feat(multitask): forward max_time_spotoptim to SpotOptim

pyproject.toml

@@ -31,7 +31,9 @@ dependencies = [
     "ruff>=0.15.6",
     "scikit-learn>=1.8.0",
     "shap>=0.49.1",
-    "spotforecast2-safe>=21.0.0,<22",
+    # 21.2.0 added the max_time_spotoptim config field that SpotOptimStrategy
+    # forwards as SpotOptim's max_time.
+    "spotforecast2-safe>=21.2.0,<22",
     # spotoptim 1.0 is sequential-only and lean: torch/tensorboard moved to its
     # ``[torch]`` extra. sf2 forwards tensorboard_* kwargs into SpotOptim, so we
     # pin the extra to keep the TensorBoard tuning dashboards working (they were

src/spotforecast2/multitask/strategies.py

@@ -230,6 +230,13 @@ def prepare_forecaster(
         task its own ``tensorboard_path``, when several tasks share one
         log directory.
 
+        Time budget: when ``config.max_time_spotoptim`` is set (wall-clock
+        minutes, a first-class `ConfigMulti`/`ConfigEntsoe` field in
+        sf2-safe >= 21.2.0), it is forwarded as SpotOptim's ``max_time``;
+        the search stops after ``n_trials_spotoptim`` evaluations or that
+        time limit, whichever comes first. ``None`` (the default) means no
+        time limit.
+
         Args:
             task: A `BaseTask` (or compatible) instance that supplies ``cv_ts``,
                 ``config``, ``logger``, ``save_tuning_results``, and
@@ -351,6 +358,16 @@ def prepare_forecaster(
             kwargs_spotoptim.update(tb_kwargs)
             task.logger.info("  SpotOptim TensorBoard: %s", tb_kwargs)
 
+        # Wall-clock budget: forward ``config.max_time_spotoptim`` (minutes,
+        # sf2-safe >= 21.2.0) as SpotOptim's ``max_time``. The search then
+        # stops at ``n_trials_spotoptim`` evaluations or the time limit,
+        # whichever comes first. ``None`` (the default) forwards nothing and
+        # leaves SpotOptim's own default (no limit) in charge.
+        max_time = getattr(task.config, "max_time_spotoptim", None)
+        if max_time is not None:
+            kwargs_spotoptim["max_time"] = float(max_time)
+            task.logger.info("  SpotOptim max_time: %.2f min", float(max_time))
+
         tuning_results, _ = spotoptim_search_forecaster(
             forecaster=forecaster,
             y=y_train,

tests/test_multitask_strategies.py

@@ -204,3 +204,49 @@ def fake_search_forecaster(*args, **kwargs):
 
     # No tensorboard attrs -> empty kwargs dict -> None passed through.
     assert captured["kwargs_spotoptim"] is None
+
+
+def test_spotoptim_strategy_forwards_max_time(monkeypatch):
+    """config.max_time_spotoptim (minutes) must reach kwargs_spotoptim as max_time."""
+    import pandas as pd
+
+    import spotforecast2.model_selection as ms
+
+    captured = {}
+
+    def fake_search_forecaster(*args, **kwargs):
+        captured["kwargs_spotoptim"] = kwargs.get("kwargs_spotoptim")
+        results = pd.DataFrame({"params": [{"alpha": 1.0}], "lags": [[1, 2]]})
+        return results, object()
+
+    monkeypatch.setattr(ms, "spotoptim_search_forecaster", fake_search_forecaster)
+
+    task = _make_fake_task(max_time_spotoptim=2.5)
+    SpotOptimStrategy(search_space={"alpha": (0.1, 1.0)}).prepare_forecaster(
+        task, "A", _FakeForecaster(), y_train=None
+    )
+
+    assert captured["kwargs_spotoptim"]["max_time"] == 2.5
+
+
+def test_spotoptim_strategy_none_max_time_not_forwarded(monkeypatch):
+    """max_time_spotoptim=None (the config default) must forward nothing."""
+    import pandas as pd
+
+    import spotforecast2.model_selection as ms
+
+    captured = {}
+
+    def fake_search_forecaster(*args, **kwargs):
+        captured["kwargs_spotoptim"] = kwargs.get("kwargs_spotoptim")
+        results = pd.DataFrame({"params": [{"alpha": 1.0}], "lags": [[1, 2]]})
+        return results, object()
+
+    monkeypatch.setattr(ms, "spotoptim_search_forecaster", fake_search_forecaster)
+
+    task = _make_fake_task(max_time_spotoptim=None)
+    SpotOptimStrategy(search_space={"alpha": (0.1, 1.0)}).prepare_forecaster(
+        task, "A", _FakeForecaster(), y_train=None
+    )
+
+    assert captured["kwargs_spotoptim"] is None