V0.9.46 更新一批代码 (#190)

* 0.9.46 start coding

* 0.9.46 PSI 测试开发

* 0.9.46 update

* 0.9.46 新增 optuna 超参分析

* 0.9.46 disk cache 增加默认path

* 0.9.46 新增 rolling_tanh 函数

* 0.9.46 新增CCF因子函数

* 0.9.46 新增最大回撤分析组件

* 0.9.46 新增期货调仓函数

Author: zengbin93

.github/workflows/pythonpackage.yml

@@ -5,7 +5,7 @@ name: Python package
 
 on:
   push:
-    branches: [ master, V0.9.45 ]
+    branches: [ master, V0.9.46 ]
   pull_request:
     branches: [ master ]
 

README.md

@@ -58,7 +58,7 @@ pip install [email protected]:waditu/czsc.git -U
 
 直接从github指定分支安装最新版：
 ```
-pip install git+https://github.com/waditu/[email protected].41 -U
+pip install git+https://github.com/waditu/[email protected].46 -U
 ```
 
 从`pypi`安装：

czsc/__init__.py

@@ -45,6 +45,8 @@
     ExitsOptimize,
 )
 from czsc.utils import (
+    format_standard_kline,
+
     KlineChart,
     WordWriter,
     BarGenerator,
@@ -81,6 +83,7 @@
     holds_performance,
     net_value_stats,
     subtract_fee,
+    top_drawdowns,
 
     home_path,
     DiskCache,
@@ -94,6 +97,9 @@
     DataClient,
     set_url_token,
     get_url_token,
+
+    optuna_study,
+    optuna_good_params,
 )
 
 # 交易日历工具
@@ -121,6 +127,8 @@
     show_stoploss_by_direction,
     show_cointegration,
     show_out_in_compare,
+    show_optuna_study,
+    show_drawdowns,
 )
 
 from czsc.utils.bi_info import (
@@ -144,12 +152,14 @@
     rolling_compare,
     rolling_scale,
     rolling_slope,
+    rolling_tanh,
+    feature_adjust,
 )
 
-__version__ = "0.9.45"
+__version__ = "0.9.46"
 __author__ = "zengbin93"
 __email__ = "[email protected]"
-__date__ = "20240308"
+__date__ = "20240318"
 
 
 def welcome():

czsc/connectors/research.py

@@ -45,6 +45,7 @@ def get_raw_bars(symbol, freq, sdt, edt, fq='前复权', **kwargs):
     :param kwargs:
     :return:
     """
+    raw_bars = kwargs.get('raw_bars', True)
     kwargs['fq'] = fq
     file = glob.glob(os.path.join(cache_path, "*", f"{symbol}.parquet"))[0]
     freq = czsc.Freq(freq)
@@ -54,5 +55,5 @@ def get_raw_bars(symbol, freq, sdt, edt, fq='前复权', **kwargs):
     kline = kline[(kline['dt'] >= pd.to_datetime(sdt)) & (kline['dt'] <= pd.to_datetime(edt))]
     if kline.empty:
         return []
-    _bars = czsc.resample_bars(kline, freq, raw_bars=True, base_freq='1分钟')
+    _bars = czsc.resample_bars(kline, freq, raw_bars=raw_bars, base_freq='1分钟')
     return _bars

czsc/connectors/tq_connector.py

@@ -204,7 +204,14 @@ def is_trade_time(trade_time: Optional[str] = None):
 
