,size,name):
        self.input= tf.placeholder(dtype=tf.float32, shape=(None, input_size))
        with tf.variable_scope(name_or_scope=name):
            self.fc1 = tf.layers.dense(self.input, size, activation=tf.nn.relu)
            self.fc2 = tf.layers.dense(self.fc1, size, activation=tf.nn.relu)
            self.Q= tf.layers.dense(self.fc2, output_size, activation=tf.nn.relu)
        self.trainable_var = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, name)            
if __name__=='__main__':
    os.environ["CUDA_VISIBLE_DEVICES"]="0"
    tf.disable_v2_behavior()
    sess=tf.Session()
    model=Model(4,2,3,"eval")
    target=Model(4,2,3,"target")
    init = tf.initialize_all_variables() 
    sess.run(init)
    print_info(sess,model.trainable_var)
    print("\n\n")
    print_info(sess,target.trainable_var)
    print("\n\n")
    for i in range(len(model.trainable_var)):
        sess.run(target.trainable_var[i].assign(model.trainable_var[i]))
    print_info(sess,target.trainable_var)    
    sess.close()
    '''
    input_size=4
    name='test'
    input= tf.placeholder(dtype=tf.float32, shape=(None, input_size))
    with tf.variable_scope(name_or_scope=name):
        W1 = tf.get_variable('W1', shape=[input_size,3], initializer=tf.glorot_uniform_initializer())
    out=tf.matmul(input, W1)
    trainable_var = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, name)
    sess=tf.Session()
    init = tf.initialize_all_variables()
    sess.run(init)
    data=np.array([1,1,1,1])
    in1=add_batch_dimension(data)
    print_info(sess,trainable_var)
    print("\n\n")
    print(sess.run(out, feed_dict={input:in1}))
    sess.close()
    '''