BrandCultivation/models/rank/data/utils.py

import pandas as pd
import re

def one_hot_embedding(dataframe, onehout_feat):
    """对数据的指定特征做embedding编码"""
    # 先将指定的特征进行Categorical处理
    for feat, categories in onehout_feat.items():
        dataframe[feat] = pd.Categorical(dataframe[feat], categories=categories, ordered=False)
    dataframe = pd.get_dummies(
        dataframe,
        columns=list(onehout_feat.keys()),
        prefix_sep="_",
        dtype=int,
    )

    # 清理列名中的特殊字符，避免 LightGBM 报错
    # 将 JSON 特殊字符替换为下划线或删除
    dataframe.columns = [
        re.sub(r'[",\\\n\r\t\b\f]', '_', col)  # 替换 JSON 特殊字符
        .replace(' ', '_')                      # 替换空格
        for col in dataframe.columns
    ]

    return dataframe

def sample_data_clear(data, config):
    for feature, rules, in config.CLEANING_RULES.items():    
        if rules["type"] == "num":
            data[feature] = pd.to_numeric(data[feature], errors="coerce")
        if rules["method"] == "fillna":
            if rules["opt"] == "fill":
                data[feature] = data[feature].fillna(rules["value"]).infer_objects(copy=False)
            elif rules["opt"] == "mean":
                data[feature] = data[feature].fillna(data[feature].mean()).infer_objects(copy=False)
            data[feature] = data[feature].infer_objects(copy=False)
    return data