import sys
import pandas as pd
import numpy as np
from sklearn.model_selection import KFold
import xgboost as xgb
from xgboost import XGBRegressor
from lightgbm import LGBMRegressor
from sklearn.linear_model import (
    HuberRegressor, RANSACRegressor, TheilSenRegressor,
    Lasso, ElasticNet, Ridge
)
from sklearn.cross_decomposition import PLSRegression
from sklearn.preprocessing import StandardScaler, RobustScaler
from sklearn.ensemble import RandomForestRegressor
from scipy.stats import pearsonr
import warnings
import torch
import matplotlib.pyplot as plt
import seaborn as sns
from concurrent.futures import ThreadPoolExecutor, as_completed
from itertools import combinations
import time

train_df = pd.read_pickle('train_df.pkl')
test_df = pd.read_pickle('test_df.pkl')

length = len(train_df)
df = pd.concat([train_df, test_df], axis=0)
LABEL_COLUMN = 'label'
feature_cols = [col for col in train_df.columns if col != LABEL_COLUMN]

X = train_df[feature_cols].values
y = train_df[LABEL_COLUMN].values

# 先将X转为numpy数组，再删除全为0的列，并同步更新feature_cols
X_np = np.asarray(X)
nonzero_col_idx = np.where((X_np != 0).any(axis=0))[0]
X = X_np[:, nonzero_col_idx]
feature_cols = [feature_cols[i] for i in nonzero_col_idx]

X_np = np.asarray(X)
y_np = np.asarray(y)
corrs = np.array([np.corrcoef(X_np[:, i], y_np)[0, 1] for i in range(X_np.shape[1])])

# 对于相关性为负的特征，将该列取负
X_adj = X_np.copy()
neg_idx = np.where(corrs < 0)[0]
X_adj[:, neg_idx] = -X_adj[:, neg_idx]

# 找到相关性绝对值大于0.01的特征索引
selected_idx = np.where(np.abs(corrs) > 0.01)[0]

# 取出这些特征对应的X的列
X_selected = X_adj[:, selected_idx]
selected_features = [feature_cols[i] for i in selected_idx]

def max_ic_factor_selection(X, y, feature_cols, threshold=0.9):
    X = np.asarray(X)
    n_features = X.shape[1]
    corr_matrix = np.corrcoef(X, rowvar=False)
    used = set()
    selected_idx = []
    for i in range(n_features):
        if i in used:
            continue
        # 找到与第i个特征高度相关的特征
        group = [i]
        for j in range(i+1, n_features):
            if j not in used and abs(corr_matrix[i, j]) > threshold:
                group.append(j)
        # 组内选与y相关性（IC）最大的特征
        if len(group) == 1:
            selected_idx.append(group[0])
        else:
            ic_list = [abs(pearsonr(X[:, k], y)[0]) for k in group]
            best_k = group[np.argmax(ic_list)]
            selected_idx.append(best_k)
        used.update(group)
    X_new = X[:, selected_idx]
    feature_cols_new = [feature_cols[i] for i in selected_idx]
    return X_new, feature_cols_new

# 在训练前进行最大IC因子合成，减少共线性
n_train = train_df.shape[0]
X_selected, selected_features = max_ic_factor_selection(X_selected, y[:n_train], selected_features, threshold=0.9)

X_train = X_selected
X_test = test_df[selected_features].values

y_train = y
y_test = test_df[LABEL_COLUMN].values
breakpoint()

kf = KFold(n_splits=5, shuffle=True, random_state=42)

import math

# 余弦退火调度函数
def cosine_annealing(epoch, initial_lr=0.01, T_max=5000, eta_min=1e-4):
    return eta_min + (initial_lr - eta_min) * (1 + math.cos(math.pi * epoch / T_max)) / 2

# XGBoost参数（更复杂的树结构+更强正则+早停机制）
xgb_params = {
    'n_estimators': 10000,           # 增加树的数量
    'learning_rate': 0.01,
    'max_depth': 10,                # 增加树的深度
    'subsample': 0.85,              # 增加样本采样比例
    'colsample_bytree': 0.85,       # 增加特征采样比例
    'tree_method': 'hist',
    'device': 'gpu',
    'predictor': 'gpu_predictor',
    'random_state': 42,
    'reg_alpha': 5,                 # 增大L1正则
    'reg_lambda': 10,               # 增大L2正则
    'min_child_weight': 5,          # 增大叶子节点最小样本权重和
    'gamma': 0.2,                   # 增大分裂所需的最小损失减少
    'early_stopping_round': 100,
    'verbose_eval': 100,
    'eval_metric': 'rmse',
    'callbacks': [
        xgb.callback.LearningRateScheduler(cosine_annealing)
    ]
}
print("start training")
val_scores = []
test_preds = np.zeros(X_test.shape[0])

for train_idx, val_idx in kf.split(X_train):
    X_tr, X_val = X_train[train_idx], X_train[val_idx]
    y_tr, y_val = y_train[train_idx], y_train[val_idx]
    model = XGBRegressor(**xgb_params)
    model.fit(
        X_tr, y_tr,
        eval_set=[(X_val, y_val)],
        # eval_metric='rmse',
    )
    val_pred = model.predict(X_val)
    val_score = np.sqrt(np.mean((val_pred - y_val) ** 2))  # RMSE
    val_scores.append(val_score)
    test_preds += model.predict(X_test) / kf.n_splits

print(f"平均验证RMSE: {np.mean(val_scores):.6f}")

# 保存预测结果到csv
result_df = pd.DataFrame({
    'ID': np.arange(1, len(test_preds) + 1),
    'prediction': test_preds
})
result_df.to_csv('xgb_prediction-3.csv', index=False)
print('预测结果已保存到 xgb_prediction.csv')