staran 0.5.0__py3-none-any.whl → 0.6.1__py3-none-any.whl

staran/models/daifa_models.py

@@ -0,0 +1,361 @@
+"""
+新疆工行代发长尾客户专用模型定义
+
+包含两个核心模型：
+1. 代发长尾客户提升3k预测模型
+2. 代发长尾客户防流失1.5k预测模型
+
+基于新疆工行代发长尾客户数据库和业务规则
+"""
+
+from typing import Dict, List
+from .config import create_model_config
+from .target import create_target_definition
+from .registry import ModelRegistry, register_model
+import os
+import json
+from datetime import datetime
+
+
+def save_model_registry(output_path: str):
+    """保存模型注册信息到文件"""
+    
+    def convert_to_serializable(obj):
+        """递归转换对象为可序列化格式"""
+        if isinstance(obj, datetime):
+            return obj.isoformat()
+        elif hasattr(obj, '__dict__'):
+            result = {}
+            for key, value in obj.__dict__.items():
+                result[key] = convert_to_serializable(value)
+            return result
+        elif hasattr(obj, 'value'):  # 枚举类型
+            return obj.value
+        elif isinstance(obj, (list, tuple)):
+            return [convert_to_serializable(item) for item in obj]
+        elif isinstance(obj, dict):
+            return {k: convert_to_serializable(v) for k, v in obj.items()}
+        else:
+            return obj
+    
+    data = {
+        "models": {},
+        "version_history": {},
+        "saved_at": str(datetime.now())
+    }
+    
+    # 获取所有注册的模型
+    for model_id, entry in ModelRegistry._models.items():
+        data["models"][model_id] = {
+            "model_config": convert_to_serializable(entry.model_config),
+            "target_definition": convert_to_serializable(entry.target_definition),
+            "registered_at": entry.registered_at.isoformat(),
+            "status": entry.status,
+            "performance_metrics": entry.performance_metrics
+        }
+    
+    data["version_history"] = ModelRegistry._version_history.copy()
+    
+    with open(output_path, 'w', encoding='utf-8') as f:
+        json.dump(data, f, indent=2, ensure_ascii=False)
+    
+    print(f"✅ 模型注册信息已保存到: {output_path}")
+    return output_path
+
+
+def create_daifa_longtail_upgrade_model() -> Dict:
+    """创建代发长尾客户提升3k预测模型"""
+    
+    # 模型配置
+    model_config = create_model_config(
+        name="xinjiang_icbc_daifa_longtail_upgrade_3k",
+        model_type="classification",
+        algorithm="gradient_boosting",
+        version="1.0.0",
+        schema_name="daifa_longtail",
+        table_types=["daifa_longtail_behavior", "daifa_longtail_asset_avg", 
+                    "daifa_longtail_asset_config", "daifa_longtail_monthly_stat"],
+        hyperparameters={
+            "n_estimators": 300,
+            "learning_rate": 0.05,
+            "max_depth": 12,
+            "min_samples_split": 20,
+            "min_samples_leaf": 10,
+            "subsample": 0.8,
+            "random_state": 42
+        },
+        bank_code="xinjiang_icbc",
+        business_domain="代发长尾客户",
+        description="新疆工行代发长尾客户下个月资产提升3k预测模型",
+        tags=["daifa", "longtail", "upgrade", "3k", "xinjiang_icbc"]
+    )
+    
+    # 目标定义 - 预测下个月提升3k
+    target_definition = create_target_definition(
+        name="daifa_longtail_upgrade_3k_target",
+        target_type="binary_classification",
+        description="新疆工行代发长尾客户下个月资产提升3000元预测目标",
+        sql_query="""
+        WITH customer_baseline AS (
+            -- 获取代发长尾客户基础信息（当月）
+            SELECT 
+                b.party_id,
+                b.asset_total_bal as current_asset,
+                b.salary_amount as current_salary,
+                b.longtail_score,
+                b.upgrade_potential,
+                CASE 
+                    WHEN b.asset_total_bal BETWEEN 10000 AND 100000 THEN 1 
+                    ELSE 0 
+                END as is_daifa_longtail
+            FROM xinjiang_icbc_daifa_hlwj_monthly_stat_wy b
+            WHERE b.data_dt = '{baseline_date}'
+        ),
+        
+        next_month_performance AS (
+            -- 计算下个月的资产变化
+            SELECT 
+                party_id,
+                asset_total_bal as next_month_asset,
+                salary_amount as next_month_salary,
+                monthly_deposit_amount,
+                monthly_withdraw_amount
+            FROM xinjiang_icbc_daifa_hlwj_monthly_stat_wy
+            WHERE data_dt = '{next_month_date}'
+        ),
+        
+        asset_change AS (
+            -- 计算资产变化情况
+            SELECT 
+                cb.party_id,
+                cb.current_asset,
+                nmp.next_month_asset,
+                (nmp.next_month_asset - cb.current_asset) as asset_change,
+                nmp.monthly_deposit_amount,
+                cb.upgrade_potential
+            FROM customer_baseline cb
+            INNER JOIN next_month_performance nmp ON cb.party_id = nmp.party_id
+            WHERE cb.is_daifa_longtail = 1  -- 只关注代发长尾客户
+        )
+        
+        SELECT 
+            party_id,
+            CASE 
+                -- 代发长尾客户资产提升3k的判断标准
+                WHEN asset_change >= 3000  -- 资产增长达到3000元
+                     AND monthly_deposit_amount > asset_change * 0.7  -- 主要通过存入实现
+                     AND upgrade_potential >= 0.6  -- 提升潜力评分较高
+                THEN 1 
+                ELSE 0 
+            END as upgrade_3k_target,
+            
+            -- 辅助分析字段
+            current_asset,
+            next_month_asset,
+            asset_change,
+            monthly_deposit_amount,
+            upgrade_potential
+            
+        FROM asset_change
+        """,
+        target_column="upgrade_3k_target",
+        class_labels=["no_upgrade", "upgrade_3k"],
+        class_weights={"no_upgrade": 1.0, "upgrade_3k": 2.5},  # 提升类样本权重更高
+        time_window="1_month",
+        prediction_horizon="1_month",
+        bank_code="xinjiang_icbc",
+        business_rules={
+            "min_asset_threshold": 10000,      # 代发长尾最小资产
+            "max_asset_threshold": 100000,     # 代发长尾最大资产
+            "upgrade_target_amount": 3000,     # 提升目标金额
+            "deposit_contribution_ratio": 0.7, # 存入贡献占比
+            "min_upgrade_potential": 0.6       # 最小提升潜力
+        }
+    )
+    
+    return {
+        "model_config": model_config,
+        "target_definition": target_definition,
+        "model_type": "upgrade_prediction"
+    }
+
+
+def create_daifa_longtail_churn_model() -> Dict:
+    """创建代发长尾客户防流失1.5k预测模型"""
+    
+    # 模型配置
+    model_config = create_model_config(
+        name="xinjiang_icbc_daifa_longtail_churn_1_5k",
+        model_type="classification",
+        algorithm="random_forest",  # 防流失模型使用随机森林
+        version="1.0.0",
+        schema_name="daifa_longtail",
+        table_types=["daifa_longtail_behavior", "daifa_longtail_asset_avg", 
+                    "daifa_longtail_asset_config", "daifa_longtail_monthly_stat"],
+        hyperparameters={
+            "n_estimators": 200,
+            "max_depth": 10,
+            "min_samples_split": 15,
+            "min_samples_leaf": 8,
+            "max_features": "sqrt",
+            "random_state": 42,
+            "class_weight": "balanced"  # 处理不平衡数据
+        },
+        bank_code="xinjiang_icbc",
+        business_domain="代发长尾客户",
+        description="新疆工行代发长尾客户下个月流失1.5k资产风险预测模型",
+        tags=["daifa", "longtail", "churn", "1_5k", "risk_prevention"]
+    )
+    
+    # 目标定义 - 预测下个月流失1.5k风险
+    target_definition = create_target_definition(
+        name="daifa_longtail_churn_1_5k_target",
+        target_type="binary_classification",
+        description="新疆工行代发长尾客户下个月流失1500元资产风险预测目标",
+        sql_query="""
+        WITH customer_baseline AS (
+            -- 获取代发长尾客户基础信息（当月）
+            SELECT 
+                b.party_id,
+                b.asset_total_bal as current_asset,
+                b.salary_amount as current_salary,
+                b.longtail_score,
+                b.churn_risk,
+                b.login_days,
+                CASE 
+                    WHEN b.asset_total_bal BETWEEN 10000 AND 100000 THEN 1 
+                    ELSE 0 
+                END as is_daifa_longtail
+            FROM xinjiang_icbc_daifa_hlwj_monthly_stat_wy b
+            WHERE b.data_dt = '{baseline_date}'
+        ),
+        
+        next_month_performance AS (
+            -- 计算下个月的资产变化和行为
+            SELECT 
+                party_id,
+                asset_total_bal as next_month_asset,
+                monthly_withdraw_amount,
+                login_days as next_month_login_days
+            FROM xinjiang_icbc_daifa_hlwj_monthly_stat_wy
+            WHERE data_dt = '{next_month_date}'
+        ),
+        
+        churn_analysis AS (
+            -- 分析流失风险情况
+            SELECT 
+                cb.party_id,
+                cb.current_asset,
+                nmp.next_month_asset,
+                (cb.current_asset - nmp.next_month_asset) as asset_decrease,
+                nmp.monthly_withdraw_amount,
+                cb.churn_risk,
+                cb.login_days,
+                nmp.next_month_login_days
+            FROM customer_baseline cb
+            INNER JOIN next_month_performance nmp ON cb.party_id = nmp.party_id
+            WHERE cb.is_daifa_longtail = 1  -- 只关注代发长尾客户
+        )
+        
+        SELECT 
+            party_id,
+            CASE 
+                -- 代发长尾客户流失1.5k的判断标准
+                WHEN asset_decrease >= 1500  -- 资产减少达到1500元
+                     AND monthly_withdraw_amount >= 1500  -- 主要通过取出导致
+                     AND (
+                         churn_risk >= 0.7  -- 流失风险评分高
+                         OR next_month_login_days <= login_days * 0.5  -- 活跃度大幅下降
+                     )
+                THEN 1 
+                ELSE 0 
+            END as churn_1_5k_target,
+            
+            -- 辅助分析字段
+            current_asset,
+            next_month_asset,
+            asset_decrease,
+            monthly_withdraw_amount,
+            churn_risk,
+            login_days,
+            next_month_login_days
+            
+        FROM churn_analysis
+        """,
+        target_column="churn_1_5k_target",
+        class_labels=["no_churn", "churn_1_5k"],
+        class_weights={"no_churn": 1.0, "churn_1_5k": 3.0},  # 流失类样本权重更高
+        time_window="1_month",
+        prediction_horizon="1_month",
+        bank_code="xinjiang_icbc",
+        business_rules={
+            "min_asset_threshold": 10000,        # 代发长尾最小资产
+            "max_asset_threshold": 100000,       # 代发长尾最大资产
+            "churn_threshold_amount": 1500,      # 流失阈值金额
+            "min_churn_risk": 0.7,               # 最小流失风险
+            "activity_decline_ratio": 0.5        # 活跃度下降比例
+        }
+    )
+    
+    return {
+        "model_config": model_config,
+        "target_definition": target_definition,
+        "model_type": "churn_prevention"
+    }
+
+
+def create_both_daifa_models(output_dir: str = "./xinjiang_models") -> Dict:
+    """创建两个代发长尾客户模型并注册"""
+    
+    # 确保输出目录存在
+    os.makedirs(output_dir, exist_ok=True)
+    
+    # 创建提升模型
+    upgrade_model = create_daifa_longtail_upgrade_model()
+    upgrade_id = register_model(
+        upgrade_model["model_config"], 
+        upgrade_model["target_definition"]
+    )
+    
+    # 创建防流失模型
+    churn_model = create_daifa_longtail_churn_model()
+    churn_id = register_model(
+        churn_model["model_config"], 
+        churn_model["target_definition"]
+    )
+    
+    # 保存注册信息到指定目录
+    registry_path = os.path.join(output_dir, "model_registry.json")
+    save_model_registry(registry_path)
+    
+    return {
+        "upgrade_model": {
+            "model_id": upgrade_id,
+            "config": upgrade_model["model_config"],
+            "target": upgrade_model["target_definition"]
+        },
+        "churn_model": {
+            "model_id": churn_id,
+            "config": churn_model["model_config"],
+            "target": churn_model["target_definition"]
+        },
+        "registry_path": registry_path,
+        "output_dir": output_dir
+    }
+
+
+def get_available_daifa_models() -> List[str]:
+    """获取所有可用的代发长尾客户模型"""
+    return [
+        "daifa_longtail_upgrade_3k",   # 代发长尾客户提升3k模型
+        "daifa_longtail_churn_1_5k"    # 代发长尾客户防流失1.5k模型
+    ]
+
+
+# 导出函数
+__all__ = [
+    'create_daifa_longtail_upgrade_model',
+    'create_daifa_longtail_churn_model', 
+    'create_both_daifa_models',
+    'get_available_daifa_models'
+]

