json - 使用 JSON 文件中的值初始化多个类似乎会创建重复项

Question

我有一个名为 Dense 的主类，它有许多变量，我有另一个名为 Sequential 的类，它将这些层堆叠在一个数组中，并允许它们之间轻松通信。

在我的程序中的某个时刻，我最终将每个单独的 Dense 类的所有变量保存在一个 JSON 文件中以备后用。它看起来有点像这样:

    {
     "layer: 0":{
         .
         .
         .
         variables
         .
         .
         .
     },

     "layer: 1":{
         .
         .
         .
         variables
         .
         .
         .
     }
    }

然后，当我将这些 dicts 加载回我的图层时，当我稍后使用它们时，Sequential 类中最后一个图层的值似乎会复制到其他每个图层。

这是我在重新创建代码方面所能做的最好的事情

from abc import ABCMeta, abstractmethod
import os
import json

class Base_Layer(metaclass=ABCMeta):
    @abstractmethod
    def __init__(self, name:str, **kwargs) -> None:
        self.name = name


    @abstractmethod
    def get_summary(self) -> dict:
        return summary


    @abstractmethod
    def load(self, layer_data:dict) -> None:
        pass





class Dense(Base_Layer):
    def __init__(self, layer_shape:tuple):
        super().__init__("Dense")

        self.layer_shape = layer_shape


    def get_summary(self):
        summary = {
                    "name":self.name,
                    "layer_shape":self.layer_shape
                  }
        return summary


    def load(self, layer_data):
        self.name = layer_data["name"]
        self.layer_shape = tuple(layer_data["layer_shape"])





class Sequential:
    def __init__(self, model=[]):
        self.model = model


    def save(self, file_path):

        summaries = []
        for layer in self.model:
            summaries.append(layer.get_summary())


        saved_model = {}
        for i in range(len(summaries)):
            saved_model["layer: %s" % i] = {}
            for key in summaries[i]:
                saved_model["layer: %s" % i][key] = summaries[i][key]

        with open(file_path, "w+") as json_file:
            json_file.write(json.dumps(saved_model, indent=2))


    def load(self, file_path):
        layers = {
                    "Dense":Dense((0, 0)),
                 }

        # Try to open the file at file_path.
        try:
            with open(file_path, "r") as json_file:
                model_layers = json.loads(json_file.read())

                for i in range(len(model_layers)):
                    layer_data = model_layers["layer: %s" % i]
                    self.model.append(layers[layer_data["name"]])
                    self.model[-1].load(layer_data)

        # Gets called if the program can't find the file_path.
        except Exception as e:
            raise FileNotFoundError("Can't find file path %s. Try saving the model or enter a correct file path." % file_path)



seq_model = Sequential([
                            Dense((128, 784)),
                            Dense((128, 128)),
                            Dense((128, 128)),
                            Dense((32, 128)),
                            Dense((10, 32)),
                      ])

file_path = os.path.dirname(os.path.realpath(__file__)) + "/recreate_error_test.json"
seq_model.save(file_path)
seq_model.load(file_path)




for layer in seq_model.model:
    print(layer.layer_shape)

这实际上输出:
(128, 784)
(128, 128)
(128, 128)
(32, 128)
(10, 32)
(10, 32)
(10, 32)
(10, 32)
(10, 32)
(10, 32)

奇怪的是为什么它打印两次，前 7 个值是正确的，接下来的 7 个显示了我在实际代码中面临的问题。

对这个问题的任何帮助将不胜感激，谢谢。

Answer 1

问题是重用 layer[layer_data["name"]] ，这与添加到列表中的引用相同。 Dense.load() 不会重新创建一个新的 Dense 对象，它会填充它。

这里对加载函数进行了一些重写以查看元素创建。

import copy

(...)



    def load(self, file_path):
        layers = {
                    "Dense":Dense((0, 0)),
                 }

        # Try to open the file at file_path.
        try:
            with open(file_path, "r") as json_file:
                model_layers = json.loads(json_file.read())

                for i in range(len(model_layers)):
                    layer_data = model_layers["layer: %s" % i]
                    print("%s %s" %( i,layer_data["name"]))
                    # OK could be new_layer=Dense((0, 0)) since it recreate a new object
                    new_layer=copy.copy(layers[layer_data["name"]])
                    # NOK same reference is reused
                    # new_layer=layers[layer_data["name"]]
                    new_layer.load(layer_data)
                    self.model.append(new_layer)

        # Gets called if the program can't find the file_path.
        except Exception as e:
            raise FileNotFoundError("Can't find file path %s. Try saving the model or enter a correct file path." % file_path)

这是使用类名而不需要复制的另一种解决方案:

def load(self, file_path):
    layerclass = {
                "Dense":Dense,
             }

    # Try to open the file at file_path.
    try:
        with open(file_path, "r") as json_file:
            model_layers = json.loads(json_file.read())

            for i in range(len(model_layers)):
                layer_data = model_layers["layer: %s" % i]
                print("%s %s" %( i,layer_data["name"]))
                # requires that all layerclass have a __init__ constructor with layer_shape tuple.
                new_layer=layerclass[layer_data["name"]](layer_shape=(0,0))
                new_layer.load(layer_data)
                self.model.append(new_layer)

    # Gets called if the program can't find the file_path.
    except Exception as e:
        raise FileNotFoundError("Can't find file path %s. Try saving the model or enter a correct file path." % file_path)