 
 def get_daily_backup(api: TqApi, **kwargs):
-    """获取每日账户中需要备份的信息"""
+    """获取每日账户中需要备份的信息
+
+    https://doc.shinnytech.com/tqsdk/latest/reference/tqsdk.objs.html?highlight=account#tqsdk.objs.Order
+    https://doc.shinnytech.com/tqsdk/latest/reference/tqsdk.objs.html?highlight=account#tqsdk.objs.Position
+    https://doc.shinnytech.com/tqsdk/latest/reference/tqsdk.objs.html?highlight=account#tqsdk.objs.Account
+
+    :param api: TqApi, 天勤API实例
+    """
     orders = api.get_order()
     trades = api.get_trade()
     position = api.get_position()
@@ -229,3 +236,61 @@ def get_daily_backup(api: TqApi, **kwargs):
         "account": account,
     }
     return backup
+
+
+def adjust_portfolio(api: TqApi, portfolio, account=None, **kwargs):
+    """调整账户组合
+
+    **注意：** 此函数会阻塞，直到调仓完成；使用前请仔细阅读 TargetPosTask 的源码和文档，确保了解其工作原理
+
+    :param api: TqApi, 天勤API实例
+    :param account: str, 天勤账户
+    :param portfolio: dict, 组合配置，key 为合约代码，value 为配置信息; 样例数据：
+
+        {
+            "[email protected]": {"target_volume": 10, "price": "PASSIVE", "offset_priority": "今昨,开"},
+            "[email protected]": {"target_volume": 0, "price": "ACTIVE", "offset_priority": "今昨,开"},
+            "[email protected]": {"target_volume": 30, "price": "PASSIVE", "offset_priority": "今昨,开"}
+        }
+
+    :param kwargs: dict, 其他参数
+    """
+    symbol_infos = {}
+    for symbol, conf in portfolio.items():
+        quote = api.get_quote(symbol)
+
+        lots = conf.get("target_volume", 0)
+        price = conf.get("price", "PASSIVE")
+        offset_priority = conf.get("offset_priority", "今昨,开")
+
+        # 踩坑记录：TargetPosTask 的 symbol 必须是合约代码
+        contract = quote.underlying_symbol if "@" in symbol else symbol
+        target_pos = TargetPosTask(api, contract, price=price, offset_priority=offset_priority, account=account)
+        target_pos.set_target_volume(int(lots))
+        symbol_infos[symbol] = {"quote": quote, "target_pos": target_pos, "lots": lots}
+
+    while True:
+        api.wait_update()
+
+        completed = []
+        for symbol, info in symbol_infos.items():
+            quote = info["quote"]
+            target_pos: TargetPosTask = info["target_pos"]
+            lots = info["lots"]
+            contract = quote.underlying_symbol if "@" in symbol else symbol
+
+            logger.info(f"调整仓位：{quote.datetime} - {contract}; 目标持仓：{lots}手; 当前持仓：{target_pos._pos.pos}手")
+
+            if target_pos._pos.pos == lots:
+                completed.append(True)
+                logger.info(f"调仓完成：{quote.datetime} - {contract}; {lots}手")
+            else:
+                completed.append(False)
+
+        if all(completed):
+            break
+
+    if kwargs.get("close_api", True):
+        api.close()
+
+    return api

czsc/features/__init__.py

@@ -24,4 +24,8 @@
     VPF002,
     VPF003,
     VPF004,
+)
+
+from .tas import (
+    CCF
 )

czsc/features/tas.py

@@ -0,0 +1,51 @@
+"""
+技术指标因子
+"""
+import inspect
+import hashlib
+import pandas as pd
+
+
+def CCF(df, **kwargs):
+    """使用 CZSC 库中的 factor 识别因子，主要用于识别缠论/形态因子
+
+    :param df: 标准K线数据，DataFrame结构
+    :param kwargs: 其他参数
+
+        - czsc_factor: dict, 缠论因子配置，样例：
+
+            {
+                "signals_all": ["日线_D1_表里关系V230101_向上_任意_任意_0"],
+                "signals_any": [],
+                "signals_not": ["日线_D1_涨跌停V230331_涨停_任意_任意_0"],
+            }
+
+        - freq: str, default '日线'，K线级别
+        - tag: str, default None，标签，用于区分不同的因子
+
+    :return: pd.DataFrame
+    """
+    from czsc.objects import Factor
+    from czsc.utils import format_standard_kline
+    from czsc.traders.base import generate_czsc_signals
+    from czsc.traders.sig_parse import get_signals_config
+
+    czsc_factor = kwargs.get('czsc_factor', None)
+    freq = kwargs.get('freq', '日线')
+    assert czsc_factor is not None and isinstance(czsc_factor, dict), "factor 参数必须指定"
+    tag = kwargs.get('tag', hashlib.sha256(f"{czsc_factor}_{freq}".encode()).hexdigest().upper()[:6])
+
+    factor_name = inspect.stack()[0][3]
+    factor_col = f'F#{factor_name}#{tag}'
+
+    czsc_factor = Factor.load(czsc_factor)
+    signals_seq = czsc_factor.signals_all + czsc_factor.signals_any + czsc_factor.signals_not
+    signals_config = get_signals_config([x.signal for x in signals_seq])
+
+    bars = format_standard_kline(df, freq=freq)
+    dfs = generate_czsc_signals(bars, signals_config, init_n=300, sdt=bars[0].dt, df=True)
+    dfs[factor_col] = dfs.apply(czsc_factor.is_match, axis=1).astype(int)
+
+    df = pd.merge(df, dfs[['dt', factor_col]], on='dt', how='left')
+    df[factor_col] = df[factor_col].fillna(0)
+    return df

czsc/features/utils.py

@@ -187,6 +187,26 @@ def rolling_scale(df: pd.DataFrame, col: str, window=300, min_periods=100, new_c
     return df
 
