669 lines
27 KiB
Python
669 lines
27 KiB
Python
"""
|
||
逆向分析模块 — 从涨幅榜复盘，提取起爆前共性特征，发现新规律
|
||
|
||
核心逻辑：
|
||
1. 拉Binance 24h涨幅榜Top N，把“已经涨起来的币”只作为事后标签。
|
||
2. 对未被推荐的暴涨币，截掉最近24h起爆段，只回溯其起爆前/启动点K线。
|
||
3. 用full_pa_analysis()提取涨前特征：连续K、起爆点、供需区(Q≥7)、静K蓄力、量价模式。
|
||
4. 检查板块联动（同板块是否有其他币也暴涨），并用未大涨高成交额样本做对照组。
|
||
5. 统计涨前共性特征占比和对照组lift，只生成候选规则，不能直接变成追涨买入依据。
|
||
6. 返回结构化结果供feishu推送和策略迭代中心展示。
|
||
"""
|
||
|
||
import json
|
||
import os
|
||
import sys
|
||
import time
|
||
from datetime import datetime, timedelta
|
||
from collections import defaultdict, Counter
|
||
|
||
import requests
|
||
import pandas as pd
|
||
|
||
sys.path.insert(0, os.path.dirname(__file__))
|
||
from app.db.altcoin_db import get_conn, record_missed_explosion, upsert_strategy_rule_candidate
|
||
from app.core.pa_engine import full_pa_analysis, classify_candles, calc_atr
|
||
from app.core.sector_map import get_sector_for_coin, COIN_TO_SECTORS, SECTOR_MEMBERS
|
||
from app.config import config_loader
|
||
from app.services.altcoin_screener import (
|
||
STABLECOINS,
|
||
WRAPPED,
|
||
GOLD_METAL,
|
||
EXCLUDED_BASES,
|
||
EXCLUDED_BASE_SUFFIXES,
|
||
)
|
||
|
||
BINANCE_API = "https://api.binance.com/api/v3"
|
||
|
||
|
||
FEATURE_LABELS = {
|
||
"has_ignition_point": "起爆点(静K→动K转折)",
|
||
"has_q7_zone": "Q≥7高质量供需区",
|
||
"has_q7_demand_nearby": "Q≥7需求区在起爆价附近(<3%)",
|
||
"has_continuous_k": "连续K多头加速",
|
||
"has_static_accumulation": "静K蓄力(占比≥15%)",
|
||
"has_volume_surge_before": "起爆前2倍+放量",
|
||
"has_bullish_breakout_pattern": "连续5K收盘价抬高",
|
||
}
|
||
|
||
|
||
def _is_altcoin_usdt_symbol(symbol_str):
|
||
if not symbol_str or not symbol_str.endswith("USDT"):
|
||
return False
|
||
base = symbol_str.replace("USDT", "").upper()
|
||
if base in STABLECOINS or base in WRAPPED or base in GOLD_METAL:
|
||
return False
|
||
if base in EXCLUDED_BASES or base.endswith(EXCLUDED_BASE_SUFFIXES):
|
||
return False
|
||
return base.isascii()
|
||
|
||
|
||
# ==================== 数据获取 ====================
|
||
|
||
def fetch_24h_tickers():
|
||
"""获取Binance所有USDT交易对的24h行情"""
|
||
try:
|
||
resp = requests.get(f"{BINANCE_API}/ticker/24hr", timeout=15)
|
||
if resp.status_code != 200:
|
||
return []
|
||
return resp.json()
|
||
except Exception as e:
|
||
print(f"[reverse_analysis] fetch_24h_tickers failed: {e}")
|
||
return []
|
||
|
||
|
||
def fetch_klines_before(symbol, lookback_hours=72, interval="1h"):
|
||
"""
|
||
获取起爆前的K线数据
|
||
取最近(lookback_hours+24)根1H K线，截掉最后24h（起爆段），只看起爆前的蓄力期
|
||
"""
|
||
total_limit = lookback_hours + 24 # 起爆前 + 起爆段
|
||
try:
|
||
resp = requests.get(f"{BINANCE_API}/klines", params={
|
||
"symbol": symbol.replace("/", ""),
|
||
"interval": interval,
|
||
"limit": total_limit,
|
||
}, timeout=10)
|
||
if resp.status_code != 200:
|
||
return [], []
|
||
raw = resp.json()
|
||
klines = [{"time": k[0], "open": float(k[1]), "high": float(k[2]),
|
||
"low": float(k[3]), "close": float(k[4]), "volume": float(k[5])}
|
