Use Torch as the only forecast model

crypto_spot_bot/config.py

@@ -98,9 +98,7 @@ class Settings:
     kelly_max_fraction: float
     time_series_forecast_enabled: bool
     time_series_min_candles: int
-    time_series_validation_window: int
     time_series_forecast_horizon: int
-    time_series_ewma_lambda: float
     time_series_min_edge_percent: float
     time_series_max_adjustment: float
     time_series_lstm_enabled: bool
@@ -222,9 +220,7 @@ def load_settings(env_file: str | Path | None = None) -> Settings:
         kelly_max_fraction=_float_env("KELLY_MAX_FRACTION", 0.20),
         time_series_forecast_enabled=_bool_env("TIME_SERIES_FORECAST_ENABLED", True),
         time_series_min_candles=_int_env("TIME_SERIES_MIN_CANDLES", 120),
-        time_series_validation_window=_int_env("TIME_SERIES_VALIDATION_WINDOW", 30),
         time_series_forecast_horizon=_int_env("TIME_SERIES_FORECAST_HORIZON", 3),
-        time_series_ewma_lambda=_float_env("TIME_SERIES_EWMA_LAMBDA", 0.94),
         time_series_min_edge_percent=_float_env("TIME_SERIES_MIN_EDGE_PERCENT", 0.04),
         time_series_max_adjustment=_float_env("TIME_SERIES_MAX_ADJUSTMENT", 0.08),
         time_series_lstm_enabled=_bool_env("TIME_SERIES_LSTM_ENABLED", True),

crypto_spot_bot/dashboard.py

@@ -215,9 +215,7 @@ def _safe_config(settings: Settings) -> dict[str, Any]:
         "kelly_max_fraction": settings.kelly_max_fraction,
         "time_series_forecast_enabled": settings.time_series_forecast_enabled,
         "time_series_min_candles": settings.time_series_min_candles,
-        "time_series_validation_window": settings.time_series_validation_window,
         "time_series_forecast_horizon": settings.time_series_forecast_horizon,
-        "time_series_ewma_lambda": settings.time_series_ewma_lambda,
         "time_series_min_edge_percent": settings.time_series_min_edge_percent,
         "time_series_max_adjustment": settings.time_series_max_adjustment,
         "time_series_lstm_enabled": settings.time_series_lstm_enabled,
@@ -257,16 +255,19 @@ def _time_series_model_artifact(settings: Settings) -> dict[str, Any]:
             "symbol_count": 0,
             "models": [],
         }
+    artifact_type = str(data.get("type", "")).strip()
     symbols = data.get("symbols")
     rows = list(symbols.values()) if isinstance(symbols, dict) else []
     models = sorted(
         {
-            _forecast_model_label(str(row.get("model", row.get("architecture", "lstm"))))
+            _forecast_model_label(
+                str(row.get("model", row.get("architecture", "lstm"))),
+                torch_artifact=artifact_type == "pytorch_recurrent_forecaster",
+            )
             for row in rows
             if isinstance(row, dict)
         }
     )
-    artifact_type = str(data.get("type", "")).strip()
     if artifact_type != "pytorch_recurrent_forecaster":
         return {
             "available": False,
@@ -286,16 +287,16 @@ def _time_series_model_artifact(settings: Settings) -> dict[str, Any]:
     }
 
 
-def _forecast_model_label(model: str) -> str:
+def _forecast_model_label(model: str, *, torch_artifact: bool = False) -> str:
     normalized = model.strip().lower()
-    if normalized == "torch_lstm":
+    if normalized in {"torch_lstm", "lstm"} and torch_artifact:
         return "PyTorch LSTM"
-    if normalized == "torch_gru":
+    if normalized in {"torch_gru", "gru"} and torch_artifact:
         return "PyTorch GRU"
     if normalized == "lstm":
-        return "устаревший LSTM"
+        return "устаревший артефакт"
     if normalized == "gru":
-        return "GRU"
+        return "устаревший артефакт"
     return model
 
