python 读取 json 配置文件

案例学习

import numpy as np
import json
import os.path
from collections import OrderedDict

# if config file not exists, use this default config
default_config = """{
  "input": {
    "nodes": 2,
    "functions": 
      [
        "lambda x: np.sin(128 * np.pi * x)",
        "lambda x: x"
      ],
    "length": 5000
  },
  "reservoir": {
    "start_node": 95,
    "end_node": 100,
    "step": 2,
    "degree_function": "lambda x: np.sqrt(x)",
    "sigma": 0.5,
    "bias": 1,
    "leakage_rate": 0.3,
    "regression_parameter": 1e-8
  },
  "output": {
    "nodes": 2
  },
  "training": {
    "init": 1000,
    "train": 3000,
    "test": 2000,
    "error": 1000
  }
}"""


class Reservoir:
    def __init__(self):
        config_file_name = 'reservoir.config'
        global config
        if os.path.isfile(config_file_name):
            with open(config_file_name) as config_file:
                config = json.load(config_file, object_pairs_hook=OrderedDict)
        else:
            config = json.loads(default_config, object_pairs_hook=OrderedDict)
            print('Config file not exist, using default config instead!')

        # Input layer
        self.M = config["input"]["nodes"]
        self.input_len = config["input"]["length"]
        self.input_func = []
        dataset = []
        for i in range(self.M):
            self.input_func.append(eval(config["input"]["functions"][i]))
            dataset.append(self.input_func[i](
                np.arange(self.input_len) / self.input_len))
        self.dataset = np.array(list(zip(*dataset))).T # shape = (M, length)

        # Reservoir layer
        self.start_node = config["reservoir"]["start_node"]
        self.N = self.start_node
        self.step = config["reservoir"]["step"]
        self.end_node = config["reservoir"]["end_node"]
        self.degree_func = eval(config["reservoir"]["degree_function"])
        self.D = self.degree_func(self.start_node)
        self.sigma = config["reservoir"]["sigma"]
        self.bias = config["reservoir"]["bias"]
        self.alpha = config["reservoir"]["leakage_rate"]
        self.beta = config["reservoir"]["regression_parameter"]

        # Output layer
        self.P = config["output"]["nodes"]

        # Training relevant
        self.init_len = config["training"]["init"]
        self.train_len = config["training"]["train"]
        self.test_len = config["training"]["test"]
        self.error_len = config["training"]["error"]

# Invoke automatically when exit, write the progress back to config file
def exit_handler():
    global config
    with open('reservoir.config', 'w') as config_file:
        config_file.write(json.dumps(config, indent = 4))
    print('Program finished! Current node = ' +
          str(config["reservoir"]["start_node"]))


if __name__ == '__main__':
    atexit.register(exit_handler)

比较有意思的是可以读取函数，然后生成数据

func_str = "lambda x: np.sin(x/np.pi)"
N = 20
data = eval(func_str)(np.arange(N)/N)

输出：

array([0.        , 0.01591482, 0.03182561, 0.04772834, 0.06361898,
       0.07949351, 0.0953479 , 0.11117814, 0.12698022, 0.14275013,
       0.15848389, 0.1741775 , 0.18982699, 0.2054284 , 0.22097778,
       0.23647118, 0.25190469, 0.26727439, 0.28257638, 0.29780681])