||
for k in raw]
|
||
|
||
# 分割：前lookback_hours是起爆前，后24h是起爆段
|
||
explosion_start = max(0, len(klines) - 24)
|
||
pre_explosion = klines[:explosion_start]
|
||
explosion_segment = klines[explosion_start:]
|
||
|
||
return pre_explosion, explosion_segment
|
||
except Exception as e:
|
||
print(f"[reverse_analysis] fetch_klines_before({symbol}) failed: {e}")
|
||
return [], []
|
||
|
||
|
||
def get_recommended_symbols(hours=72):
|
||
"""获取过去N小时内被推荐过的币种列表"""
|
||
conn = get_conn()
|
||
cutoff = (datetime.now() - timedelta(hours=float(hours or 72))).isoformat()
|
||
rows = conn.execute("""
|
||
SELECT symbol FROM recommendation
|
||
WHERE rec_time >= %s
|
||
""", (cutoff,)).fetchall()
|
||
conn.close()
|
||
return set(r["symbol"] for r in rows)
|
||
|
||
|
||
# ==================== 特征提取 ====================
|
||
|
||
def extract_pre_explosion_features(symbol, pre_klines, explosion_klines, config):
|
||
"""
|
||
对起爆前K线做full_pa_analysis，提取特征字典
|
||
"""
|
||
features = {
|
||
"has_ignition_point": False,
|
||
"ignition_count": 0,
|
||
"ignition_details": [],
|
||
"has_q7_zone": False,
|
||
"q7_zones": [],
|
||
"q_max": 0,
|
||
"has_continuous_k": False,
|
||
"continuous_k_count": 0,
|
||
"continuous_k_details": [],
|
||
"has_static_accumulation": False,
|
||
"static_k_count": 0,
|
||
"static_ratio": 0.0,
|
||
"has_volume_surge_before": False,
|
||
"vol_surge_ratio": 0.0,
|
||
"has_bullish_breakout_pattern": False,
|
||
}
|
||
|
||
if not pre_klines or len(pre_klines) < 30:
|
||
return features
|
||
|
||
# 转DataFrame做PA分析
|
||
df = pd.DataFrame(pre_klines)
|
||
df["time"] = pd.to_datetime(df["time"], unit="ms")
|
||
pa_result = full_pa_analysis(df, timeframe="1h")
|
||
|
||
feature_config = config.get("feature_extraction", {})
|
||
|
||
# 1. 起爆点特征
|
||
if feature_config.get("check_ignition", True):
|
||
ignition_points = pa_result.get("ignition_points", [])
|
||
bullish_ignitions = [ip for ip in ignition_points if ip["direction"] == 1]
|
||
features["has_ignition_point"] = len(bullish_ignitions) > 0
|
||
features["ignition_count"] = len(bullish_ignitions)
|
||
features["ignition_details"] = bullish_ignitions
|
||
|
||
# 2. 供需区特征（特别是Q≥7）
|
||
if feature_config.get("check_supply_demand", True):
|
||
zones = pa_result.get("zones", [])
|
||
q7_zones = [z for z in zones if z["q_score"] >= 7]
|
||
demand_q7 = [z for z in q7_zones if z["type"] == "demand"]
|
||
features["has_q7_zone"] = len(q7_zones) > 0
|
||
features["q7_zones"] = q7_zones
|
||
features["q_max"] = max((z["q_score"] for z in zones), default=0)
|
||
# 特别标注：是否有Q≥7需求区在起爆前价格附近
|
||
if demand_q7 and len(pre_klines) > 0:
|
||
last_price = pre_klines[-1]["close"]
|
||
nearby_demand = [z for z in demand_q7 if abs(z["top"] - last_price) / last_price < 0.03]
|
||
features["has_q7_demand_nearby"] = len(nearby_demand) > 0
|
||
|
||
# 3. 连续K加速特征
|
||
if feature_config.get("check_continuous_k", True):
|
||
continuous_k = pa_result.get("continuous_k", [])
