人工手动筛选特征

一、相关性分析

计算特征之间的相关性，并移除高度相关的特征（相关系数大于 0.95）

import pandas as pd  
import numpy as np  

# 创建一个示例数据框  
data = {  
    'A': [1, 2, 3, 4, 5],  
    'B': [5, 4, 3, 2, 1],  
    'C': [1, 1, 2, 2, 3],  
    'D': [5, 3, 2, 4, 5],  
    'E': [2, 4, 6, 8, 10]  
}  

X = pd.DataFrame(data)  
print("原始数据框:")  
print(X)  

# 计算相关矩阵  
corr_matrix = X.corr().abs()  
print("\n相关矩阵:")  
print(corr_matrix)  

# 获取上三角矩阵  
upper = corr_matrix.where(np.triu(np.ones(corr_matrix.shape), k=1).astype(bool))  
print("\n上三角矩阵:")  
print(upper)  

# 找出任何相关系数大于 0.95 的列  
to_drop = [column for column in upper.columns if any(upper[column] > 0.95)]  
print("\n要移除的特征:")  
print(to_drop)  

# 移除高相关特征  
X_reduced = X.drop(to_drop, axis=1)  
print("\n去除高度相关后的数据框:")  
print(X_reduced)

二、重要性分析

提取和展示模型中各特征的重要性，尤其是查看特征重要性为零的特征以及非零特征。

import pandas as pd  
import lightgbm as lgb  
from sklearn.datasets import load_breast_cancer  
from sklearn.model_selection import train_test_split  

# 加载数据集  
data = load_breast_cancer()  
X = pd.DataFrame(data.data, columns=data.feature_names)  
y = pd.Series(data.target)  

# 拆分训练和测试数据  
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)  

# 创建 LightGBM 数据集  
d_train = lgb.Dataset(X_train, label=y_train)  

# 训练 LightGBM 模型  
model = lgb.train(params={'objective': 'binary'}, train_set=d_train, num_boost_round=100)  

# 获取特征重要性  
fea_iptc = pd.DataFrame({  
    'feature_name': X.columns,  
    'importance': model.feature_importance()  
})  

# 显示特征重要性数据框  
print(fea_iptc)  

# 找到特征重要性为0的特征  
fea_importance_0 = fea_iptc['feature_name'].loc[fea_iptc['importance'] == 0].tolist()  
print('fea_importance_0: ', len(fea_importance_0), fea_importance_0)  

# 找到特征重要性不为0的特征  
fea_importance_not_0 = fea_iptc['feature_name'].loc[fea_iptc['importance'] != 0].tolist()  
print('fea_importance_not_0: ', len(fea_importance_not_0), fea_importance_not_0)

三、iv和psi分析

使用信息值（IV）和群体稳定指数（PSI）筛选特征

import pandas as pd  

# 示例数据：特征名称与它们的 IV 和 PSI 值  
data = {  
    'name': ['feature1', 'feature2', 'feature3', 'feature4', 'feature5'],  
    'iv': [0.02, 0.15, 0.03, 0.01, 0.5],  # IV 值  
    'psi': [0.04, 0.02, 0.06, 0.07, 0.03]  # PSI 值  
}  

# 创建 DataFrame  
iv_df = pd.DataFrame(data)  

# 定义筛选特征的函数  
def select_features_by_iv_psi(feature_stats, iv_threshold=0.01, psi_threshold=0.05):  
    """根据IV和PSI值筛选特征"""  
    high_iv_features = feature_stats[feature_stats['iv'] >= iv_threshold]  
    selected_features = high_iv_features[high_iv_features['psi'] < psi_threshold]  
    return selected_features['name'].tolist()  

# 可选的特征列表  
selected_features_list = ['feature1', 'feature2', 'feature3']  

# 过滤特征  
filtered_iv_df = iv_df[iv_df['name'].isin(selected_features_list)]  

# 使用筛选函数  
selected_features = select_features_by_iv_psi(filtered_iv_df)  

# 输出筛选结果  
print("Selected features:", len(selected_features), selected_features)