Build the pruning tests' objective from the unit that fits

Two optimizer test files built the objective they measure by reaching for
HoldoutEvaluationStrategy.evaluate. That method is gone: the strategies declare
how a run is evaluated and the fitting unit carries it out, so the objective
comes from there now. The tests themselves are unchanged -- they still check
that a bad trial is pruned, that disabling the pruner completes every trial,
and that a real model reports each epoch to its trial.

Which partitions a trial records is still read off the strategy class, the same
way the job reads it, so the declaration stays connected to what it produces.

These four failures were not caught earlier because the verification runs had
been narrowed to the directories this work was touching -- units, dag, spike,
api and evaluation -- after the full suite was dropped for containing a test
that builds the app against the real ~/.DashAI. Deselecting that one test was
the right answer; shrinking the suite to what seemed relevant was not, and it
is precisely the change that removes a caller elsewhere that this hides.

Whole suite: 3682 passed, one test deselected.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: Felipedino

tests/back/optimizers/test_optuna_pruning_integration.py

@@ -29,18 +29,32 @@
 from DashAI.back.evaluation.holdout import HoldoutEvaluationStrategy
 from DashAI.back.models.base_model import BaseModel
 from DashAI.back.optimizers.optuna_optimizer import OptunaOptimizer
+from DashAI.back.units.fit_model_unit import FitModelUnit
 
 
 def _holdout_evaluate(model, input_dataset, output_dataset, metric):
-    """The real holdout evaluation path, on a strategy with no factory.
-
-    `evaluate` reads which partitions its strategy scores, so it needs a real
-    instance rather than None for self. Building one through __init__ would
-    need a `ModelFactory` this test does not have, and does not need: the only
-    thing read off the instance is a class attribute.
+    """The real objective of a holdout search: one fit, then score validation.
+
+    It used to be ``HoldoutEvaluationStrategy.evaluate``. The strategies now
+    only declare how a run is evaluated, and the unit that fits carries it out,
+    so the objective the optimizer measures comes from there. Which partitions
+    a trial records is still the strategy's declaration -- the job reads
+    SCORED_SPLITS off it and hands it over, minus the test partition, which a
+    trial may never score -- so it is read off the class here too.
     """
-    strategy = HoldoutEvaluationStrategy.__new__(HoldoutEvaluationStrategy)
-    return strategy.evaluate(model, input_dataset, output_dataset, metric)
+    trial_splits = [
+        split.name
+        for split in HoldoutEvaluationStrategy.SCORED_SPLITS
+        if split.name != "TEST"
+    ]
+    unit = FitModelUnit(
+        optimizer={"component": "", "params": {}},
+        goal_metric="",
+        run_id=None,
+        artifact_prefix=None,
+        trial_splits=trial_splits,
+    )
+    return unit._score_one_trial(model, input_dataset, output_dataset, metric)
 
 
 EPOCHS = 12

tests/back/optimizers/test_optuna_real_model.py

@@ -40,18 +40,32 @@
 from DashAI.back.metrics.classification.accuracy import Accuracy
 from DashAI.back.models.mlp_image_classifier import MLPImageClassifier
 from DashAI.back.optimizers.optuna_optimizer import OptunaOptimizer
+from DashAI.back.units.fit_model_unit import FitModelUnit
 
 
 def _holdout_evaluate(model, input_dataset, output_dataset, metric):
-    """The real holdout evaluation path, on a strategy with no factory.
-
-    `evaluate` reads which partitions its strategy scores, so it needs a real
-    instance rather than None for self. Building one through __init__ would
-    need a `ModelFactory` this test does not have, and does not need: the only
-    thing read off the instance is a class attribute.
+    """The real objective of a holdout search: one fit, then score validation.
+
+    It used to be ``HoldoutEvaluationStrategy.evaluate``. The strategies now
+    only declare how a run is evaluated, and the unit that fits carries it out,
+    so the objective the optimizer measures comes from there. Which partitions
+    a trial records is still the strategy's declaration -- the job reads
+    SCORED_SPLITS off it and hands it over, minus the test partition, which a
+    trial may never score -- so it is read off the class here too.
     """
-    strategy = HoldoutEvaluationStrategy.__new__(HoldoutEvaluationStrategy)
-    return strategy.evaluate(model, input_dataset, output_dataset, metric)
+    trial_splits = [
+        split.name
+        for split in HoldoutEvaluationStrategy.SCORED_SPLITS
+        if split.name != "TEST"
+    ]
+    unit = FitModelUnit(
+        optimizer={"component": "", "params": {}},
+        goal_metric="",
+        run_id=None,
+        artifact_prefix=None,
+        trial_splits=trial_splits,
+    )
+    return unit._score_one_trial(model, input_dataset, output_dataset, metric)
 
 
 EPOCHS = 3