 
@@ -746,12 +747,6 @@ HTML = r"""
       const names = {
         torch_lstm: 'PyTorch LSTM',
         torch_gru: 'PyTorch GRU',
-        lstm: 'Устаревший LSTM',
-        naive: 'Baseline',
-        drift: 'Drift',
-        ewma: 'EWMA',
-        ar1: 'AR(1)',
-        ar3: 'AR(3)',
         none: '-'
       };
       return names[key] || String(model || '-');
@@ -937,7 +932,6 @@ HTML = r"""
         ['Kelly размер', `${yesNo(config.kelly_sizing_enabled)} · ${num(config.kelly_fraction, 2)}x · max ${num((config.kelly_max_fraction || 0) * 100, 1)}%`],
         ['Прогноз временных рядов', yesNo(config.time_series_forecast_enabled)],
         ['Модельный горизонт', `${config.time_series_forecast_horizon} свечи`],
-        ['Walk-forward окно', `${config.time_series_validation_window} свечей`],
         ['Мин. edge прогноза', `${num(config.time_series_min_edge_percent, 3)}%`],
         ['Нейропрогноз', modelArtifactSummary(config)],
         ['Файл модели', config.time_series_lstm_model_path || '-'],

crypto_spot_bot/time_series.py

@@ -44,62 +44,52 @@ class TimeSeriesForecaster:
         closes = [float(candle.close) for candle in candles if candle.close > 0]
         min_candles = max(30, self.settings.time_series_min_candles)
         if len(closes) < min_candles:
-            return _empty_forecast(True, "недостаточно свечей для прогноза")
+            return _empty_forecast(True, "недостаточно свечей для PyTorch прогноза")
         returns = _log_returns(closes)
         if len(returns) < 20:
-            return _empty_forecast(True, "недостаточно доходностей для прогноза")
+            return _empty_forecast(True, "недостаточно доходностей для PyTorch прогноза")
+
+        artifact = self._load_lstm_artifact()
+        model = _torch_recurrent_model_name(symbol, artifact)
+        if not model or not _can_use_torch_recurrent(returns, symbol, artifact):
+            return _empty_forecast(True, "нет валидной PyTorch LSTM/GRU модели для пары")
+        entry = _torch_recurrent_entry(symbol, artifact)
+        prediction = _torch_recurrent_predict(returns, symbol, artifact)
+        if entry is None or prediction is None:
+            return _empty_forecast(True, "PyTorch LSTM/GRU модель не смогла построить прогноз")
 
-        validation_window = min(
-            max(8, self.settings.time_series_validation_window),
-            max(8, len(returns) // 3),
-        )
-        lstm_artifact = self._load_lstm_artifact()
-        candidates = _validate_candidates(returns, validation_window, self.settings, symbol, lstm_artifact)
-        best = min(candidates, key=lambda item: item["mae"])
-        baseline = next(item for item in candidates if item["model"] == "naive")
-        latest_prediction = _predict_next_return(best["model"], returns, self.settings, symbol, lstm_artifact)
         horizon = max(1, self.settings.time_series_forecast_horizon)
-        expected_return = latest_prediction * horizon
+        expected_return = prediction * horizon
         expected_price = closes[-1] * math.exp(expected_return)
-
-        ewma_vol = _ewma_volatility(returns, self.settings.time_series_ewma_lambda)
-        garch_vol = _fixed_garch_volatility(returns)
-        vol_one_step = max(ewma_vol, garch_vol)
-        volatility_percent = vol_one_step * math.sqrt(horizon) * 100
+        model_mae = _torch_validation_mae(entry, returns)
+        baseline_mae = max(_float_entry(entry, "baseline_mae_percent", model_mae * 100) / 100, model_mae)
+        uncertainty_one_step = max(model_mae, _return_scale(returns) * 0.25, 1e-9)
+        uncertainty = uncertainty_one_step * math.sqrt(horizon)
+        volatility_percent = uncertainty * 100
         expected_return_percent = (math.exp(expected_return) - 1) * 100
-        probability_up = _normal_cdf(expected_return / max(vol_one_step * math.sqrt(horizon), 1e-9))
-        baseline_mae = float(baseline["mae"])
-        model_mae = float(best["mae"])
-        skill = (baseline_mae - model_mae) / baseline_mae if baseline_mae > 0 else 0.0
-        skill = _clamp(skill, -1.0, 1.0)
+        probability_up = _normal_cdf(expected_return / max(uncertainty, 1e-9))
+        skill = _clamp(_float_entry(entry, "skill", 0.0), -1.0, 1.0)
         min_edge = max(0.0, self.settings.time_series_min_edge_percent)
-        usable_skill = skill > 0.02 and best["model"] != "naive"
         confidence_adjustment = _confidence_adjustment(
             expected_return_percent=expected_return_percent,
             probability_up=probability_up,
             skill=skill,
             min_edge=min_edge,
             max_adjustment=self.settings.time_series_max_adjustment,
-            usable_skill=usable_skill,
-        )
-        block_entry = bool(
-            usable_skill
-            and expected_return_percent <= -min_edge
-            and probability_up <= 0.45
         )
+        block_entry = bool(expected_return_percent <= -min_edge and probability_up <= 0.45)
         reason = _reason(
-            model=best["model"],
+            model=model,
             expected_return_percent=expected_return_percent,
             probability_up=probability_up,
             skill=skill,
             block_entry=block_entry,
-            usable_skill=usable_skill,
         )
         return TimeSeriesForecast(
             enabled=True,
             usable=True,
-            model=best["model"],
-            volatility_model="max(EWMA,GARCH-like)",
+            model=model,
+            volatility_model="torch validation MAE",
             expected_return_percent=round(expected_return_percent, 4),
             expected_price=round(expected_price, 8),
             volatility_percent=round(volatility_percent, 4),
@@ -111,10 +101,7 @@ class TimeSeriesForecaster:
             skill=round(skill, 4),
             horizon=horizon,
             reason=reason,
-            candidates=[
-                {"model": item["model"], "mae_percent": round(float(item["mae"]) * 100, 4)}
-                for item in sorted(candidates, key=lambda item: item["mae"])
-            ],
+            candidates=[{"model": model, "mae_percent": round(model_mae * 100, 4)}],
         )
 