|
||
bullish_cont = [ck for ck in continuous_k if ck["type"] == "bullish_continue"]
|
||
features["has_continuous_k"] = len(bullish_cont) > 0
|
||
features["continuous_k_count"] = len(bullish_cont)
|
||
features["continuous_k_details"] = bullish_cont
|
||
|
||
# 4. 静K蓄力特征
|
||
candles_class = pa_result.get("candles_class", [])
|
||
static_ks = [c for c in candles_class if c["type"] == "static"]
|
||
features["static_k_count"] = len(static_ks)
|
||
features["static_ratio"] = len(static_ks) / len(candles_class) if candles_class else 0
|
||
features["has_static_accumulation"] = features["static_ratio"] >= 0.15 # 静K占比≥15%
|
||
|
||
# 5. 起爆前放量特征
|
||
if feature_config.get("check_volume_pattern", True) and len(pre_klines) >= 20:
|
||
recent_10 = pre_klines[-10:]
|
||
older_10 = pre_klines[-20:-10]
|
||
avg_recent_vol = sum(k["volume"] for k in recent_10) / len(recent_10)
|
||
avg_older_vol = sum(k["volume"] for k in older_10) / len(older_10) if older_10 else 1
|
||
vol_surge_ratio = avg_recent_vol / avg_older_vol if avg_older_vol > 0 else 0
|
||
features["has_volume_surge_before"] = vol_surge_ratio >= 2.0
|
||
features["vol_surge_ratio"] = round(vol_surge_ratio, 2)
|
||
|
||
# 6. 多头突破模式（最后几根K线close连续抬高）
|
||
if len(pre_klines) >= 5:
|
||
last_5_closes = [k["close"] for k in pre_klines[-5:]]
|
||
bullish_breakout = all(last_5_closes[i] > last_5_closes[i - 1] for i in range(1, len(last_5_closes)))
|
||
features["has_bullish_breakout_pattern"] = bullish_breakout
|
||
|
||
# 7. 4H 和 1D 多周期分析（从config读取lookback_hours，默认168/720）
|
||
lookback_4h = config.get("lookback_hours_4h", 168)
|
||
lookback_1d = config.get("lookback_hours_1d", 720)
|
||
features["has_4h_ignition"] = False
|
||
features["has_daily_demand_zone"] = False
|
||
features["daily_trend_up"] = False
|
||
|
||
try:
|
||
# 4H K线分析
|
||
pre_4h, _ = fetch_klines_before(symbol, lookback_4h, interval="4h")
|
||
if pre_4h and len(pre_4h) >= 8:
|
||
df_4h = pd.DataFrame(pre_4h)
|
||
df_4h["time"] = pd.to_datetime(df_4h["time"], unit="ms")
|
||
pa_4h = full_pa_analysis(df_4h, timeframe="4h")
|
||
ignition_4h = pa_4h.get("ignition_points", [])
|
||
features["has_4h_ignition"] = any(ip["direction"] == 1 for ip in ignition_4h)
|
||
except Exception:
|
||
pass
|
||
|
||
try:
|
||
# 1D K线分析
|
||
pre_1d, _ = fetch_klines_before(symbol, lookback_1d, interval="1d")
|
||
if pre_1d and len(pre_1d) >= 5:
|
||
df_1d = pd.DataFrame(pre_1d)
|
||
df_1d["time"] = pd.to_datetime(df_1d["time"], unit="ms")
|
||
pa_1d = full_pa_analysis(df_1d, timeframe="1d")
|
||
zones_1d = pa_1d.get("zones", [])
|
||
demand_zones_1d = [z for z in zones_1d if z["type"] == "demand" and z["q_score"] >= 5]
|
||
features["has_daily_demand_zone"] = len(demand_zones_1d) > 0
|
||
|
||
# 日线趋势：最后5根日线close是否整体向上
|
||
closes_1d = [k["close"] for k in pre_1d[-5:]]
|
||
features["daily_trend_up"] = closes_1d[-1] > closes_1d[0] if len(closes_1d) >= 3 else False
|
||
except Exception:
|
||
pass
|
||
|
||
return features
|
||
|
||
|
||
def check_sector_alignment(symbol, top_gainers, config):
|
||
"""
|
||
检查板块联动：同板块其他币是否也在涨幅榜
|
||
返回: {sector_name, sector_coin_count, sector_top_gainer_count, is_sector_hot}
|
||
"""
|
||
if not config.get("feature_extraction", {}).get("check_sector_alignment", True):
|
||
return {"sector": "", "sector_coin_count": 0, "sector_top_gainer_count": 0, "is_sector_hot": False}
|
||
|
||
sectors = get_sector_for_coin(symbol)
|
||
if not sectors:
|
||
return {"sector": "未知", "sector_coin_count": 0, "sector_top_gainer_count": 0, "is_sector_hot": False}
|
||
|
||
# 主板块取第一个
|
||
primary_sector = sectors[0]
|
||
sector_coins = SECTOR_MEMBERS.get(primary_sector, [])
|
||
|
||
# 统计同板块有多少币在涨幅榜
|
||
sector_in_gainers = []
|
||
for g in top_gainers:
|
||
if g["symbol"] in sector_coins:
|
||
sector_in_gainers.append(g)
|
||
|
||
# 板块联动阈值：≥3只同板块币涨幅榜 → 板块热
|
||
is_hot = len(sector_in_gainers) >= 3
|
||
|
||
return {
|
||
"sector": primary_sector,
|
||
"sector_coin_count": len(sector_coins),
|
||
"sector_top_gainer_count": len(sector_in_gainers),
|
||
"sector_gainer_symbols": [g["symbol"] for g in sector_in_gainers],
|
||
"is_sector_hot": is_hot,
|
||
}
|
||
|
||
|
||
# ==================== 共性特征统计与规律发现 ====================
|
||
|
||
def compute_pattern_summary(all_features, total_count, control_features=None):
|
||
"""
|
||
统计所有top gainer起爆前窗口的共性特征占比
|
||
返回: [{feature_name, count, percentage, description}]
|
||
"""
|
||
if total_count == 0:
|
||
return []
|
||
|
||
feature_counts = Counter()
|
||
feature_details = defaultdict(list)
|
||
|
||
for feat in all_features:
|
||
for key, value in feat.items():
|
||
if key.startswith("has_") and value:
|
||
feature_counts[key] += 1
|
||
# 收集具体描述
|
||
detail_key = key.replace("has_", "") + "_details"
|
||
if detail_key in feat and feat[detail_key]:
|
||
feature_details[key].extend(feat[detail_key])
|
||
|
||
control_features = control_features or []
|
||
control_total = len(control_features)
|
||
control_counts = Counter()
|
||
for feat in control_features:
|
||
for key, value in feat.items():
|
||
if key.startswith("has_") and value:
|
||
control_counts[key] += 1
|
||
|
||
summary = []
|
||
for feat_key, count in feature_counts.most_common():
|
||
pct = round(count / total_count * 100, 1)
|
||
control_count = control_counts.get(feat_key, 0)
|
||
control_pct = round(control_count / control_total * 100, 1) if control_total else 0.0
|
||
lift = round((pct + 1.0) / (control_pct + 1.0), 2) if control_total else 0.0
|
||
label = FEATURE_LABELS.get(feat_key, feat_key)
|
||
summary.append({
|
||
"feature": feat_key,
|
||
"label": label,
|
||
"count": count,
|
||
"percentage": pct,
|
||
"control_count": control_count,
|
||
"control_percentage": control_pct,
|
||
"lift": lift,
|
||
"total": total_count,
|
||
"control_total": control_total,
|
||
})
|
||
|
||
# 补充：即使未达到阈值的特征也记录（只是不触发learned_rule）
|
||
for feat_key, label in FEATURE_LABELS.items():
|
||
