release: v1.30.2 — 无未来函数指标扩容16个（注册指标50个）+ 内置策略补齐54个：WebUI回测全覆盖，新增前缀一致性无未来函数回归测试

2026-09-12 18:04:20 +08:00 · 2026-09-03 03:07:57 +08:00
parent 1433f4eed0
commit 62e80d892e
11 changed files with 2496 additions and 7 deletions
@@ -3,7 +3,7 @@
 核心保证：**同一 df + 同参数下，向量化路径与逐 bar 路径的输出逐位一致**
 （performance / trades / equity_curve / positions 全比对）。

- 对拍覆盖：全部 19 个内置策略（默认参数）+ ma_cross/macd/boll/rsi 的非默认
+- 对拍覆盖：全部内置策略（默认参数，当前 54 个）+ ma_cross/macd/boll/rsi 的非默认
  参数组合 + warmup / 极低资金（买不足 1 手的退化路径）/ 非默认费率与成交价模式；
 - 约束检测：``_vectorize_eligibility`` 的显式约束（无掩码 / 缠论注入）与
  ``signal_path`` 的 auto/vector/loop 语义。
@@ -115,7 +115,7 @@ def _run_both(
    return loop, vec


-# ── 对拍：19 个内置策略 × 默认参数 ───────────────────────────────────────────
+# ── 对拍：全部内置策略 × 默认参数 ───────────────────────────────────────────


 #: 已知「默认参数下不会交易」的策略：MyTT 的 WR 是 0~100 刻度（100=超卖），
@@ -1,4 +1,4 @@
-"""MyTT.py 新增指标函数（SAR/VWAP/AROON/FK）的数值正确性与边界测试。
+"""MyTT.py 新增指标函数（SAR/VWAP/AROON/FK + V4.3 十六个无未来函数指标）的数值正确性与边界测试。

 这些测试针对 MyTT.py 里函数本身，不经过 indicator.py 注册层。
 注册层的端到端覆盖在 test_indicator.py::TestComputeIndicators::test_all_registered_indicators_run。
@@ -214,3 +214,346 @@ class TestFK:
        close = np.array([100 - i for i in range(100)], dtype=float)
        fk = MyTT.FK(close)
        assert bool(fk[-1]) is True
+
+
+# ═══════════════════════════════════════════════════════════════════════════
+# V4.3 新增：16 个无未来函数指标
+# ═══════════════════════════════════════════════════════════════════════════
+
+#: 新指标的构造器：统一接收 (open, high, low, close, vol) 五元组，返回输出元组。
+#: 用于「无未来函数」前缀一致性回归（见 TestNoLookahead）。
+_V43_INDICATORS = {
+    "HMA": lambda o, h, lo, c, v: (MyTT.HMA(c, 16),),
+    "KAMA": lambda o, h, lo, c, v: (MyTT.KAMA(c),),
+    "SUPERTREND": lambda o, h, lo, c, v: MyTT.SUPERTREND(c, h, lo),
+    "CHANDELIER": lambda o, h, lo, c, v: MyTT.CHANDELIER(c, h, lo),
+    "ICHIMOKU": lambda o, h, lo, c, v: MyTT.ICHIMOKU(h, lo, c),
+    "UOS": lambda o, h, lo, c, v: MyTT.UOS(c, h, lo),
+    "CMO": lambda o, h, lo, c, v: (MyTT.CMO(c),),
+    "TSI": lambda o, h, lo, c, v: MyTT.TSI(c),
+    "FISHER": lambda o, h, lo, c, v: MyTT.FISHER(h, lo),
+    "SQUEEZE": lambda o, h, lo, c, v: MyTT.SQUEEZE(c, h, lo),
+    "CHOP": lambda o, h, lo, c, v: (MyTT.CHOP(h, lo, c),),
+    "AD": lambda o, h, lo, c, v: (MyTT.AD(c, h, lo, v),),
+    "CMF": lambda o, h, lo, c, v: (MyTT.CMF(c, h, lo, v),),
+    "EFI": lambda o, h, lo, c, v: (MyTT.EFI(c, v),),
+    "BBP": lambda o, h, lo, c, v: (MyTT.BBP(c),),
+    "BBW": lambda o, h, lo, c, v: (MyTT.BBW(c),),
+}
+
+
+class TestNoLookahead:
+    """无未来函数回归：指标在全序列上前缀段输出 == 仅用前缀数据计算的输出。
+
+    未来函数（如 ZIG）的致命特征是：后到的数据会改写历史输出。本测试
+    把 200 根 K 线截断到前 120 根，两组输出在重叠段必须逐位一致——
+    任何引用了 t+1 及之后数据的实现都会当场爆红。
+    """
+
+    PREFIX = 120
+
+    @pytest.mark.parametrize("name", sorted(_V43_INDICATORS))