     def _load_lstm_artifact(self) -> dict[str, Any]:
@@ -162,85 +149,8 @@ def _log_returns(closes: list[float]) -> list[float]:
     return [math.log(closes[index] / closes[index - 1]) for index in range(1, len(closes))]
 
 
-def _validate_candidates(
-    returns: list[float],
-    validation_window: int,
-    settings: Settings,
-    symbol: str | None = None,
-    lstm_artifact: dict[str, Any] | None = None,
-) -> list[dict[str, float | str]]:
-    models = ["naive", "drift", "ewma", "ar1", "ar3"]
-    torch_model = _torch_recurrent_model_name(symbol, lstm_artifact or {})
-    if torch_model and _can_use_torch_recurrent(returns, symbol, lstm_artifact or {}):
-        models.append(torch_model)
-    rows: list[dict[str, float | str]] = []
-    start = max(8, len(returns) - validation_window)
-    for model in models:
-        errors: list[float] = []
-        for index in range(start, len(returns)):
-            history = returns[:index]
-            if len(history) < 8:
-                continue
-            predicted = _predict_next_return(model, history, settings, symbol, lstm_artifact)
-            errors.append(abs(predicted - returns[index]))
-        mae = sum(errors) / len(errors) if errors else 1e9
-        rows.append({"model": model, "mae": mae})
-    return rows
-
-
-def _predict_next_return(
-    model: str,
-    returns: list[float],
-    settings: Settings | None = None,
-    symbol: str | None = None,
-    lstm_artifact: dict[str, Any] | None = None,
-) -> float:
-    if model == "naive":
-        return 0.0
-    if model == "drift":
-        window = returns[-24:] if len(returns) >= 24 else returns
-        return sum(window) / len(window) if window else 0.0
-    if model == "ewma":
-        return _ewma_mean(returns, 0.82)
-    if model == "ar1":
-        return _ar_predict(returns, 1)
-    if model == "ar3":
-        return _ar_predict(returns, 3)
-    if model in {"torch_lstm", "torch_gru"}:
-        return _torch_recurrent_predict(returns, symbol, lstm_artifact or {})
-    return 0.0
-
-
-def _ewma_mean(values: list[float], decay: float) -> float:
-    if not values:
-        return 0.0
-    estimate = values[0]
-    alpha = 1 - _clamp(decay, 0.01, 0.99)
-    for value in values[1:]:
-        estimate = alpha * value + (1 - alpha) * estimate
-    return estimate
-
-
-def _ar_predict(returns: list[float], lag_count: int) -> float:
-    if len(returns) <= lag_count + 6:
-        return _predict_next_return("drift", returns)
-    rows: list[list[float]] = []
-    targets: list[float] = []
-    for index in range(lag_count, len(returns)):
-        rows.append([1.0] + [returns[index - lag] for lag in range(1, lag_count + 1)])
-        targets.append(returns[index])
-    coeffs = _ols(rows, targets)
-    if not coeffs:
-        return _predict_next_return("drift", returns)
-    features = [1.0] + [returns[-lag] for lag in range(1, lag_count + 1)]
-    prediction = sum(coeff * feature for coeff, feature in zip(coeffs, features))
-    recent_abs = sorted(abs(value) for value in returns[-48:]) if len(returns) >= 8 else [0.01]
-    cap = max(recent_abs[int(len(recent_abs) * 0.9)], 0.0002)
-    return _clamp(prediction, -cap, cap)
-
-
-def _torch_recurrent_model_name(symbol: str | None, lstm_artifact: dict[str, Any]) -> str | None:
-    entry = _torch_recurrent_entry(symbol, lstm_artifact)
+def _torch_recurrent_model_name(symbol: str | None, artifact: dict[str, Any]) -> str | None:
+    entry = _torch_recurrent_entry(symbol, artifact)
     if not entry:
         return None
     architecture = str(entry.get("architecture", "")).strip().lower()
@@ -250,11 +160,13 @@ def _torch_recurrent_model_name(symbol: str | None, lstm_artifact: dict[str, Any
     return model if model in {"torch_lstm", "torch_gru"} else None
 
