feat: fill configs and write to disk on load

.vscode/settings.json

@@ -1,7 +1,7 @@
 {
     "python.analysis.typeCheckingMode": "strict",
     "python.testing.pytestArgs": [
-        "."
+        "psycop"
     ],
     "python.testing.unittestEnabled": false,
     "python.testing.pytestEnabled": true,

psycop/common/model_training_v2/config/baseline_test_config.cfg

@@ -27,7 +27,6 @@ validation_outcome_col_name = "outcome"
 [trainer.preprocessing_pipeline.*.age_filter]
 @preprocessing = "age_filter"
 min_age = 4
-max_age = 99
 age_col_name = "pred_age"
 
 [trainer.task]

psycop/common/model_training_v2/config/config_utils.py

@@ -17,12 +17,31 @@
 def load_baseline_config(config_path: Path) -> BaselineSchema:
     """Loads the baseline config from disk and resolves it."""
     cfg = Config().from_disk(config_path)
-    resolved = BaselineRegistry.resolve(cfg)
+
+    # Fill with defaults and write to disk if relevant
+    filled = BaselineRegistry.fill(cfg, validate=False)
+    resolved = BaselineRegistry.resolve(filled)
+
+    # Writing to disk happens after resolution, to ensure that the
+    # config is valid
+    if cfg != filled:
+        filled.to_disk(config_path)
+
     return BaselineSchema(**resolved)
 
 
 def load_hyperparam_config(config_path: Path) -> dict[str, Any]:
     """Loads the baseline config from disk and resolves it."""
     cfg = Config().from_disk(config_path)
-    resolved = BaselineRegistry.resolve(cfg)
+
+    # Fill with defaults and write to disk if relevant
+    filled = BaselineRegistry.fill(cfg, validate=False)
+    resolved = BaselineRegistry.resolve(filled)
+
+    # Writing to disk happens after resolution, to ensure that the
+    # config is valid
+    if cfg != filled:
+        filled.to_disk(config_path)
+
+    resolved = BaselineRegistry.resolve(filled)
     return resolved

psycop/common/model_training_v2/hyperparameter_suggester/test_hyperparam_search.cfg

@@ -1,4 +1,14 @@
 [model]
 @estimator_steps = "logistic_regression_suggester"
-C = {"low": 0.1, "high": 1.0, "logarithmic": "True"}
-l1_ratio = {"low": 0.01, "high": 1.0, "logarithmic": "True"}
+solvers = ["saga"]
+penalties = ["l1","l2","elasticnet"]
+
+[model.C]
+low = 0.1
+high = 1.0
+logarithmic = "True"
+
+[model.l1_ratio]
+low = 0.01
+high = 1.0
+logarithmic = "True"

psycop/common/model_training_v2/trainer/preprocessing/pipeline.py

@@ -21,9 +21,7 @@ def apply(self, data: PolarsFrame) -> pd.DataFrame:
 
 
 @BaselineRegistry.preprocessing.register("baseline_preprocessing_pipeline")
-class BaselinePreprocessingPipeline(
-    PreprocessingPipeline,
-):  # TODO: #406 Does registering this into a registry remove protocol checking? E.g. the __init__ method does not adhered to PreprocessingPipeline
+class BaselinePreprocessingPipeline(PreprocessingPipeline):
     def __init__(self, *args: PresplitStep) -> None:
         self.steps = list(args)
 

psycop/common/model_training_v2/trainer/preprocessing/steps/row_filters.py

@@ -9,7 +9,12 @@
 
 @BaselineRegistry.preprocessing.register("age_filter")
 class AgeFilter(PresplitStep):
-    def __init__(self, min_age: int, max_age: int, age_col_name: str):
+    def __init__(
+        self,
+        min_age: int,
+        max_age: int = 999,
+        age_col_name: str = "pred_age",
+    ):
         self.min_age = min_age
         self.max_age = max_age
         self.age = pl.col(age_col_name)