+    def test_prefix_stability(self, name):
+        ohlcv = _ohlcv(200)
+        full = _V43_INDICATORS[name](*ohlcv)
+        part = _V43_INDICATORS[name](*[x[: self.PREFIX] for x in ohlcv])
+        for j, (f, p) in enumerate(zip(full, part)):
+            # ICHIMOKU 迟行带引用未来数据画图（文档已声明仅作图示），
+            # 它是唯一允许前缀不一致的输出，单独跳过（见 TestICHIMOKU）。
+            if name == "ICHIMOKU" and j == 4:
+                continue
+            assert np.allclose(f[: self.PREFIX], p, equal_nan=True), (
+                f"{name} 输出#{j} 前缀不一致：疑似引用了未来数据"
+            )
+
+
+class TestHMA:
+    def test_warmup_and_length(self):
+        close = _ohlcv()[3]
+        hma = MyTT.HMA(close, 16)
+        assert len(hma) == len(close)
+        assert np.isnan(hma[:14]).all()  # 最内层 WMA(16) 窗口预热
+        assert not np.isnan(hma[19])
+
+    def test_rising_market_follows_price(self):
+        close = np.arange(100, dtype=float) * 0.5 + 10
+        hma = MyTT.HMA(close, 16)
+        # 单边上涨中低滞后均线应贴在价格下方且不发散
+        assert (hma[20:] <= close[20:] + 1e-6).all()
+        assert hma[-1] > close[-2]  # 跟随上涨
+
+
+class TestKAMA:
+    def test_warmup_starts_at_n(self):
+        close = _ohlcv()[3]
+        kama = MyTT.KAMA(close, N=10)
+        assert np.isnan(kama[:10]).all()
+        assert not np.isnan(kama[10])
+
+    def test_strong_trend_hugs_price(self):
+        # 单边强趋势：效率比≈1，KAMA 平滑系数取快速极值，稳态滞后 ≈ (1-sc)/sc ≈ 1.25 根
+        close = np.cumsum(np.ones(100))  # 每根 +1 的完美趋势
+        kama = MyTT.KAMA(close, N=10)
+        assert np.abs(kama[-1] - close[-1]) < 2.0
+
+    def test_flat_market_flat_kama(self):
+        kama = MyTT.KAMA(np.full(60, 10.0), N=10)
+        assert np.allclose(kama[10:], 10.0)
+
+
+class TestSUPERTREND:
+    def test_direction_values(self):
+        _, high, low, close, _ = _ohlcv()
+        st, direction = MyTT.SUPERTREND(close, high, low)
+        assert set(np.unique(direction).tolist()) <= {1, -1}
+        assert np.isfinite(st).all()
+
+    def test_rising_market_st_below_price(self):
+        # 持续上涨：趋势为多，ST（下轨）应持续低于最低价
+        high = np.arange(80, dtype=float) + 1
+        low = np.arange(80, dtype=float)
+        close = high.copy()
+        st, direction = MyTT.SUPERTREND(close, high, low, N=10, M=3.0)
+        assert direction[-1] == 1
+        assert (st[10:] <= low[10:] + 1e-6).all()
+
+    def test_reversal_flips_direction(self):
+        # V 型反转：方向必须从 1 翻到 -1
+        half = np.arange(40, dtype=float)
+        close = np.concatenate([half + 1, 40 - half])
+        high = close + 0.5
+        low = close - 0.5
+        _, direction = MyTT.SUPERTREND(close, high, low)
+        assert direction[0] != direction[-1]
+
+    def test_empty_input(self):
+        st, direction = MyTT.SUPERTREND(np.array([]), np.array([]), np.array([]))
+        assert len(st) == 0 and len(direction) == 0
