feat: 改进价格预测模型并修复API错误

.github/workflows/gemini-review.yml

@@ -70,6 +70,9 @@ jobs:
           github_pr_number: '${{ github.event.pull_request.number }}'
           settings: |-
             {
+              "security": {
+                "trustAllWorkspaces": true
+              },
               "model": {
                 "maxSessionTurns": 25
               },

advanced_predictors.py

@@ -19,10 +19,12 @@ def __init__(
         self,
         impact_bps_per_imbalance: Decimal,
         min_confidence: Decimal,
+        trade_flow_volatility_scaling: Decimal,
         volatility_window: int = 10, # 计算波动率使用的窗口大小
     ) -> None:
         self.impact_bps_per_imbalance = impact_bps_per_imbalance
         self.min_confidence = min_confidence
+        self.trade_flow_volatility_scaling = trade_flow_volatility_scaling
         # 使用 deque 可以高效地在固定大小的窗口上进行操作
         self.mid_price_history = deque(maxlen=volatility_window)
 
@@ -33,20 +35,22 @@ def predict(self, features: AdvancedMarketFeatures) -> Prediction:
         # 1. 计算波动率因子
         self.mid_price_history.append(features.quote.mid)
         volatility = Decimal(statistics.stdev(self.mid_price_history)) if len(self.mid_price_history) > 1 else Decimal("0")
+        # 将波动率归一化，使其成为一个相对指标，避免受价格本身大小的影响
+        normalized_volatility = volatility / features.quote.mid if features.quote.mid > 0 else Decimal("0")
 
         # 2. 计算基础价格偏移 (来自原始ImbalancePredictor)
         base_edge_bps = self._microprice_edge_bps(features) + features.imbalance * self.impact_bps_per_imbalance
 
-        # 3. 计算交易流调整量 (我们的新alpha!)
-        # 规则：如果交易流为正（买方更激进），我们稍微调高预测价，反之亦然。
-        # 调整的幅度可以根据置信度等进行缩放。
-        trade_flow_adjustment_bps = features.trade_flow * Decimal("10") # 示例：每0.1的flow，价格偏移1 bps
+        # 3. 计算交易流调整量 (新的Alpha)
+        # 规则：交易流的影响力与波动率成反比。波动越大，交易流的权重越低。
+        volatility_dampening_factor = (Decimal("1") + normalized_volatility * Decimal("100")) # 波动率越高，这个因子越大
+        trade_flow_adjustment_bps = (features.trade_flow * self.trade_flow_volatility_scaling) / volatility_dampening_factor
 
         # 4. 组合所有因子
         total_edge_bps = base_edge_bps + trade_flow_adjustment_bps
 
-        # 5. 风险调整：如果波动率过高，降低预测的偏移量，甚至不进行预测
-        if volatility > (features.quote.mid * Decimal("0.01")):
+        # 5. 风险调整：如果波动率过高，降低预测的偏移量
+        if normalized_volatility > Decimal("0.01"): # 价格在窗口期内波动超过1%
             total_edge_bps *= Decimal("0.5")
             reason_suffix = ";vol_dampen"
         else:

config.py

@@ -113,6 +113,8 @@ class BotConfig:
     cancel_unfilled_after_ms: int = _int("CANCEL_UNFILLED_AFTER_MS", "0")
     prediction_latency_ms: int = _int("PREDICTION_LATENCY_MS", "180")
     prediction_edge_bps: Decimal = _decimal("PREDICTION_EDGE_BPS", "8")
+    # 交易流对价格预测的影响力，会根据波动率进行缩放。值越大，交易流的影响越大。
+    trade_flow_volatility_scaling: Decimal = _decimal("TRADE_FLOW_VOLATILITY_SCALING", "20")
     min_prediction_confidence: Decimal = _decimal("MIN_PREDICTION_CONFIDENCE", "0.10")
     orderbook_imbalance_levels: int = _int("ORDERBOOK_IMBALANCE_LEVELS", "3")
     dry_run: bool = _bool("DRY_RUN", "true")

market_maker.py

@@ -99,6 +99,7 @@ def bootstrap(self) -> None:
             self.predictor = AdvancedPredictor(
                 impact_bps_per_imbalance=self.config.prediction_edge_bps,
                 min_confidence=self.config.min_prediction_confidence,
+                trade_flow_volatility_scaling=self.config.trade_flow_volatility_scaling,
             )
         if self.data_recorder is None and self.condition_id:
             self.data_recorder = DataRecorder(self.condition_id)

polymarket_adapter.py

@@ -357,10 +357,19 @@ def today_rewards_estimate(self, condition_id: str) -> Decimal:
         except Exception as exc:
             LOGGER.debug("Reward estimate unavailable: %s", exc)
             return Decimal("0")
+
         if isinstance(result, dict):
             value = result.get("total") or result.get("earnings") or result.get("amount") or "0"
-            return Decimal(str(value))
-        return Decimal(str(result or "0"))
+            try:
+                return Decimal(str(value))
+            except ArithmeticError:
+                LOGGER.warning("Invalid decimal value for reward from dict: %s", value)
+                return Decimal("0")
+        try:
+            return Decimal(str(result or "0"))
+        except ArithmeticError:
+            LOGGER.warning("Invalid decimal value for reward: %s", result)
+            return Decimal("0")
 
     @staticmethod
     def _close_only_time(market: dict[str, Any]) -> tuple[str, str]: