qlib: # 数据路径 provider_uri: /Users/jiewen/PyCharmMiscProject/ML-Alpha-Research-System/qlib_data_30min_868_hfq_bin region: cn market: all data: freq: 30min start_time: "2000-01-01" end_time: "2026-06-30" fields: - $open - $close - $high - $low - $volume - $vwap # qlib 标准 label 表达式(与 Alpha158 一致,T+1 open -> T+2 close) label_expr: "Ref($close, -2)/Ref($close, -1) - 1" splits: train: start: "2000-01-01" end: "2021-12-31" valid: start: "2022-01-01" end: "2023-12-31" test: start: "2024-01-01" end: "2025-06-01" holdout: start: "2025-06-02" end: "2026-06-30" label: forward_days: 5 gp_forward_steps: 5 dump: # 原始 CSV/Parquet → qlib .bin 转换配置 raw_data_dir: data/raw/csv_by_symbol qlib_dir: data/qlib_data/cn_data include_fields: open,close,high,low,volume,factor date_field: date symbol_field: symbol file_suffix: .csv freq: day max_workers: 8 # 列名映射(适配你的 parquet 中文列名) column_aliases: 时间: date datetime: date date: date 代码: symbol symbol: symbol code: symbol 开盘价: open open: open 收盘价: close close: close 最高价: high high: high 最低价: low low: low 成交量: volume volume: volume 成交额: amount amount: amount factor: factor experiment: experiment_name: ml_alpha_research recorder_prefix: gp_lgb mlruns_uri: mlruns backtest: benchmark: null # 30min 数据无指数日线;有日线数据时可改 SH000300 freq: null # null = 使用 data.freq(当前 30min) topk: 50 n_drop: 5 account: 100000000 open_cost: 0.0005 close_cost: 0.0015 min_cost: 5 limit_threshold: 0.095 deal_price: close output: root: outputs reports_dir: outputs/reports