+
+
+class TestCHANDELIER:
+    def test_stops_bracket_price(self):
+        _, high, low, close, _ = _ohlcv()
+        long_stop, short_stop = MyTT.CHANDELIER(close, high, low, N=22, M=22, K=3.0)
+        # 吊灯止损锚定通道极值：多头止损在 N 日最高价下方、空头止损在 N 日最低价上方
+        # （下跌段中 long_stop 可以高于当根 high——这正是吊灯线滞后等待离场的行为）
+        valid = slice(22, None)  # TR[0] 为 NaN（REF 前收盘缺失）→ ATR 自 22 起有效
+        assert (long_stop[valid] < MyTT.HHV(high, 22)[valid]).all()
+        assert (short_stop[valid] > MyTT.LLV(low, 22)[valid]).all()
+
+    def test_matches_manual_formula(self):
+        _, high, low, close, _ = _ohlcv()
+        long_stop, _ = MyTT.CHANDELIER(close, high, low, N=22, M=22, K=2.0)
+        expected = MyTT.HHV(high, 22) - MyTT.ATR(close, high, low, 22) * 2.0
+        assert np.allclose(long_stop, expected, equal_nan=True)
+
+
+class TestICHIMOKU:
+    def test_five_outputs_lengths(self):
+        _, high, low, close, _ = _ohlcv()
+        outs = MyTT.ICHIMOKU(high, low, close)
+        assert len(outs) == 5
+        for arr in outs:
+            assert len(arr) == len(close)
+
+    def test_tenkan_formula(self):
+        _, high, low, close, _ = _ohlcv()
+        tenkan, _, _, _, _ = MyTT.ICHIMOKU(high, low, close, P1=9, P2=26, P3=52)
+        expected = (MyTT.HHV(high, 9) + MyTT.LLV(low, 9)) / 2
+        assert np.allclose(tenkan, expected, equal_nan=True)
+
+    def test_span_is_shifted_past(self):
+        # 先行带 = 26 期前的 (转换线+基准线)/2：i 处的值来自 i-26（过去）
+        _, high, low, close, _ = _ohlcv()
+        _, _, span_a, _, _ = MyTT.ICHIMOKU(high, low, close, P1=9, P2=26, P3=52, SHIFT=26)
+        tenkan = (MyTT.HHV(high, 9) + MyTT.LLV(low, 9)) / 2
+        kijun = (MyTT.HHV(high, 26) + MyTT.LLV(low, 26)) / 2
+        raw = (tenkan + kijun) / 2
+        assert np.allclose(span_a[26:], raw[:-26], equal_nan=True)
+
+    def test_chikou_tail_nan(self):
+        # 迟行带 = 当前收盘画回 26 期前：末尾 26 个槽位无对应未来数据 → NaN
+        _, high, low, close, _ = _ohlcv()
+        *_, chikou = MyTT.ICHIMOKU(high, low, close)
+        assert np.isnan(chikou[-26:]).all()
+        assert not np.isnan(chikou[:-26]).any()
+
+
+class TestUOS:
+    def test_range_zero_to_hundred(self):
+        _, high, low, close, _ = _ohlcv()
+        uos, uos_ma = MyTT.UOS(close, high, low)
+        valid = uos[28:]  # bp[0] 为 NaN（REF 前收盘缺失）→ P3=28 窗口自 28 起有效
+        assert (valid >= 0).all() and (valid <= 100).all()
+
+    def test_new_low_oversold(self):
+        # 持续创新低 → UOS 应处于超卖区（<50）
+        low = np.linspace(50, 1, 60)
+        close = low.copy()
+        high = low + 0.5
+        uos, _ = MyTT.UOS(close, high, low)
+        assert uos[-1] < 50
+
+    def test_flat_market_neutral(self):
+        flat = np.full(60, 10.0)
+        uos, _ = MyTT.UOS(flat, flat, flat)
+        assert np.allclose(uos[28:], 50.0, equal_nan=True)  # 除零保护取中性