if feat_key not in feature_counts:
|
||
summary.append({
|
||
"feature": feat_key,
|
||
"label": label,
|
||
"count": 0,
|
||
"percentage": 0.0,
|
||
"control_count": control_counts.get(feat_key, 0),
|
||
"control_percentage": round(control_counts.get(feat_key, 0) / control_total * 100, 1) if control_total else 0.0,
|
||
"lift": 0.0,
|
||
"total": total_count,
|
||
"control_total": control_total,
|
||
})
|
||
|
||
# 排序：优先看相对对照组的提升，再看涨幅榜占比
|
||
summary.sort(key=lambda x: (x.get("lift", 0), x["percentage"]), reverse=True)
|
||
return summary
|
||
|
||
|
||
def _rule_for_feature(feat_key, pct, lift=0):
|
||
lift_note = f"，相对对照组提升{lift}x" if lift else ""
|
||
if feat_key == "has_ignition_point":
|
||
return {
|
||
"type": "bonus",
|
||
"description": f"涨幅榜{pct}%有起爆点(静K→动K){lift_note} → 起爆点是爆发前候选信号",
|
||
"conditions": {"has_ignition_point": True},
|
||
"score_adjust": 2,
|
||
"source": "reverse_analysis",
|
||
}
|
||
if feat_key == "has_q7_zone":
|
||
return {
|
||
"type": "bonus",
|
||
"description": f"涨幅榜{pct}%有Q≥7供需区{lift_note} → 高质量供需区是爆发支撑候选",
|
||
"conditions": {"has_q7_zone": True},
|
||
"score_adjust": 2,
|
||
"source": "reverse_analysis",
|
||
}
|
||
if feat_key == "has_q7_demand_nearby":
|
||
return {
|
||
"type": "bonus",
|
||
"description": f"涨幅榜{pct}%有Q≥7需求区在起爆价附近{lift_note} → 需求区支撑是爆发候选因子",
|
||
"conditions": {"has_q7_demand_nearby": True},
|
||
"score_adjust": 3,
|
||
"source": "reverse_analysis",
|
||
}
|
||
if feat_key == "has_continuous_k":
|
||
return {
|
||
"type": "bonus",
|
||
"description": f"涨幅榜{pct}%有连续K多头加速{lift_note} → 趋势加速是爆发前兆候选",
|
||
"conditions": {"has_continuous_k_bullish": True},
|
||
"score_adjust": 2,
|
||
"source": "reverse_analysis",
|
||
}
|
||
if feat_key == "has_static_accumulation":
|
||
return {
|
||
"type": "bonus",
|
||
"description": f"涨幅榜{pct}%有静K蓄力(占比≥15%){lift_note} → 蓄力是爆发候选条件",
|
||
"conditions": {"static_ratio_min": 0.15},
|
||
"score_adjust": 1,
|
||
"source": "reverse_analysis",
|
||
}
|
||
if feat_key == "has_volume_surge_before":
|
||
return {
|
||
"type": "bonus",
|
||
"description": f"涨幅榜{pct}%起爆前2倍+放量{lift_note} → 量能先行是爆发预警候选",
|
||
"conditions": {"vol_surge_ratio_min": 2.0},
|
||
"score_adjust": 2,
|
||
"source": "reverse_analysis",
|
||
}
|
||
if feat_key == "has_bullish_breakout_pattern":
|
||
return {
|
||
"type": "bonus",
|
||
"description": f"涨幅榜{pct}%连续5K收盘价抬高{lift_note} → 价格加速是爆发前形态候选",
|
||
"conditions": {"has_bullish_breakout_pattern": True},
|
||
"score_adjust": 1,
|
||
"source": "reverse_analysis",
|
||
}
|
||
return None
|
||
|
||
|
||
def discover_new_rules(pattern_summary, all_features, sector_alignments, significance_pct=60.0, min_lift=1.5):
|
||
"""
|
||
当共性特征占比≥significance_pct时，自动生成learned_rule
|
||
返回: [{rule_id, description, conditions, score_adjust}]
|
||
"""
|
||
new_rules = []
|
||
total_analyzed = len(all_features)
|
||
significant_features = [
|
||
p for p in pattern_summary
|
||
if p["percentage"] >= significance_pct and (not p.get("control_total") or p.get("lift", 0) >= min_lift)
|
||
]
|
||
|
||
for pattern in significant_features:
|
||
feat_key = pattern["feature"]
|
||
pct = pattern["percentage"]
|
||
count = int(pattern.get("count") or 0)
|
||
lift = float(pattern.get("lift") or 0)
|
||
rule = _rule_for_feature(feat_key, pct, lift=lift)
|
||
if not rule:
|
||
continue
|
||
|
||
# 新体系：逆向分析只生成候选规则，不直接写 learned_rules，避免涨幅榜小样本污染已发布策略。
|
||
rule["candidate_id"] = upsert_strategy_rule_candidate(
|
||
source="reverse_analysis",
|
||
rule_type=rule.get("type", "bonus"),
|
||
signal_name=feat_key,
|
||
rule_description=rule.get("description", ""),
|
||
support_count=int(count),
|
||
success_count=int(count),
|
||
fail_count=int(pattern.get("control_count") or 0),
|
||
confidence_score=round(min(95, pct * max(lift, 1.0) / 2), 1),
|
||
sample_size=int(total_analyzed),
|
||
status="candidate",
|
||
notes=f"逆向涨幅榜规律，只分析起爆前窗口并做对照组校验(lift={lift})，用于提前埋伏候选，不能作为涨后追买依据，仍需等待推荐样本验证后再发布",
|
||
source_ref=f"reverse:{feat_key}",
|
||
)
|
||
new_rules.append(rule)
|
||
|
||
# 检查板块联动规律
|
||
hot_sectors = [sa for sa in sector_alignments if sa.get("is_sector_hot")]
|
||
if len(hot_sectors) >= 2:
|
||
# 多板块联动规律
|
||
sector_names = [sa["sector"] for sa in hot_sectors]
|
||
rule = {
|
||
"type": "bonus",
|
||
"description": f"多板块联动({', '.join(sector_names)}) → 市场整体情绪升温，蓄力币起爆概率高",
|
||
"conditions": {"multi_sector_hot": True, "sectors": sector_names},
|
||
"score_adjust": 2,
|
||
"source": "reverse_analysis",
|
||
}
|
||
rule["candidate_id"] = upsert_strategy_rule_candidate(
|
||
source="reverse_analysis",
|
||
rule_type=rule.get("type", "bonus"),
|
||
signal_name="multi_sector_hot",
|
||
rule_description=rule.get("description", ""),
|
||
support_count=len(hot_sectors),
|
||
success_count=len(hot_sectors),
|
||
fail_count=0,
|
||
confidence_score=60,
|
||
sample_size=len(sector_alignments),
|
||
status="candidate",
|
||
notes="板块联动候选规律，用于提前识别市场情绪扩散，不能作为涨后追买依据，需等待推荐样本验证后再发布",
|
||
source_ref="reverse:multi_sector_hot",
|
||
)
|
||
new_rules.append(rule)
|
||
|
||
return new_rules
|
||
|
||
|
||
# ==================== 主流程 ====================
|
||
|
||
def run_reverse_analysis():
|
||
"""
|
||
执行完整逆向分析流程：
|
||
1. 拉涨幅榜Top N，只把它作为“哪些币后来涨了”的标签
|
||
2. 过滤掉已推荐的币
|
||
3. 对每个暴涨币回溯起爆前K线，截掉涨后窗口再做PA分析
|
||
4. 统计共性特征，发现规律
|
||
5. 写入DB，返回结构化结果
|
||
"""
|
||
config = config_loader.get_reverse_params()
|
||
top_n = config.get("top_n_gainers", 30)
|
||
lookback_hours = config.get("lookback_hours", 72)
|
||
min_gain_pct = config.get("min_gain_pct", 10.0)
|
||
significance_pct = config.get("significance_threshold_pct", 60.0)
|
||
control_sample_size = config.get("control_sample_size", top_n)
|
||
min_lift = config.get("min_lift", 1.5)
|
||
|
||
# 1. 拉涨幅榜
|
||
tickers = fetch_24h_tickers()
|
||
if not tickers:
|
||
print("[reverse_analysis] 无法获取24h行情数据")
|
||
return {"error": "无法获取24h行情数据", "top_gainers": [], "pattern_summary": [], "new_rules": []}
|
||
|
||
# 排序涨幅榜
|
||
gainers = []
|
||
eligible_universe = []
|
||
for t in tickers:
|
||
symbol_str = t["symbol"]
|
||
if not _is_altcoin_usdt_symbol(symbol_str):
|
||
continue
|
||
base = symbol_str.replace("USDT", "")
|
||
formatted = f"{base}/USDT"
|
||
change_pct = float(t["priceChangePercent"])
|
||
volume_24h = float(t["quoteVolume"])
|
||
last_price = float(t["lastPrice"])
|
||
item = {
|
||
"symbol": formatted,
|
||
"gain_pct": round(change_pct, 2),
|
||
"price": last_price,
|
||
"volume_24h": volume_24h,
|
||
"sector": get_sector_for_coin(formatted),
|
||
}
|
||
eligible_universe.append(item)
|
||
|
||
if change_pct >= min_gain_pct:
|
||
gainers.append(item)
|
||
|
||
# 按涨幅排序，取Top N
|
||
gainers.sort(key=lambda x: x["gain_pct"], reverse=True)
|
||
gainers = gainers[:top_n]
|
||
|
||
# 2. 过滤掉已推荐的币
|
||
recommended = get_recommended_symbols(hours=lookback_hours)
|
||
unrecommended_gainers = [g for g in gainers if g["symbol"] not in recommended]
|
||
|
||
print(f"[reverse_analysis] 涨幅榜共{len(gainers)}只>{min_gain_pct}%, 其中{len(unrecommended_gainers)}只未被推荐")
|
||
|
||
gainer_symbols = {g["symbol"] for g in gainers}
|
||
control_candidates = [
|
||
item for item in eligible_universe
|
||
if item["symbol"] not in gainer_symbols and item["gain_pct"] < min_gain_pct
|
||
]
|
||
# 对照组取成交额较高但未大涨的山寨，保证“没涨”不是因为死币没流动性。
|
||
control_candidates.sort(key=lambda x: (x["volume_24h"], -abs(x["gain_pct"])), reverse=True)
|
||
control_group = control_candidates[:control_sample_size]
|
||
|
||
# 3. 对每个未被推荐的暴涨币做PA分析
|
||
all_features = []
|
||
sector_alignments = []
|
||
missed_details = []
|
||
|
||
for gainer in unrecommended_gainers:
|
||
symbol = gainer["symbol"]
|
||
pre_klines, explosion_klines = fetch_klines_before(symbol, lookback_hours, "1h")
|
||
|
||
features = extract_pre_explosion_features(symbol, pre_klines, explosion_klines, config)
|
||
all_features.append(features)
|
||
|
||
# 板块联动检测
|
||
sector_alignment = check_sector_alignment(symbol, gainers, config)
|
||
sector_alignments.append(sector_alignment)
|
||
gainer["sector_info"] = sector_alignment
|
||
|
||
# 构建DB记录的特征列表（用于features_detected字段）
|
||
detected_features = []
|
||
for key, value in features.items():
|
||
if key.startswith("has_") and value:
|
||
detected_features.append(key.replace("has_", ""))
|
||
elif key in ("ignition_count", "q_max", "static_ratio", "vol_surge_ratio", "static_k_count", "continuous_k_count"):
|
||
detected_features.append(f"{key}={value}")
|
||
|
||
# 计算起爆前价格（涨幅反推）
|
||
price_before = gainer["price"] / (1 + gainer["gain_pct"] / 100)
|
||
|
||
# 判断为什么没推荐（简化版）
|
||
conn = get_conn()
|
||
screened = conn.execute("""
|
||
SELECT symbol, state, score, signals FROM screening_log
|
||
WHERE symbol=%s AND layer='细筛'
|
||
ORDER BY scan_time DESC LIMIT 1
|
||
""", (symbol,)).fetchone()
|
||
conn.close()
|
||
|
||
if screened:
|
||
reason = f"细筛淘汰(state={screened['state']}, score={screened['score']})"
|
||
else:
|
||
