Initial commit

2025-06-10 01:16:09 +08:00
commit 05bab03b8c
155 changed files with 19979 additions and 0 deletions
@@ -0,0 +1,160 @@
+import json
+from typing import Dict, TypedDict
+import sys 
+import os
+sys.path.append(os.path.abspath("/Users/jack/Project/chan.py"))
+import xgboost as xgb
+
+from Chan import CChan
+from ChanConfig import CChanConfig
+from ChanModel.Features import CFeatures
+from Common.CEnum import AUTYPE, DATA_SRC, KL_TYPE, BSP_TYPE
+from Common.CTime import CTime
+from Plot.PlotDriver import CPlotDriver
+
+
+class T_SAMPLE_INFO(TypedDict):
+    feature: CFeatures
+    is_buy: bool
+    open_time: CTime
+
+
+def plot(chan, plot_marker):
+    plot_config = {
+        "plot_kline": True,
+        "plot_bi": True,
+        "plot_seg": True,
+        "plot_zs": True,
+        "plot_bsp": True,
+        "plot_marker": True,
+        "plot_macd": True,
+    }
+    plot_para = {
+        "figure": {
+            "x_range": 400,
+        },
+        "marker": {
+            "markers": plot_marker
+        }
+    }
+    plot_driver = CPlotDriver(
+        chan,
+        plot_config=plot_config,
+        plot_para=plot_para,
+    )
+    plot_driver.save2img("label.png")
+
+
+def stragety_feature(last_klu):
+    return {
+        "open_klu_rate": (last_klu.close - last_klu.open)/last_klu.open,
+    }
+
+
+if __name__ == "__main__":
+    """
+    本demo主要演示如何记录策略产出的买卖点的特征
+    然后将这些特征作为样本，训练一个模型(以XGB为demo)
+    用于预测买卖点的准确性
+
+    请注意，demo训练预测都用的是同一份数据，这是不合理的，仅仅是为了演示
+    """
+    code = "BTC/USDT"
+    begin_time = "2024-12-1"
+    end_time = None
+    data_src = DATA_SRC.CCXT
+    lv_list = [KL_TYPE.K_1M]
+
+    config = CChanConfig({
+        "trigger_step": True,  # 打开开关！
+        "bi_strict": True,
+        "skip_step": 0,
+        "divergence_rate": float("inf"),
+        "bsp2_follow_1": False,
+        "bsp3_follow_1": False,
+        "min_zs_cnt": 0,
+        "bs1_peak": False,
+        "macd_algo": "peak",
+        "bs_type": '1,2,3a,1p,2s,3b',
+        "print_warning": True,
+        "zs_algo": "normal",
+    })
+
+    chan = CChan(
+        code=code,
+        begin_time=begin_time,
+        end_time=end_time,
+        data_src=data_src,
+        lv_list=lv_list,
+        config=config,
+        autype=AUTYPE.QFQ,
+    )
+
+    bsp_dict: Dict[int, T_SAMPLE_INFO] = {}  # 存储策略产出的bsp的特征
+
+    # 跑策略，保存买卖点的特征
+    index = 0
+    for chan_snapshot in chan.step_load():
+        last_klu = chan_snapshot[0][-1][-1]
+        bsp_list = chan_snapshot.get_bsp()
+        if not bsp_list:
+            continue
+        last_bsp = bsp_list[-1]
+        cur_lv_chan = chan_snapshot[0]
+        if last_bsp.klu.idx not in bsp_dict and cur_lv_chan[-2].idx == last_bsp.klu.klc.idx and last_bsp.is_buy:
+            # 假如策略是：买卖点分形第三元素出现时交易
+            bsp_dict[last_bsp.klu.idx] = {
+                "feature": last_bsp.features,
+                "is_buy": last_bsp.is_buy,
+                "open_time": last_klu.time,
+            }
+            bsp_dict[last_bsp.klu.idx]['feature'].add_feat(stragety_feature(last_klu))  # 开仓K线特征
+            print(index, last_bsp.klu.time, last_bsp.is_buy)
+            index += 1
+
+    # 生成libsvm样本特征
+    bsp_academy = [bsp.klu.idx for bsp in chan.get_bsp()]
+    feature_meta = {}  # 特征meta
+    cur_feature_idx = 0
+    plot_marker = {}
+    fid = open("feature.libsvm", "w")
+    for bsp_klu_idx, feature_info in bsp_dict.items():
+        label = int(bsp_klu_idx in bsp_academy)  # 以买卖点识别是否准确为label
+        features = []  # List[(idx, value)]
+        for feature_name, value in feature_info['feature'].items():
+            if feature_name not in feature_meta:
+                feature_meta[feature_name] = cur_feature_idx
+                cur_feature_idx += 1
+            features.append((feature_meta[feature_name], value))
+        features.sort(key=lambda x: x[0])
+        feature_str = " ".join([f"{idx}:{value}" for idx, value in features])
+        fid.write(f"{label} {feature_str}\n")
+        plot_marker[feature_info["open_time"].to_str()] = ("√" if label else "×", "down" if feature_info["is_buy"] else "up")
+    fid.close()
+
+    with open("feature.meta", "w") as fid:
+        # meta保存下来，实盘预测时特征对齐用
+        fid.write(json.dumps(feature_meta))
+
+    # 画图检查label是否正确
+    plot(chan, plot_marker)
+
+    # load sample file & train model
+    dtrain = xgb.DMatrix("feature.libsvm?format=libsvm")  # load sample
+    param = {'max_depth': 2, 'eta': 0.3, 'objective': 'binary:logistic', 'eval_metric': 'auc'}
+    evals_result = {}
+    bst = xgb.train(
+        param,
+        dtrain=dtrain,
+        num_boost_round=50,
+        evals=[(dtrain, "train")],
+        evals_result=evals_result,
+        verbose_eval=True,
+    )
+    bst.save_model("buy_model.json")
+
+    # load model
+    model = xgb.Booster()
+    model.load_model("buy_model.json")
+    # predict
+    print(model.predict(dtrain))