File size: 5,987 Bytes
73b57d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
"""
阶段1:被动单聚合。

利用上交所 L2 语义——被动方 order_id 才是真实的挂单 ID:
  - active_buy  → ask_order_id(卖方限价单被吃掉)
  - active_sell → bid_order_id(买方限价单被吃掉)

按被动方 order_id 聚合,提取每笔被动挂单的行为指纹特征。
"""

from __future__ import annotations

import numpy as np
import pandas as pd


def compute_vwap(trades: pd.DataFrame) -> float:
    """从 trades 计算当日 VWAP(成交量加权均价)。"""
    valid = trades
    if "is_cancellation" in trades.columns:
        valid = trades[~trades["is_cancellation"]]
    if valid.empty:
        return np.nan
    qty = valid["qty"].astype(float)
    return float((valid["price"] * qty).sum() / qty.sum())


def _compute_amount(trades: pd.DataFrame) -> pd.Series:
    """成交额 = price * qty(yuan)。"""
    return trades["price"].astype(float) * trades["qty"].astype(float)


def extract_passive_orders(
    trades: pd.DataFrame,
    vwap: float | None = None,
) -> pd.DataFrame:
    """
    从单日 trades 提取被动单特征。

    Args:
        trades: 单日逐笔成交 DataFrame。
                必须有列: bs_flag_desc, bid_order_id, ask_order_id,
                          price, amount, time_ms
        vwap: 当日 VWAP。为 None 则自动计算。

    Returns:
        DataFrame,每行一个被动 order_id,含行为指纹特征。
    """
    if vwap is None or np.isnan(vwap):
        vwap = compute_vwap(trades)

    # 过滤撤单 + 空 bs_flag_desc(集合竞价/中性)
    if "is_cancellation" in trades.columns:
        trades = trades[~trades["is_cancellation"]]
    trades = trades[trades["bs_flag_desc"].isin(["active_buy", "active_sell"])].copy()

    # 计算成交额
    trades["_amount"] = _compute_amount(trades)

    rows = []

    # active_buy → ask_order_id 是真正的卖方挂单(被动卖出)
    buy_trades = trades[trades["bs_flag_desc"] == "active_buy"]
    rows.extend(_aggregate_side(buy_trades, "ask_order_id", "ask", vwap))

    # active_sell → bid_order_id 是真正的买方挂单(被动买入=承接)
    sell_trades = trades[trades["bs_flag_desc"] == "active_sell"]
    rows.extend(_aggregate_side(sell_trades, "bid_order_id", "bid", vwap))

    df = pd.DataFrame(rows)
    if df.empty:
        return df

    # 日内时段归一化 [0, 1]:9:30=0, 15:00=1
    df["time_centroid"] = (df["time_centroid_raw"] - 34_200_000) / (
        54_000_000 - 34_200_000
    )
    df["time_centroid"] = df["time_centroid"].clip(0, 1)

    # 价格偏离 (bps)
    if vwap and not np.isnan(vwap):
        df["avg_price_bps"] = ((df["avg_price"] / vwap) - 1) * 10000
    else:
        df["avg_price_bps"] = 0.0

    # 对数额外特征
    df["log_amount"] = np.log1p(df["total_amount"])
    df["log_duration"] = np.log1p(df["duration_sec"])
    df["log_trade_count"] = np.log1p(df["trade_count"])

    # 过滤无意义的单笔单(无节奏信息)
    df["cv_interval"] = df["cv_interval"].fillna(0)

    return df


def _aggregate_side(
    trades: pd.DataFrame,
    passive_col: str,
    side: str,
    vwap: float,
) -> list[dict]:
    """聚合某个方向(bid 或 ask)的被动单。"""
    rows = []
    for oid, grp in trades.groupby(passive_col):
        if oid == 0:
            continue  # 跳过无效 order_id
        grp = grp.sort_values("time_ms")
        times = grp["time_ms"].values.astype(float)
        prices = grp["price"].values.astype(float)
        qtys = grp["qty"].values.astype(float)
        amounts = grp["_amount"].values.astype(float)

        # 成交间隔
        if len(times) > 1:
            intervals = np.diff(times) / 1000.0  # 秒
            cv = float(intervals.std() / intervals.mean()) if intervals.mean() > 0 else 0.0
        else:
            cv = 0.0

        # VWAP for this passive order (weighted by qty)
        total_qty = qtys.sum()
        order_vwap = float(np.average(prices, weights=qtys)) if total_qty > 0 else prices.mean()

        rows.append({
            "order_id": int(oid),
            "side": side,
            "total_amount": float(amounts.sum()),
            "total_qty": float(total_qty),
            "avg_price": order_vwap,
            "trade_count": len(grp),
            "time_centroid_raw": float(np.average(times, weights=amounts)) if amounts.sum() > 0 else float(times.mean()),
            "first_time_ms": int(times.min()),
            "last_time_ms": int(times.max()),
            "duration_sec": float((times.max() - times.min()) / 1000.0),
            "cv_interval": cv,
            "price_min": float(prices.min()),
            "price_max": float(prices.max()),
            "price_range_bps": float((prices.max() - prices.min()) / vwap * 10000) if vwap and vwap > 0 else 0.0,
        })

    return rows


def select_candidates(
    passive_orders: pd.DataFrame,
    top_n: int = 150,
) -> pd.DataFrame:
    """取成交额 top N 的被动单作为主力候选。"""
    if passive_orders.empty:
        return passive_orders
    return passive_orders.nlargest(top_n, "total_amount").copy()


# 聚类 & 匹配用的 7 维特征列
FEATURE_COLS = [
    "log_amount",
    "avg_price_bps",
    "time_centroid",
    "log_duration",
    "cv_interval",
    "side_num",      # 0=bid, 1=ask — clustering 阶段填入
    "price_range_bps",
]


def prepare_features(candidates: pd.DataFrame) -> np.ndarray:
    """将候选 DataFrame 转为 (N, 7) 聚类特征矩阵。"""
    df = candidates.copy()
    df["side_num"] = df["side"].map({"bid": 0, "ask": 1}).fillna(0.5)
    # 日级 min-max 归一化
    feats = df[FEATURE_COLS].values.astype(np.float64)
    # 处理 inf/nan
    feats = np.nan_to_num(feats, nan=0.0, posinf=0.0, neginf=0.0)
    # 每列 min-max
    f_min = feats.min(axis=0, keepdims=True)
    f_max = feats.max(axis=0, keepdims=True)
    denom = f_max - f_min
    denom[denom == 0] = 1.0
    feats = (feats - f_min) / denom
    return feats