如何生成嵌套JSON预处理配置的所有参数迭代组合？

配置与需求 #

预处理配置文件示例：

"Preprocessing": {"BOW":{"ngram_range":[1,2], "max_features":[100, 200]},"RemoveStopWords": {"Parameter1": ["..."]}}

需求：基于该配置生成所有参数迭代组合，且保留BOW、RemoveStopWords的层级结构，传入Preprocessing对象。

期望输出 #

[
  {"BOW":{"ngram_range":1, "max_features":100}, "RemoveStopWords":{"Parameter1": "..."}},
  {"BOW":{"ngram_range":2, "max_features":100}, "RemoveStopWords":{"Parameter1": "..."}},
  {"BOW":{"ngram_range":1, "max_features":200}, "RemoveStopWords":{"Parameter1": "..."}},
  {"BOW":{"ngram_range":2, "max_features":200}, "RemoveStopWords":{"Parameter1": "..."}}
]

当前错误输出 #

[
  {"ngram_range":1, "max_features":100, "Parameter1":"..."},
  {"ngram_range":2, "max_features":200, "Parameter1":"..."},
  {"ngram_range":1, "max_features":100, "Parameter1":"..."},
  {"ngram_range":2, "max_features":200, "Parameter1":"..."}
]

修正后的Python代码 #

import itertools
import json

# 原始预处理配置
preprocessing_config = {
    "BOW": {"ngram_range": [1, 2], "max_features": [100, 200]},
    "RemoveStopWords": {"Parameter1": ["..."]}
}

def generate_param_combinations(config):
    # 为每个预处理步骤生成参数的所有可能组合
    step_candidates = []
    for step_name, params in config.items():
        param_keys = list(params.keys())
        param_values = list(params.values())
        # 生成当前步骤的所有参数组合（字典形式）
        step_param_combos = [dict(zip(param_keys, vals)) for vals in itertools.product(*param_values)]
        # 包装为(步骤名, 参数组合)的元组列表，方便后续组装
        step_candidates.append([(step_name, combo) for combo in step_param_combos])
    
    # 生成跨步骤的笛卡尔积组合
    all_combos = itertools.product(*step_candidates)
    
    # 组装成最终的嵌套结构
    result = []
    for combo in all_combos:
        preprocessing_dict = {}
        for step_name, params in combo:
            preprocessing_dict[step_name] = params
        result.append(preprocessing_dict)
    
    return result

# 生成并输出结果
final_result = generate_param_combinations(preprocessing_config)
print(json.dumps(final_result, indent=2))

关键修复点 #

• 保留了BOW和RemoveStopWords的层级结构，未将参数平铺开
• 先为每个预处理步骤单独生成参数组合，再通过笛卡尔积得到跨步骤的所有组合
• 最终组装时将每个步骤的参数组合对应到原步骤名称下，完全匹配预期的嵌套格式

内容的提问来源于stack exchange，提问作者Daniel Kennett