 
-def _torch_recurrent_entry(symbol: str | None, lstm_artifact: dict[str, Any]) -> dict[str, Any] | None:
-    symbols = lstm_artifact.get("symbols")
+def _torch_recurrent_entry(symbol: str | None, artifact: dict[str, Any]) -> dict[str, Any] | None:
+    if artifact.get("type") != "pytorch_recurrent_forecaster":
+        return None
+    symbols = artifact.get("symbols")
     entry = symbols.get(symbol.upper()) if symbol and isinstance(symbols, dict) else None
     if not isinstance(entry, dict):
-        default = lstm_artifact.get("default")
+        default = artifact.get("default")
         entry = default if isinstance(default, dict) else None
     if not isinstance(entry, dict):
         return None
@@ -263,8 +175,8 @@ def _torch_recurrent_entry(symbol: str | None, lstm_artifact: dict[str, Any]) ->
     return entry
 
 
-def _can_use_torch_recurrent(returns: list[float], symbol: str | None, lstm_artifact: dict[str, Any]) -> bool:
-    entry = _torch_recurrent_entry(symbol, lstm_artifact)
+def _can_use_torch_recurrent(returns: list[float], symbol: str | None, artifact: dict[str, Any]) -> bool:
+    entry = _torch_recurrent_entry(symbol, artifact)
     if not entry:
         return False
     lookback = int(_clamp(_float_entry(entry, "lookback", 0.0), 4.0, 512.0))
@@ -276,12 +188,12 @@ def _can_use_torch_recurrent(returns: list[float], symbol: str | None, lstm_arti
 def _torch_recurrent_predict(
     returns: list[float],
     symbol: str | None,
-    lstm_artifact: dict[str, Any],
-) -> float:
-    entry = _torch_recurrent_entry(symbol, lstm_artifact)
-    model_name = _torch_recurrent_model_name(symbol, lstm_artifact)
+    artifact: dict[str, Any],
+) -> float | None:
+    entry = _torch_recurrent_entry(symbol, artifact)
+    model_name = _torch_recurrent_model_name(symbol, artifact)
     if not entry or not model_name:
-        return _predict_next_return("drift", returns)
+        return None
     lookback = int(_clamp(_float_entry(entry, "lookback", 0.0), 4.0, 512.0))
     hidden_size = int(_clamp(_float_entry(entry, "hidden_size", 0.0), 1.0, 512.0))
     num_layers = int(_clamp(_float_entry(entry, "num_layers", 1.0), 1.0, 8.0))
@@ -289,7 +201,7 @@ def _torch_recurrent_predict(
     scale = max(_float_entry(entry, "scale", _return_scale(returns)), 1e-8)
     clip = _clamp(_float_entry(entry, "clip", 8.0), 1.0, 50.0)
     if len(returns) < lookback:
-        return _predict_next_return("drift", returns)
+        return None
 
     normalized = [_clamp((value - mean) / scale, -clip, clip) for value in returns[-lookback:]]
     try:
@@ -301,17 +213,17 @@ def _torch_recurrent_predict(
             num_layers=num_layers,
         )
         if hidden is None:
-            return _predict_next_return("drift", returns)
+            return None
         head_weight = _float_vector(entry.get("head_weight"))
         head_bias = _float_entry(entry, "head_bias", 0.0)
         if len(head_weight) != hidden_size:
-            return _predict_next_return("drift", returns)
+            return None
         normalized_prediction = sum(weight * value for weight, value in zip(head_weight, hidden)) + head_bias
         if not math.isfinite(normalized_prediction):
-            return _predict_next_return("drift", returns)
+            return None
         prediction = _clamp(normalized_prediction, -clip, clip) * scale + mean
     except (IndexError, KeyError, TypeError, ValueError, OverflowError):
-        return _predict_next_return("drift", returns)
+        return None
 
     recent_abs = sorted(abs(value) for value in returns[-48:]) if len(returns) >= 8 else [0.01]
     cap = max(recent_abs[int(len(recent_abs) * 0.9)], 0.0002)
@@ -440,6 +352,13 @@ def _torch_gate_values(
     return gates
 