 
+def rolling_tanh(df: pd.DataFrame, col: str, window=300, min_periods=100, new_col=None, **kwargs):
+    """对序列进行滚动 tanh 变换
+
+    双曲正切函数：https://baike.baidu.com/item/%E5%8F%8C%E6%9B%B2%E6%AD%A3%E5%88%87%E5%87%BD%E6%95%B0/15469414
+
+    :param df: pd.DataFrame, 待计算的数据
+    :param col: str, 待计算的列
+    :param window: int, 滚动窗口大小, 默认为300
+    :param min_periods: int, 最小计算周期, 默认为100
+    :param new_col: str, 新列名，默认为 None, 表示使用 f'{col}_scale' 作为新列名
+    """
+    if kwargs.get("copy", False):
+        df = df.copy()
+    new_col = new_col if new_col else f'{col}_tanh'
+    df = df.sort_values("dt", ascending=True).reset_index(drop=True)
+    df[new_col] = df[col].rolling(window=window, min_periods=min_periods).apply(lambda x: np.tanh(scale(x))[-1])    # type: ignore
+    df[new_col] = df[new_col].fillna(0)
+    return df
+
+
 def rolling_slope(df: pd.DataFrame, col: str, window=300, min_periods=100, new_col=None, **kwargs):
     """计算序列的滚动斜率
 
@@ -234,3 +254,77 @@ def __lr_slope(x):
 
     df[new_col] = df[new_col].fillna(0)
     return df
+
+
+def feature_adjust_V230101(df: pd.DataFrame, fcol, **kwargs):
+    """特征调整函数：对特征进行调整，使其符合持仓权重的定义
+
+    方法说明：对因子进行滚动相关系数计算，然后对因子值用 maxabs_scale 进行归一化，最后乘以滚动相关系数的符号
+
+    :param df: pd.DataFrame, 必须包含 dt、symbol、price 列，以及因子列
+    :param fcol: str 因子列名
+    :param kwargs: dict
+    """
+    window = kwargs.get("window", 1000)
+    min_periods = kwargs.get("min_periods", 200)
+
+    df = df.copy().sort_values("dt", ascending=True).reset_index(drop=True)
+    df['n1b'] = df['price'].shift(-1) / df['price'] - 1
+    df['corr'] = df[fcol].rolling(window=window, min_periods=min_periods).corr(df['n1b'])
+    df['corr'] = df['corr'].shift(5).fillna(0)
+
+    df = rolling_scale(df, col=fcol, window=window, min_periods=min_periods,
+                       new_col='weight', method='maxabs_scale', copy=True)
+    df['weight'] = df['weight'] * np.sign(df['corr'])
+
+    df.drop(['n1b', 'corr'], axis=1, inplace=True)
+    return df
+
+
+def feature_adjust_V240323(df: pd.DataFrame, fcol, **kwargs):
+    """特征调整函数：对特征进行调整，使其符合持仓权重的定义
+
+    方法说明：对因子进行滚动相关系数计算，然后对因子值用 scale + tanh 进行归一化，最后乘以滚动相关系数的符号
+
+    :param df: pd.DataFrame, 必须包含 dt、symbol、price 列，以及因子列
+    :param fcol: str 因子列名
+    :param kwargs: dict
+    """
+    window = kwargs.get("window", 1000)
+    min_periods = kwargs.get("min_periods", 200)
+
+    df = df.copy().sort_values("dt", ascending=True).reset_index(drop=True)
+    df['n1b'] = df['price'].shift(-1) / df['price'] - 1
+    df['corr'] = df[fcol].rolling(window=window, min_periods=min_periods).corr(df['n1b'])
+    df['corr'] = df['corr'].shift(5).fillna(0)
+
+    df = rolling_tanh(df, col=fcol, window=window, min_periods=min_periods, new_col='weight')
+    df['weight'] = df['weight'] * np.sign(df['corr'])
+
+    df.drop(['n1b', 'corr'], axis=1, inplace=True)
+    return df
+
+
+def feature_adjust(df: pd.DataFrame, fcol, method, **kwargs):
+    """特征调整函数：对特征进行调整，使其符合持仓权重的定义
+
+    :param df: pd.DataFrame, 待调整的数据
+    :param fcol: str, 因子列名
+    :param method: str, 调整方法
+
+        - V230101: 对因子进行滚动相关系数计算，然后对因子值用 maxabs_scale 进行归一化，最后乘以滚动相关系数的符号
+        - V240323: 对因子进行滚动相关系数计算，然后对因子值用 scale + tanh 进行归一化，最后乘以滚动相关系数的符号
+
+    :param kwargs: dict
+
+        - window: int, 滚动窗口大小
+        - min_periods: int, 最小计算周期
+
+    :return: pd.DataFrame, 新增 weight 列
+    """
+    if method == "V230101":
+        return feature_adjust_V230101(df, fcol, **kwargs)
+    elif method == "V240323":
+        return feature_adjust_V240323(df, fcol, **kwargs)
+    else:
+        raise ValueError(f"Unknown method: {method}")

czsc/utils/__init__.py

@@ -10,20 +10,21 @@
 from .echarts_plot import kline_pro, heat_map
 from .word_writer import WordWriter
 from .corr import nmi_matrix, single_linear, cross_sectional_ic
-from .bar_generator import BarGenerator, freq_end_time, resample_bars
+from .bar_generator import BarGenerator, freq_end_time, resample_bars, format_standard_kline
 from .bar_generator import is_trading_time, get_intraday_times, check_freq_and_market
 from .io import dill_dump, dill_load, read_json, save_json
 from .sig import check_pressure_support, check_gap_info, is_bis_down, is_bis_up, get_sub_elements, is_symmetry_zs
 from .sig import same_dir_counts, fast_slow_cross, count_last_same, create_single_signal
 from .plotly_plot import KlineChart
 from .trade import cal_trade_price, update_nbars, update_bbars, update_tbars, risk_free_returns, resample_to_daily
 from .cross import CrossSectionalPerformance, cross_sectional_ranker
-from .stats import daily_performance, net_value_stats, subtract_fee, weekly_performance, holds_performance
+from .stats import daily_performance, net_value_stats, subtract_fee, weekly_performance, holds_performance, top_drawdowns
 from .signal_analyzer import SignalAnalyzer, SignalPerformance
 from .cache import home_path, get_dir_size, empty_cache_path, DiskCache, disk_cache
 from .index_composition import index_composition
 from .data_client import DataClient, set_url_token, get_url_token
 from .oss import AliyunOSS
+from .optuna import optuna_study, optuna_good_params
 
 
 sorted_freqs = ['Tick', '1分钟', '2分钟', '3分钟', '4分钟', '5分钟', '6分钟', '10分钟', '12分钟',

czsc/utils/bar_generator.py

@@ -39,6 +39,30 @@ def get_intraday_times(freq='1分钟', market="A股"):
     return freq_market_times[f"{freq}_{market}"]
 
 
+def format_standard_kline(df: pd.DataFrame, freq: str):
+    """格式化标准K线数据为 CZSC 标准数据结构 RawBar 列表
+
+    :param df: 标准K线数据，DataFrame结构
+
+        ===================  =========  ======  =======  ======  =====  ===========  ===========
+        dt                   symbol       open    close    high    low          vol       amount
+        ===================  =========  ======  =======  ======  =====  ===========  ===========
+        2023-11-17 00:00:00  689009.SH   33.52    33.41   33.69  33.38  1.97575e+06  6.61661e+07
+        2023-11-20 00:00:00  689009.SH   33.4     32.91   33.45  32.25  5.15016e+06  1.68867e+08
+        ===================  =========  ======  =======  ======  =====  ===========  ===========
+
+    :param freq: K线级别
+    :return: list of RawBar
+    """
+    # from czsc.objects import RawBar, Freq
+    bars = []
+    for i, row in df.iterrows():
+        bar = RawBar(id=i, symbol=row['symbol'], dt=row['dt'], open=row['open'], close=row['close'],
+                     high=row['high'], low=row['low'], vol=row['vol'], amount=row['amount'], freq=Freq(freq))
+        bars.append(bar)
+    return bars
+
+
 def check_freq_and_market(time_seq: List[AnyStr], freq: Optional[AnyStr] = None):
     """检查时间序列是否为同一周期，是否为同一市场