Improve learning discovery and dashboard i18n

app/behavior/engine.py

@@ -48,10 +48,27 @@ _MAX_DECISION_TRACES = 30
 _MAX_LATENCY_MEASUREMENTS = 50
 _MAX_FEEDBACK_LOG = 50
 _ACTION_LOGBOOK_TOLERANCE = timedelta(seconds=10)
-_CONTEXT_TRIGGER_TOLERANCE = timedelta(seconds=3)
+_CONTEXT_TRIGGER_TOLERANCE = timedelta(minutes=4)
 _OWN_ACTION_TOLERANCE = timedelta(seconds=20)
-_SAFE_ACTIVE_DOMAINS = frozenset({"cover", "fan", "humidifier", "light", "switch"})
+_SAFE_ACTIVE_DOMAINS = frozenset({
+    "button",
+    "cover",
+    "fan",
+    "humidifier",
+    "input_button",
+    "light",
+    "switch",
+})
 _AUTOMATION_CONTEXT_DOMAINS = frozenset({"automation", "script"})
+_LIGHT_TARGET_ATTRIBUTES = frozenset({
+    "brightness",
+    "color_temp",
+    "color_temp_kelvin",
+    "effect",
+    "hs_color",
+    "rgb_color",
+    "xy_color",
+})
 logger = logging.getLogger(__name__)
 
 
@@ -339,7 +356,7 @@ class BehaviorEngine:
             now=now,
             min_support=self._settings.min_behavior_actions,
             window_minutes=self._settings.prediction_window_minutes,
-            causal_window_seconds=self._settings.prediction_interval_seconds * 2,
+            causal_window_seconds=max(self._settings.prediction_interval_seconds * 2, 240),
             timezone_name=self._settings.timezone,
         )
         if prediction is not None:
@@ -439,7 +456,11 @@ class BehaviorEngine:
                     self._ha_reader.call_service(
                         domain,
                         service,
-                        {"entity_id": actuator_entity_id},
+                        _service_data_for_prediction(
+                            actuator_entity_id,
+                            domain,
+                            prediction,
+                        ),
                     )
                     decision_to_service_ms = _elapsed_ms(service_started_perf)
                 except (HaClientError, ValueError) as exc:
@@ -1107,7 +1128,7 @@ class BehaviorEngine:
             blockers.append(
                 f"Sicherheit {prediction.confidence:.0%} liegt unter der Schwelle {threshold:.0%}."
             )
-        if current_state == prediction.target_state:
+        if _target_reached(record.actuator_entity_id, current_state, prediction):
             blockers.append("Zielzustand ist bereits erreicht.")
         if not self._cooldown_elapsed(
             record.behavior,
@@ -1150,12 +1171,16 @@ class BehaviorEngine:
             patterns.append(
                 BehaviorPattern(
                     target_state=point.state,
+                    target_attributes=_target_attributes_for(point),
                     minute_of_day=local.hour * 60 + local.minute,
                     weekday=local.weekday(),
                     context_states=contexts,
-                    trigger_entity_id=trigger[0] if trigger else None,
-                    trigger_from_state=trigger[1] if trigger else None,
-                    trigger_to_state=trigger[2] if trigger else None,
+                    trigger_entity_id=trigger[1] if trigger else None,
+                    trigger_from_state=trigger[2] if trigger else None,
+                    trigger_to_state=trigger[3] if trigger else None,
+                    trigger_delay_seconds=(
+                        int(trigger[0].total_seconds()) if trigger else None
+                    ),
                     source=source,
                     weight=weight,
                     observed_at=point.timestamp,
@@ -1941,6 +1966,7 @@ def predict_behavior(
     minute_of_day = local.hour * 60 + local.minute
     changed_at = current_context_changed_at or {}
     by_state: dict[str, list[float]] = {}
+    attributes_by_state: dict[str, list[tuple[float, dict[str, object]]]] = {}
     causal_support_by_state: dict[str, int] = {}
     for pattern in patterns:
         if pattern.trigger_entity_id and pattern.trigger_to_state:
@@ -1954,7 +1980,11 @@ def predict_behavior(
                 current_context.get(pattern.trigger_entity_id)
                 == pattern.trigger_to_state
                 and trigger_age is not None
-                and 0 <= trigger_age <= causal_window_seconds
+                and _trigger_age_matches(
+                    trigger_age,
+                    pattern.trigger_delay_seconds,
+                    causal_window_seconds,
+                )
             ):
                 continue
             comparable = [
@@ -1969,6 +1999,9 @@ def predict_behavior(
             )
             score = pattern.weight * (0.85 + 0.15 * context_score)
             by_state.setdefault(pattern.target_state, []).append(score)
+            attributes_by_state.setdefault(pattern.target_state, []).append(
+                (score, pattern.target_attributes)
+            )
             causal_support_by_state[pattern.target_state] = (
                 causal_support_by_state.get(pattern.target_state, 0) + 1
             )
@@ -1998,6 +2031,9 @@ def predict_behavior(
             0.45 * time_score + 0.45 * context_score + 0.10 * weekday_score
         )
         by_state.setdefault(pattern.target_state, []).append(score)
+        attributes_by_state.setdefault(pattern.target_state, []).append(
+            (score, pattern.target_attributes)
+        )
     if not by_state:
         return None
     target_state, scores = max(
@@ -2011,6 +2047,9 @@ def predict_behavior(
         return None
     return BehaviorPrediction(
         target_state=target_state,
+        target_attributes=_aggregate_target_attributes(
+            attributes_by_state.get(target_state, [])
+        ),
         confidence=round(confidence, 4),
         generated_at=now,
         matching_patterns=support,
@@ -2044,9 +2083,87 @@ def _weighted_context_score(
     return matched_weight / total_weight
 
