Upload 6 files

app/ml/gating.py

@@ -1,6 +1,9 @@
 from typing import Any, Dict, List, Tuple
 
-from sundew.gating import gate_probability_with_hysteresis, significance_score
+try:
+    from sundew.gating import gate_probability_with_hysteresis, significance_score
+except ImportError:
+    from app.ml.sundew_fallback import gate_probability_with_hysteresis, significance_score
 
 
 def _clamp(value: float, low: float = 0.0, high: float = 1.0) -> float:

app/ml/sundew_fallback.py

@@ -0,0 +1,66 @@
+"""
+Fallback implementation of Sundew gating algorithms.
+Replace with actual sundew-algorithms package for production.
+"""
+import math
+
+
+def significance_score(
+    features: dict,
+    w_mag: float = 0.35,
+    w_ano: float = 0.4,
+    w_ctx: float = 0.15,
+    w_urg: float = 0.1
+) -> float:
+    """
+    Compute significance score for a signal window.
+
+    Args:
+        features: Dict with keys: magnitude, anomaly_score, context_relevance, urgency
+        w_mag: Weight for magnitude
+        w_ano: Weight for anomaly score
+        w_ctx: Weight for context relevance
+        w_urg: Weight for urgency
+
+    Returns:
+        Significance score in [0, 1]
+    """
+    score = (
+        w_mag * min(features.get("magnitude", 0.0) / 100.0, 1.0) +
+        w_ano * features.get("anomaly_score", 0.0) +
+        w_ctx * features.get("context_relevance", 0.0) +
+        w_urg * features.get("urgency", 0.0)
+    )
+    return max(0.0, min(1.0, score))
+
+
+def gate_probability_with_hysteresis(
+    significance: float,
+    threshold: float = 0.6,
+    temperature: float = 0.1,
+    last_activation: bool = False
+) -> float:
+    """
+    Convert significance to gate probability with hysteresis.
+
+    Args:
+        significance: Significance score in [0, 1]
+        threshold: Base threshold for gating
+        temperature: Softness of the gate (0 = hard threshold)
+        last_activation: Whether previous window was active (for hysteresis)
+
+    Returns:
+        Probability of keeping the window in [0, 1]
+    """
+    # Apply hysteresis: lower threshold if last was active
+    effective_threshold = threshold - (0.1 if last_activation else 0.0)
+
+    if temperature <= 0:
+        # Hard threshold
+        return 1.0 if significance >= effective_threshold else 0.0
+
+    # Soft threshold using sigmoid
+    logit = (significance - effective_threshold) / temperature
+    prob = 1.0 / (1.0 + math.exp(-logit))
+
+    return prob