reason = "粗筛未通过(涨幅或量价不达标)"
|
||
|
||
# 写入DB
|
||
record_missed_explosion(
|
||
symbol=symbol,
|
||
price_at_detect=gainer["price"],
|
||
price_before=price_before,
|
||
gain_pct=gainer["gain_pct"],
|
||
reason_missed=reason,
|
||
features_detected=detected_features,
|
||
lesson=f"起爆前PA特征: {', '.join(detected_features[:5])}; 板块: {sector_alignment.get('sector', '未知')}",
|
||
)
|
||
|
||
missed_details.append({
|
||
"symbol": symbol,
|
||
"gain_pct": gainer["gain_pct"],
|
||
"sector": sector_alignment.get("sector", "未知"),
|
||
"is_sector_hot": sector_alignment.get("is_sector_hot", False),
|
||
"reason_missed": reason,
|
||
"features_detected": detected_features,
|
||
"q_max": features.get("q_max", 0),
|
||
"ignition_count": features.get("ignition_count", 0),
|
||
"static_ratio": features.get("static_ratio", 0),
|
||
"vol_surge_ratio": features.get("vol_surge_ratio", 0),
|
||
})
|
||
|
||
control_features = []
|
||
control_details = []
|
||
for item in control_group:
|
||
symbol = item["symbol"]
|
||
pre_klines, explosion_klines = fetch_klines_before(symbol, lookback_hours, "1h")
|
||
features = extract_pre_explosion_features(symbol, pre_klines, explosion_klines, config)
|
||
control_features.append(features)
|
||
control_details.append({
|
||
"symbol": symbol,
|
||
"gain_pct": item["gain_pct"],
|
||
"volume_24h": item["volume_24h"],
|
||
"features": [key for key, value in features.items() if key.startswith("has_") and value],
|
||
})
|
||
|
||
# 4. 统计共性特征
|
||
total_analyzed = len(all_features)
|
||
pattern_summary = compute_pattern_summary(all_features, total_analyzed, control_features=control_features)
|
||
|
||
# 5. 发现新规律
|
||
new_rules = discover_new_rules(
|
||
pattern_summary,
|
||
all_features,
|
||
sector_alignments,
|
||
significance_pct,
|
||
min_lift=min_lift,
|
||
)
|
||
|
||
# 更新meta
|
||
config_loader.update_meta("last_reverse_analysis", datetime.now().isoformat())
|
||
|
||
# 6. 返回结构化结果
|
||
results = {
|
||
"timestamp": datetime.now().isoformat(),
|
||
"analysis_scope": "pre_explosion_only",
|
||
"feature_window": {
|
||
"lookback_hours": lookback_hours,
|
||
"excluded_recent_hours": 24,
|
||
"label_usage": "top_gainers_are_outcome_labels_only",
|
||
"note": "涨幅榜只用于标记哪些币后来涨了；因子只从起爆前/启动点窗口提取，不能使用涨后结果做追买依据。",
|
||
},
|
||
"total_gainers": len(gainers),
|
||
"total_unrecommended": len(unrecommended_gainers),
|
||
"total_analyzed": total_analyzed,
|
||
"control_analyzed": len(control_features),
|
||
"top_gainers": gainers[:10], # feishu推送只取TOP10
|
||
"missed_details": missed_details,
|
||
"control_details": control_details[:10],
|
||
"pattern_summary": pattern_summary,
|
||
"new_rules": new_rules,
|
||
"sector_alignments": sector_alignments,
|
||
}
|
||
|
||
print(f"[reverse_analysis] 完成: 分析{total_analyzed}只, 发现{len(new_rules)}条新规律")
|
||
return results
|
||
|
||
|
||
if __name__ == "__main__":
|
||
results = run_reverse_analysis()
|
||
print(json.dumps(results, ensure_ascii=False, indent=2))
|