Created by: jayxio

请问paddlepaddle如何加载指定变量？比如，如果我用如下程序训练了一个CNN

def mnist_cnn_model(img):
    """
    Mnist cnn model

    Args:
        img(Varaible): the input image to be recognized

    Returns:
        Variable: the label prediction
    """
    conv_pool_1 = fluid.nets.simple_img_conv_pool(
        input=img,
        num_filters=20,
        filter_size=5,
        pool_size=2,
        pool_stride=2,
        act='relu')

    conv_pool_2 = fluid.nets.simple_img_conv_pool(
        input=conv_pool_1,
        num_filters=50,
        filter_size=5,
        pool_size=2,
        pool_stride=2,
        act='relu')
    fc = fluid.layers.fc(input=conv_pool_2, size=50, act='relu')

    logits = fluid.layers.fc(input=fc, size=10, act=None)
    softmax = fluid.layers.softmax(input=logits)
    
    return softmax, logits

然后用：

fluid.io.save_params(
        exe, dirname='./mnist', main_program=fluid.default_main_program())

保存了这个default的program里面的所有变量。之后，当我又新建另一个program中，并向计算图添加了新的变量，比如我这里使用了create_parameter在self._loss_cw中：

with fluid.program_guard(main_program=attack_main_program, startup_program=attack_startup_program):
            img_0_1_placehold = fluid.layers.data(name='img_data_scaled',shape=[1,28,28],dtype="float32")
            target_placehold = fluid.layers.data(name='target',shape=[10],dtype="float32")
            shape_placehold = fluid.layers.data(name="shape", shape=[1], dtype="float32")
            #k_placehold = fluid.layers.data(name='k',shape=[1],dtype="float32")
            c_placehold = fluid.layers.data(name='c',shape=[1],dtype="float32")
            
            # get fluid.layer object from prebuilt program
            #img_placehold_from_prebuilt_program = attack_main_program.block(0).var(self.model._input_name)
            #softmax_from_prebuilt_program = attack_main_program.block(0).var(self.model._softmax_name)
            #logits_from_prebuilt_program = attack_main_program.block(0).var(self.model._predict_name)
            
            t0,t1,t2,t3,t4 = self._loss_cw(img_0_1_placehold,
                             target_placehold,
                             shape_placehold,
                             c_placehold)#,
                             #img_placehold_from_prebuilt_program,
                             #softmax_from_prebuilt_program,
                             #logits_from_prebuilt_program)
        
            # Adam optimizer as suggested in paper
            optimizer = fluid.optimizer.Adam(learning_rate=learning_rate)
            optimizer.minimize(t2,parameter_list=['parameter'])

如何调用在'./mnist'之前已保存的数据，并正确的只加载在mnist_cnn_model函数所创建的变量上？ 我想到的解决思路如下：

fluid.io.load_params(executor=exe, dirname=param_path,
                     main_program=prog)

但是这个API默认直接加载在对应program的所有变量上。但在TensorFlow里面可以通过：

saver = tf.train.Saver(tf.contrib.framework.get_variables()[1:-4])#[0:-1][1:-4]
saver.restore(self.sess,self.weights_file)

来选取想要加载的变量的。 请求大神指点！