+
+
+class TestCMO:
+    def test_symmetric_range(self):
+        close = _ohlcv()[3]
+        cmo = MyTT.CMO(close, N=14)
+        valid = cmo[14:]
+        assert (valid >= -100).all() and (valid <= 100).all()
+
+    def test_rising_positive_falling_negative(self):
+        rise = np.cumsum(np.ones(100))
+        assert MyTT.CMO(rise, N=14)[-1] > 0  # 纯上涨 → +100 极值
+        fall = 100 - np.cumsum(np.ones(100))
+        assert MyTT.CMO(fall, N=14)[-1] < 0  # 纯下跌 → -100 极值
+
+
+class TestTSI:
+    def test_signal_line_follows(self):
+        close = _ohlcv()[3]
+        tsi, signal = MyTT.TSI(close)
+        assert len(tsi) == len(signal) == len(close)
+        valid = ~np.isnan(tsi) & ~np.isnan(signal)
+        assert valid.any()
+
+    def test_strong_rise_positive(self):
+        close = np.cumsum(np.ones(100))
+        tsi, _ = MyTT.TSI(close)
+        assert tsi[-1] > 0
+
+
+class TestFISHER:
+    def test_trigger_is_prev_value(self):
+        _, high, low, _, _ = _ohlcv()
+        fisher, trigger = MyTT.FISHER(high, low, N=9)
+        assert np.allclose(trigger[1:], fisher[:-1], equal_nan=True)
+
+    def test_strong_rise_positive_sharply(self):
+        high = np.linspace(1, 50, 100)
+        low = high - 0.5
+        fisher, _ = MyTT.FISHER(high, low, N=9)
+        assert fisher[-1] > 1.0  # 顶部区域输出尖峰
+        assert np.isfinite(fisher[8:]).all()  # 钳制保证无 inf
+
+    def test_bounded_input_clamp(self):
+        # 归一化值被钳制在 ±0.999 → 输出有限
+        _, high, low, _, _ = _ohlcv()
+        fisher, _ = MyTT.FISHER(high, low, N=3)
+        assert np.isfinite(fisher[2:]).all()
+
+
+class TestSQUEEZE:
+    def test_bool_flag_and_mom_length(self):
+        _, high, low, close, _ = _ohlcv()
+        sqz, mom = MyTT.SQUEEZE(close, high, low)
+        assert sqz.dtype == bool
+        assert len(sqz) == len(mom) == len(close)
+
+    def test_flat_market_mom_zero(self):
+        flat = np.full(60, 10.0)
+        _, mom = MyTT.SQUEEZE(flat, flat, flat)
+        # 双层 N=20 窗口（带宽层 + 回归层）→ 有效值自 2N-1 起
+        assert np.allclose(mom[39:], 0.0, atol=1e-9)
+
+
+class TestCHOP:
+    def test_range_and_warmup(self):
+        _, high, low, close, _ = _ohlcv()
+        chop = MyTT.CHOP(high, low, close, N=14)
+        assert np.isnan(chop[:14]).all()  # TR[0] 为 NaN → 14 窗口自 14 起有效
+        valid = chop[14:]
+        assert (valid >= 0).all() and (valid <= 100).all()
+
+    def test_strong_trend_low_chop(self):
+        # 完美趋势：ΣTR ≈ 区间 → chop 趋近低值
+        high = np.arange(1, 81, dtype=float)
+        low = high - 1
+        close = high.copy()
+        chop = MyTT.CHOP(high, low, close, N=14)
+        assert chop[-1] < 30
+
+    def test_oscillation_high_chop(self):
+        # 剧烈往返震荡（窗口跨 2 个以上完整周期）：路径远大于区间 → chop 高
+        t = np.arange(200)