 
+def _trigger_age_matches(
+    trigger_age_seconds: float,
+    expected_delay_seconds: int | None,
+    causal_window_seconds: int,
+) -> bool:
+    if trigger_age_seconds < 0:
+        return False
+    if expected_delay_seconds is None or expected_delay_seconds <= 10:
+        return trigger_age_seconds <= causal_window_seconds
+    tolerance = max(30, min(90, causal_window_seconds // 2))
+    return abs(trigger_age_seconds - expected_delay_seconds) <= tolerance
+
+
+def _aggregate_target_attributes(
+    weighted_attributes: list[tuple[float, dict[str, object]]],
+) -> dict[str, object]:
+    if not weighted_attributes:
+        return {}
+    result: dict[str, object] = {}
+    numeric_values: dict[str, list[tuple[float, float]]] = {}
+    categorical_values: dict[str, dict[str, float]] = {}
+    for score, attributes in weighted_attributes:
+        for key, value in attributes.items():
+            if key not in _LIGHT_TARGET_ATTRIBUTES:
+                continue
+            if isinstance(value, bool) or value is None:
+                continue
+            if isinstance(value, (int, float)):
+                numeric_values.setdefault(key, []).append((score, float(value)))
+            else:
+                categorical_values.setdefault(key, {}).setdefault(str(value), 0.0)
+                categorical_values[key][str(value)] += score
+    for key, values in numeric_values.items():
+        total_weight = sum(score for score, _ in values)
+        if total_weight <= 0:
+            continue
+        result[key] = round(sum(score * value for score, value in values) / total_weight)
+    for key, values in categorical_values.items():
+        if key in result:
+            continue
+        result[key] = max(values.items(), key=lambda item: (item[1], item[0]))[0]
+    return result
+
+
+def _target_attributes_for(point: StateHistoryPoint) -> dict[str, object]:
+    if point.state != "on":
+        return {}
+    return {
+        key: value
+        for key, value in point.attributes.items()
+        if key in _LIGHT_TARGET_ATTRIBUTES and value is not None
+    }
+
+
+def _service_data_for_prediction(
+    actuator_entity_id: str,
+    domain: str,
+    prediction: BehaviorPrediction,
+) -> dict[str, object]:
+    data: dict[str, object] = {"entity_id": actuator_entity_id}
+    if domain == "light" and prediction.target_state == "on":
+        data.update(prediction.target_attributes)
+    return data
+
+
+def _target_reached(
+    actuator_entity_id: str,
+    current_state: str,
+    prediction: BehaviorPrediction,
+) -> bool:
+    domain = actuator_entity_id.split(".", 1)[0]
+    if domain == "light" and prediction.target_state == "on" and prediction.target_attributes:
+        return False
+    return current_state == prediction.target_state
+
+
 def service_for_state(domain: str, target_state: str) -> str | None:
     if domain in {"fan", "humidifier", "light", "media_player", "remote", "switch"}:
         return {"on": "turn_on", "off": "turn_off"}.get(target_state)
+    if domain in {"button", "input_button"}:
+        return "press"
     if domain == "scene":
         return "turn_on" if target_state == "on" else None
     if domain == "cover":
@@ -2112,7 +2229,7 @@ def _recent_context_transition(
     history: dict[str, StateHistorySeries],
     context_ids: list[str],
     timestamp: datetime,
-) -> tuple[str, str, str] | None:
+) -> tuple[timedelta, str, str, str] | None:
     nearest: tuple[timedelta, str, str, str] | None = None
     for entity_id in context_ids:
         series = history.get(entity_id)
@@ -2131,7 +2248,7 @@ def _recent_context_transition(
             previous_state = point.state
     if nearest is None:
         return None
-    return nearest[1], nearest[2], nearest[3]
+    return nearest
 
 
 def _circular_minute_distance(left: int, right: int) -> int: