staran 0.2.4__py3-none-any.whl → 0.4.0__py3-none-any.whl

staran/examples/aum_longtail.py

@@ -0,0 +1,250 @@
+"""
+AUM代发长尾模型示例
+基于Staran v0.3.0架构，使用schemas模块的预定义表结构
+"""
+
+from typing import Dict, Optional
+from ..engines import create_turing_engine
+from ..features import FeatureManager, FeatureConfig, FeatureType
+from ..tools import Date
+from ..schemas.aum import get_aum_schemas
+
+
+class AUMLongtailExample:
+    """AUM代发长尾模型示例类"""
+    
+    def __init__(self, database: str = "dwegdata03000"):
+        """
+        初始化AUM长尾模型示例
+        
+        Args:
+            database: 数据库名称，默认为dwegdata03000
+        """
+        self.database = database
+        self.engine = create_turing_engine(database)
+        self.schemas = get_aum_schemas()  # 从schemas模块获取预定义的表结构
+        
+    def run(self, feature_date: Optional[str] = None, output_path: str = "file:///nfsHome/aum_longtail") -> Dict:
+        """
+        运行完整的AUM长尾模型特征工程
+        
+        Args:
+            feature_date: 特征日期，格式为YYYYMM，默认为当前月
+            output_path: 输出路径，默认为file:///nfsHome/aum_longtail
+            
+        Returns:
+            包含所有结果的字典
+        """
+        if feature_date is None:
+            feature_date = Date.today().format_compact()[:6]
+        
+        print(f"🚀 开始AUM长尾模型特征工程 - {feature_date}")
+        print("="*60)
+        
+        results = {}
+        
+        # 步骤1: 生成A表特征（行为特征表）- 只生成原始拷贝和聚合特征
+        print("📊 步骤1: 生成客户行为特征...")
+        results['behavior'] = self._generate_behavior_features('behavior', feature_date)
+        
+        # 步骤2: 生成B表特征（资产平均值表）- 完整特征
+        print("💰 步骤2: 生成资产平均值特征...")
+        results['asset_avg'] = self._generate_full_features('asset_avg', feature_date)
+        
+        # 步骤3: 生成C表特征（资产配置表）- 完整特征  
+        print("📈 步骤3: 生成资产配置特征...")
+        results['asset_config'] = self._generate_full_features('asset_config', feature_date)
+        
+        # 步骤4: 生成D表特征（月度统计表）- 完整特征
+        print("📋 步骤4: 生成月度统计特征...")
+        results['monthly_stat'] = self._generate_full_features('monthly_stat', feature_date)
+        
+        # 步骤5: 导出特征表
+        print("💾 步骤5: 导出特征表...")
+        results['exports'] = self._export_features(feature_date, output_path)
+        
+        print("="*60)
+        print("✅ AUM长尾模型特征工程完成！")
+        return results
+    
+    def _generate_behavior_features(self, table_type: str, feature_date: str) -> Dict:
+        """生成行为特征（A表）- 只生成原始拷贝和聚合特征"""
+        schema = self.schemas[table_type]
+        manager = FeatureManager(self.engine, self.database)
+        
+        # A表特征配置：只启用原始拷贝和聚合
+        config = FeatureConfig()
+        config.enable_feature(FeatureType.RAW_COPY)
+        config.enable_feature(FeatureType.AGGREGATION)
+        config.disable_feature(FeatureType.MOM)  # 不生成环比
+        config.disable_feature(FeatureType.YOY)  # 不生成同比
+        
+        print(f"   🔧 生成{schema.table_name}的特征...")
+        result = manager.generate_features(
+            schema=schema,
+            config=config,
+            feature_date=feature_date
+        )
+        
+        feature_count = manager.count_features(schema, config)
+        print(f"   ✅ A表特征生成完成: {feature_count}个特征")
+        return result
+    
+    def _generate_full_features(self, table_type: str, feature_date: str) -> Dict:
+        """生成完整特征（B、C、D表）- 聚合+5个月环比+1年同比"""
+        schema = self.schemas[table_type]
+        manager = FeatureManager(self.engine, self.database)
+        
+        # B、C、D表特征配置：完整特征集
+        config = FeatureConfig()
+        config.enable_feature(FeatureType.AGGREGATION)
+        config.enable_feature(FeatureType.MOM, mom_windows=[5])    # 5个月环比
+        config.enable_feature(FeatureType.YOY, yoy_windows=[12])  # 1年同比
+        
+        print(f"   🔧 生成{schema.table_name}的特征...")
+        result = manager.generate_features(
+            schema=schema,
+            config=config,
+            feature_date=feature_date
+        )
+        
+        feature_count = manager.count_features(schema, config)
+        print(f"   ✅ {table_type}表特征生成完成: {feature_count}个特征")
+        return result
+    
+    def _export_features(self, feature_date: str, output_path: str) -> Dict:
+        """导出所有特征表到指定路径"""
+        file_prefixes = {
+            'behavior': 'aum_behavior_features',
+            'asset_avg': 'aum_asset_avg_features',
+            'asset_config': 'aum_asset_config_features',
+            'monthly_stat': 'monthly_stat_features'
+        }
+        
+        results = {}
+        for table_type, file_prefix in file_prefixes.items():
+            print(f"   💾 导出{table_type}表...")
+            
+            # 构建特征表名
+            table_name = f"{self.schemas[table_type].table_name}_{feature_date}_f001"
+            
+            result = self.engine.download_table_data(
+                table_name=f"{self.database}.{table_name}",
+                output_path=f"{output_path}/{file_prefix}_{feature_date}.parquet",
+                mode="cluster"
+            )
+            
+            results[table_type] = result
+            print(f"   ✅ 导出 {table_type}: {result.get('status', 'unknown')}")
+        
+        return results
+    
+    def get_summary(self) -> Dict:
+        """获取示例摘要信息"""
+        summary = {
+            'database': self.database,
+            'tables': {},
+            'total_features': 0
+        }
+        
+        for table_type, schema in self.schemas.items():
+            try:
+                manager = FeatureManager(self.engine, self.database)
+                
+                if table_type == 'behavior':
+                    # A表只有原始拷贝和聚合特征
+                    config = FeatureConfig()
+                    config.enable_feature(FeatureType.RAW_COPY)
+                    config.enable_feature(FeatureType.AGGREGATION)
+                    config.disable_feature(FeatureType.MOM)
+                    config.disable_feature(FeatureType.YOY)
+                else:
+                    # B、C、D表包含完整特征：聚合+5个月MoM+1年YoY
+                    config = FeatureConfig()
+                    config.enable_feature(FeatureType.AGGREGATION, mom_windows=[5], yoy_windows=[12])
+                
+                feature_count = manager.count_features(schema, config)
+                summary['tables'][table_type] = {
+                    'table_name': schema.table_name,
+                    'field_count': len(schema.fields),
+                    'feature_count': feature_count,
+                    'features': {
+                        'total': feature_count,
+                        'aggregation': len(schema.fields),  # 估算
+                        'mom': len(schema.fields) * 5 if table_type != 'behavior' else 0,
+                        'yoy': len(schema.fields) * 1 if table_type != 'behavior' else 0
+                    }
+                }
+                summary['total_features'] += feature_count
+            except Exception as e:
+                # 在模拟模式下返回预估数量
+                base_fields = len(schema.fields)
+                if table_type == 'behavior':
+                    estimated_features = base_fields * 2  # 原始拷贝 + 聚合
+                    agg_count = base_fields
+                    mom_count = 0
+                    yoy_count = 0
+                else:
+                    estimated_features = base_fields * 8  # 聚合 + MoM + YoY 组合
+                    agg_count = base_fields
+                    mom_count = base_fields * 5
+                    yoy_count = base_fields * 1
+                
+                summary['tables'][table_type] = {
+                    'table_name': schema.table_name,
+                    'field_count': base_fields,
+                    'feature_count': estimated_features,
+                    'mode': 'estimated',
+                    'features': {
+                        'total': estimated_features,
+                        'aggregation': agg_count,
+                        'mom': mom_count,
+                        'yoy': yoy_count
+                    }
+                }
+                summary['total_features'] += estimated_features
+        
+        return summary
+
+
+# 简化的API函数
+def create_aum_example(database: str = "dwegdata03000") -> AUMLongtailExample:
+    """
+    一键创建AUM长尾模型示例
+    
+    Args:
+        database: 数据库名称，默认为dwegdata03000
+        
+    Returns:
+        AUMLongtailExample实例
+    """
+    return AUMLongtailExample(database)
+
+
+def run_aum_example(feature_date: Optional[str] = None, 
+                   database: str = "dwegdata03000",
+                   output_path: str = "file:///nfsHome/aum_longtail") -> Dict:
+    """
+    一键运行AUM长尾模型特征工程
+    
+    Args:
+        feature_date: 特征日期，格式为YYYYMM，默认为当前月
+        database: 数据库名称，默认为dwegdata03000
+        output_path: 输出路径，默认为file:///nfsHome/aum_longtail
+        
+    Returns:
+        包含所有结果的字典
+        
+    Example:
+        >>> results = run_aum_example('202507')
+        >>> print(f"生成特征数: {len(results)}")
+    """
+    example = create_aum_example(database)
+    return example.run(feature_date, output_path)
+
+
+__all__ = [
+    'AUMLongtailExample',
+    'create_aum_example', 
+    'run_aum_example'
+]