+        close = 10 + 5 * np.sin(2 * np.pi * t / 6)
+        high = close + 0.5
+        low = close - 0.5
+        chop = MyTT.CHOP(high, low, close, N=14)
+        assert chop[14:].max() > 55
+
+
+class TestADandCMF:
+    def test_ad_manual_clv(self):
+        # 单根：CLV=((C-L)-(H-C))/(H-L)，AD=CLV×VOL 累计；C=11.5 → CLV=(1.5-0.5)/2=0.5
+        close = np.array([11.5, 11.5])
+        high = np.array([12.0, 12.0])
+        low = np.array([10.0, 10.0])
+        vol = np.array([100.0, 100.0])
+        ad = MyTT.AD(close, high, low, vol)
+        assert ad[0] == pytest.approx(0.5 * 100)
+        assert ad[1] == pytest.approx(100.0)
+
+    def test_cmf_range_and_warmup(self):
+        _, high, low, close, vol = _ohlcv()
+        cmf = MyTT.CMF(close, high, low, vol, N=20)
+        assert np.isnan(cmf[:19]).all()
+        valid = cmf[19:]
+        assert (valid >= -1).all() and (valid <= 1).all()
+
+    def test_cmf_sign_matches_close_position(self):
+        # 收盘持续靠近最高价（吸筹）→ CMF 为正
+        n = 60
+        close = np.linspace(10, 20, n)
+        high = close + 0.1
+        low = close - 1.0  # 收盘贴近最高
+        vol = np.full(n, 1000.0)
+        cmf = MyTT.CMF(close, high, low, vol, N=20)
+        assert cmf[-1] > 0
+
+
+class TestEFI:
+    def test_rising_with_volume_positive(self):
+        n = 100
+        close = np.cumsum(np.ones(n))
+        vol = np.full(n, 1000.0)
+        efi = MyTT.EFI(close, vol, N=13)
+        assert efi[-1] > 0
+
+    def test_length_and_warmup(self):
+        close = _ohlcv()[3]
+        vol = _ohlcv()[4]
+        efi = MyTT.EFI(close, vol, N=13)
+        assert len(efi) == len(close)
+        assert np.isnan(efi[0])  # DIFF 首位 NaN
+
+
+class TestBBPandBBW:
+    def test_bbp_position_semantics(self):
+        # N=3 手工窗口 [10, 14, x]：mid=12、std=sqrt(8/3)；close=mid → %B 恰为 50
+        c_mid = np.array([10.0, 14.0, 12.0])
+        assert MyTT.BBP(c_mid, N=3, P=2)[-1] == pytest.approx(50.0, abs=1e-6)
+        # 位置单调：同一窗口形态下，收盘越高 %B 越大
+        lo = MyTT.BBP(np.array([10.0, 14.0, 11.0]), N=3, P=2)[-1]
+        hi = MyTT.BBP(np.array([10.0, 14.0, 13.0]), N=3, P=2)[-1]
+        assert lo < 50.0 < hi
+        # 公式口径：直接用 numpy 独立重算 (C-(mid-2sd))/(4sd)*100（RD 三位小数舍入）
+        window = np.array([10.0, 14.0, 13.0])
+        mid, sd = window.mean(), window.std()
+        expected = (13.0 - (mid - 2 * sd)) / (4 * sd) * 100
+        assert MyTT.BBP(window, N=3, P=2)[-1] == pytest.approx(expected, abs=1e-3)
+
+    def test_bbw_zero_when_flat(self):
+        bbw = MyTT.BBW(np.full(60, 10.0), N=20, P=2)
+        assert np.allclose(bbw[19:], 0.0)
+
+    def test_bbw_grows_with_volatility(self):
+        rng = np.random.default_rng(7)
+        quiet = 100 + rng.standard_normal(60) * 0.1
+        wild = 100 + rng.standard_normal(60) * 5.0
+        assert MyTT.BBW(wild)[-1] > MyTT.BBW(quiet)[-1]