import pandas as pd
import numpy as np
import xgboost as xgb
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder, StandardScaler
from sklearn.metrics import accuracy_score, classification_report
import joblib
import json

# 1. 加载数据
print("正在加载数据...")
df = pd.read_csv('bank.csv')

# 2. 数据预处理
print("正在进行数据预处理...")

# 移除 duration 列 (避免数据泄露)
if 'duration' in df.columns:
    df = df.drop('duration', axis=1)

# 分离特征和目标
X = df.drop('deposit', axis=1)
y = df['deposit']

# 处理目标变量 (yes -> 1, no -> 0)
le_target = LabelEncoder()
y = le_target.fit_transform(y)

# 识别分类特征和数值特征
categorical_cols = X.select_dtypes(include=['object']).columns.tolist()
numeric_cols = X.select_dtypes(include=['int64', 'float64']).columns.tolist()

# 保存列名信息，供 Agent 使用
feature_meta = {
    'numeric_cols': numeric_cols,
    'categorical_cols': categorical_cols,
    'all_cols': list(X.columns)
}

# 对分类特征进行 Label Encoding
# 注意：XGBoost 可以处理类别特征，但通常需要转换为数值。
# 为了简化 Agent 的推理流程，我们需要保存这些 Encoder。
encoders = {}
for col in categorical_cols:
    le = LabelEncoder()
    X[col] = le.fit_transform(X[col])
    encoders[col] = le

# 3. 划分数据集
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# 4. 训练模型
print("正在训练 XGBoost 模型...")
model = xgb.XGBClassifier(
    n_estimators=100,
    learning_rate=0.1,
    max_depth=5,
    use_label_encoder=False,
    eval_metric='logloss'
)
model.fit(X_train, y_train)

# 5. 评估模型
y_pred = model.predict(X_test)
y_pred_proba = model.predict_proba(X_test)[:, 1]

print("\n模型评估结果:")
print(f"Accuracy: {accuracy_score(y_test, y_pred):.4f}")
print("\nClassification Report:")
print(classification_report(y_test, y_pred))

# 6. 保存资产
print("\n正在保存模型和预处理工具...")
artifacts = {
    'model': model,
    'encoders': encoders,
    'target_encoder': le_target,
    'feature_meta': feature_meta
}
joblib.dump(artifacts, 'model_artifacts.pkl')

# 另外保存一份特征重要性，供参考
importances = model.feature_importances_
feature_names = X.columns
feat_imp_df = pd.DataFrame({'Feature': feature_names, 'Importance': importances})
feat_imp_df = feat_imp_df.sort_values(by='Importance', ascending=False)
print("\n特征重要性 Top 5:")
print(feat_imp_df.head())

print("\n完成！模型已保存为 'model_artifacts.pkl'")