 
+def _torch_validation_mae(entry: dict[str, Any], returns: list[float]) -> float:
+    mae_percent = _float_entry(entry, "validation_mae_percent", 0.0)
+    if mae_percent > 0:
+        return mae_percent / 100
+    return _return_scale(returns)
+
+
 def _float_entry(data: dict[str, Any], key: str, default: float) -> float:
     value = data.get(key)
     if isinstance(value, (int, float)):
@@ -486,65 +405,6 @@ def _sigmoid(value: float) -> float:
     return 1 / (1 + math.exp(-value))
 
 
-def _ols(rows: list[list[float]], targets: list[float], ridge: float = 1e-8) -> list[float] | None:
-    if not rows:
-        return None
-    columns = len(rows[0])
-    xtx = [[0.0 for _ in range(columns)] for _ in range(columns)]
-    xty = [0.0 for _ in range(columns)]
-    for row, target in zip(rows, targets):
-        for i in range(columns):
-            xty[i] += row[i] * target
-            for j in range(columns):
-                xtx[i][j] += row[i] * row[j]
-    for i in range(columns):
-        xtx[i][i] += ridge
-    return _solve_linear_system(xtx, xty)
-
-
-def _solve_linear_system(matrix: list[list[float]], vector: list[float]) -> list[float] | None:
-    size = len(vector)
-    augmented = [row[:] + [vector[index]] for index, row in enumerate(matrix)]
-    for col in range(size):
-        pivot = max(range(col, size), key=lambda row: abs(augmented[row][col]))
-        if abs(augmented[pivot][col]) < 1e-12:
-            return None
-        augmented[col], augmented[pivot] = augmented[pivot], augmented[col]
-        pivot_value = augmented[col][col]
-        for item in range(col, size + 1):
-            augmented[col][item] /= pivot_value
-        for row in range(size):
-            if row == col:
-                continue
-            factor = augmented[row][col]
-            for item in range(col, size + 1):
-                augmented[row][item] -= factor * augmented[col][item]
-    return [augmented[row][size] for row in range(size)]
-
-
-def _ewma_volatility(returns: list[float], decay: float) -> float:
-    if not returns:
-        return 0.0
-    decay = _clamp(decay, 0.80, 0.995)
-    variance = returns[0] * returns[0]
-    for value in returns[1:]:
-        variance = decay * variance + (1 - decay) * value * value
-    return math.sqrt(max(variance, 0.0))
-
-
-def _fixed_garch_volatility(returns: list[float]) -> float:
-    if not returns:
-        return 0.0
-    long_variance = sum(value * value for value in returns) / len(returns)
-    alpha = 0.08
-    beta = 0.90
-    omega = max(1e-12, (1 - alpha - beta) * long_variance)
-    variance = long_variance
-    for value in returns:
-        variance = omega + alpha * value * value + beta * variance
-    return math.sqrt(max(variance, 0.0))
-
-
 def _confidence_adjustment(
     *,
     expected_return_percent: float,
@@ -552,10 +412,7 @@ def _confidence_adjustment(
     skill: float,
     min_edge: float,
     max_adjustment: float,
-    usable_skill: bool,
 ) -> float:
-    if not usable_skill:
-        return 0.0
     edge = abs(expected_return_percent) - min_edge
     if edge <= 0:
         return 0.0
@@ -564,7 +421,7 @@ def _confidence_adjustment(
         return 0.0
     strength = _clamp(edge / max(min_edge, 0.05), 0.0, 1.0)
     probability_strength = _clamp(abs(probability_up - 0.5) / 0.25, 0.0, 1.0)
-    skill_strength = _clamp(skill / 0.18, 0.0, 1.0)
+    skill_strength = _clamp((skill + 0.03) / 0.18, 0.25, 1.0)
     return direction * _clamp(max_adjustment, 0.0, 0.18) * strength * probability_strength * skill_strength
 
 
@@ -575,10 +432,7 @@ def _reason(
     probability_up: float,
     skill: float,
     block_entry: bool,
-    usable_skill: bool,
 ) -> str:
-    if not usable_skill:
-        return f"модель {model} не лучше baseline на walk-forward проверке"
     if block_entry:
         return f"модель {model}: ожидаемое движение вниз {expected_return_percent:.3f}%, P(рост)={probability_up:.2f}"
     return f"модель {model}: прогноз {expected_return_percent:.3f}%, P(рост)={probability_up:.2f}, skill={skill:.3f}"