Kaggle-Binary Prediction with a Rainfall Dataset-(回归+特征工程+xgb)

Binary Prediction with a Rainfall Dataset

题意：

给你每天的天气信息，让你预测降雨量。

数据处理：

1.根据特征值构造天气降雨量的新特征值
2.根据时间构造月和季节特征
3.处理缺失值

建立模型：

1.建立lightgbm模型
2.建立xgboost模型，并进行网格搜索最佳参数模型
3.进行模型融合

代码：

import os
import sys
import warnings
import numpy as np
import pandas as pd
import seaborn
from matplotlib import pyplot as plt
import lightgbm
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
from sklearn.model_selection import train_test_split, GridSearchCV, cross_val_score
from xgboost import XGBRegressor


def init():
    os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'  # 仅输出错误日志
    warnings.simplefilter('ignore')  # 忽略警告日志
    pd.set_option('display.width', 1000)
    pd.set_option('display.max_colwidth', 1000)
    pd.set_option("display.max_rows", 1000)
    pd.set_option("display.max_columns", 1000)

def show_dataframe():
    print("查看特征值和特征值类型\n" + str(df_train.dtypes) + "\n" + "-"*100)
    print("查看前10行信息\n" + str(df_train.head()) + "\n" + "-"*100)
    print("查看每个特征值的各种数据统计信息\n" + str(df_train.describe()) + "\n" + "-" * 100)
    print("输出重复行的个数\n" + str(df_train.duplicated().sum()) + "\n" + "-" * 100)
    print("查看每列的缺失值个数\n" + str(df_train.isnull().sum()) + "\n" + "-" * 100)
    print("查看缺失值的具体信息\n" + str(df_train.info()) + "\n" + "-" * 100)
    print("输出X所有值出现的是什么，还有对应出现的次数\n" + str(df_train['X'].value_count()) + "\n" + "-" * 100)

def show_relation(data, colx, coly): #输出某一特征值与目标值的关系
    if data[colx].dtype == 'object' or data[colx].dtype == 'category' or len(data[colx].unique()) < 20:
        seaborn.boxplot(x=colx, y=coly, data=data)
    else:
        plt.scatter(data[colx], data[coly])
    plt.xlabel(colx)
    plt.ylabel(coly)
    plt.show()

def show_score(model_name,pred):
    mse = mean_squared_error(y_train, pred)
    mae = mean_absolute_error(y_train, pred)
    score = r2_score(y_train, pred)
    print(model_name)
    print(f"{'MSE':<10}{mse:<15.4f}")
    print(f"{'MAE':<10}{mae:<15.4f}")
    print(f"{'R²':<10}{score:<15.4f}")
    print("-"*100)

if __name__ == '__main__':
    df_train = pd.read_csv('/kaggle/input/playground-series-s5e3/train.csv')
    df_test = pd.read_csv('/kaggle/input/playground-series-s5e3/test.csv')

    init()
    #df_show()

    df_all = pd.concat([df_train.drop(['id', 'rainfall'], axis=1), df_test.drop(['id'], axis=1)], axis=0)
    df_all['cha_temp'] = df_all['maxtemp']-df_all['mintemp'] #温差
    df_all['cha_temp_dew'] = df_all['temparature'] - df_all['dewpoint'] #温度湿度差
    df_all['humidity_norm'] = df_all['humidity'] / 100  #湿度转换为0-1范围
    df_all['precip_potential'] = (df_all['humidity_norm'] * df_all['cloud']) / (df_all['windspeed'] + 1e-5) #降水潜力指数
    df_all['solar_energy'] = df_all['sunshine'] * (df_all['maxtemp'] - df_all['mintemp']) #日照能量
    df_all['month'] = pd.to_datetime(df_all['day']).dt.month  #假设有date列
    df_all['season'] = df_all['month'] % 12 // 3 + 1  #季节特征
    
    X_train = df_all[:df_train.shape[0]]
    Y_train = df_train['rainfall']
    x_train,x_val,y_train,y_test = train_test_split(X_train,Y_train,test_size=0.2,random_state=42)

    x_test = df_all[df_train.shape[0]:]

    model_lgb = lightgbm.LGBMRegressor(
        n_estimators=3000,  # 增加迭代次数配合早停
        learning_rate=0.03,  # 减小学习率
        num_leaves=15,  # 限制模型复杂度
        min_child_samples=25,  # 增加最小叶子样本数
        reg_alpha=0.1,  # L1正则化
        reg_lambda=0.1,  # L2正则化
        objective='regression_l1',  # 改用MAE损失
        verbose=100
    )
    model_lgb.fit(x_train, y_train)
    pred_lgb = model_lgb.predict(x_test)
    show_score(model_name='lgb', pred=model_lgb.predict(x_train))

    xgb_param_grid = {
        'n_estimators': [50, 100, 200,250],  # 树的数量
        'max_depth': [3, 4, 5, 6],  # 树的最大深度
        'learning_rate': [0.01,0.05,0.1, 0.3],  # 学习速率
        'subsample': [0.8, 1.0,1.3],  # 指定每次迭代中用于训练每棵树的数据比例
        #'colsample_bytree': [0.8, 1.0], #指定每次迭代中用于训练每棵树的特征比例
        #'gamma': [0, 0.1, 0.2], #最小损失减少值
        #'min_child_weight': [1, 3, 5], #子节点所需的最小样本权重和
        #'reg_alpha': [0, 0.1, 1], #控制模型的正则化强度
        #'reg_lambda': [0, 0.1, 1] #控制模型的正则化强度
    }
    model_xgb = GridSearchCV(
        estimator=XGBRegressor(random_state=42),  # 对什么模型进行搜索超参数
        param_grid=xgb_param_grid,  # 超参数的候选值
        cv=4,  # 使用3折交叉验证
    )
    model_xgb.fit(x_train, y_train)
    pred_xgb = model_xgb.predict(x_test)
    show_score(model_name='xgb', pred=model_xgb.predict(x_train))

    pred_all = pred_xgb
    submission = pd.DataFrame({
        'id': df_test['id'],
        'sales': np.where(pred_all < 0, 0, pred_all)
    })
    submission.to_csv('/kaggle/working/submission.csv', index=False)