如何将GridSearchCV的参数值拆分后与评分导出为CSV文件

我懂你现在的困扰——网格搜索结果里的参数字典被一股脑塞进CSV的同一列，查看和分析都特别麻烦。咱们只需要把字典里的每个参数单独拆成一列就搞定啦！

方法一：针对固定参数的直接提取 #

如果你的param_grid里的参数是固定的（比如就batch_size和epochs），可以直接从每个参数字典里提取对应的值，和评分结果一起写入不同列：

import csv

# 假设你的参数网格定义
param_grid=dict(batch_size=[40, 60, 80], epochs=[1000, 2000])
grid = GridSearchCV(estimator=model, param_grid=param_grid, n_jobs=-1, cv=3)
grid_result = grid.fit(x_train, y_train, validation_data=(x_test, y_test), callbacks=[es])

# 提取结果数据
means = grid_result.cv_results_['mean_test_score']
stds = grid_result.cv_results_['std_test_score']
params = grid_result.cv_results_['params']  # 注意变量名统一，避免覆盖

exportfile='/Users/test.csv'
with open(exportfile, 'w', newline='') as file:
    writer = csv.writer(file)
    # 先写入清晰的表头
    writer.writerow(['mean_test_score', 'std_test_score', 'batch_size', 'epochs'])
    # 遍历每一组结果，拆分参数写入
    for mean, stdev, param in zip(means, stds, params):
        batch_size_val = param['batch_size']
        epochs_val = param['epochs']
        writer.writerow([mean, stdev, batch_size_val, epochs_val])

这样生成的CSV里，每个参数都会单独占一列，再也不会是挤在一起的字典格式了。

方法二：动态适配任意参数网格（推荐） #

如果以后你可能会给param_grid添加更多参数（比如learning_rate、dropout_rate），可以用动态方法自动适配，不用每次修改代码：

import csv

# 示例：带更多参数的网格
param_grid=dict(batch_size=[40, 60], epochs=[1000, 2000], learning_rate=[0.001, 0.01])
grid = GridSearchCV(estimator=model, param_grid=param_grid, n_jobs=-1, cv=3)
grid_result = grid.fit(x_train, y_train, validation_data=(x_test, y_test), callbacks=[es])

means = grid_result.cv_results_['mean_test_score']
stds = grid_result.cv_results_['std_test_score']
params = grid_result.cv_results_['params']

exportfile='/Users/test.csv'
with open(exportfile, 'w', newline='') as file:
    writer = csv.writer(file)
    # 动态生成表头：评分项 + 所有参数名
    header = ['mean_test_score', 'std_test_score'] + list(param_grid.keys())
    writer.writerow(header)
    # 按表头顺序提取每个参数的值，组合成一行写入
    for mean, stdev, param in zip(means, stds, params):
        param_values = [param[key] for key in param_grid.keys()]
        writer.writerow([mean, stdev] + param_values)

这个方法会自动读取param_grid里的所有参数名作为表头，然后按顺序提取每个参数的值，不管你以后加多少参数，代码都不用改，非常灵活。

小提醒 #

注意你原代码里的变量名问题：原代码里写的param = grid_result.cv_results_['params']，后面循环又用param作为变量，这会导致变量覆盖，容易出错，所以我改成了params来存储所有参数列表。

内容的提问来源于stack exchange，提问作者D.Luipers