staran/schemas/__init__.py

@@ -1,28 +1,27 @@
 """
-Staran Schemas模块 - 数据表结构定义与文档生成
+Staran Schemas模块 - 新疆工行代发长尾客户表结构定义
 
-提供标准化的表结构定义、字段管理和文档生成功能。
-支持根据表结构生成Markdown和PDF文档供业务方使用。
+提供新疆工行代发长尾客户的标准化表结构定义和字段管理功能。
 
 主要功能:
-- 表结构标准化定义
+- 代发长尾客户表结构定义
 - 业务字段含义管理  
-- 文档自动生成 (MD/PDF)
-- 多业务领域支持
+- 新疆工行专用配置
+- 表结构文档生成
 """
 
-from .document_generator import SchemaDocumentGenerator
+from ..tools.document_generator import SchemaDocumentGenerator
 from .aum import *
 
 __all__ = [
     'SchemaDocumentGenerator',
-    # AUM业务表
-    'AUMBehaviorSchema',
-    'AUMAssetAvgSchema', 
-    'AUMAssetConfigSchema',
-    'AUMMonthlyStatSchema',
-    'get_aum_schemas',
-    'export_aum_docs'
+    # 新疆工行代发长尾客户表
+    'XinjiangICBCDaifaLongtailBehaviorSchema',
+    'XinjiangICBCDaifaLongtailAssetAvgSchema', 
+    'XinjiangICBCDaifaLongtailAssetConfigSchema',
+    'XinjiangICBCDaifaLongtailMonthlyStatSchema',
+    'get_xinjiang_icbc_daifa_longtail_schemas',
+    'export_xinjiang_icbc_daifa_longtail_docs'
 ]
 
-__version__ = "0.3.0"
+__version__ = "0.6.0"