feat: update cfg experiments (#909)

psycop/projects/cvd/model_training/cvd_baseline.cfg

@@ -6,7 +6,7 @@
 
 [logger.*.mlflow]
 @loggers = "mlflow_logger"
-experiment_name = "cvd_hyperparam_tuning_joblib"
+experiment_name = "baseline_v2_cvd"
 postpone_run_creation_to_first_log = True
 
 [trainer]
@@ -21,7 +21,7 @@ group_col_name = "dw_ek_borger"
 
 [trainer.training_data]
 @data = "parquet_vertical_concatenator"
-paths = ["E:/shared_resources/cvd/feature_set/flattened_datasets/cvd_lookbehind_experiments/train.parquet", "E:/shared_resources/cvd/feature_set/flattened_datasets/cvd_lookbehind_experiments/val.parquet"]
+paths = ["E:/shared_resources/cvd/feature_set/flattened_datasets/train.parquet", "E:/shared_resources/cvd/feature_set/flattened_datasets/val.parquet"]
 validate_on_init = False
 
 #################
@@ -45,12 +45,12 @@ age_col_name = "pred_age_in_years"
 [trainer.preprocessing_pipeline.*.layer_selector]
 @preprocessing = "filter_columns_within_subset"
 subset_rule = "pred_.+layer.+"
-keep_matching = ".+_layer_(1|2|3).+"
+keep_matching = ".+_layer_(1).+"
 
-# [trainer.preprocessing_pipeline.*.aggregation_selector]
-# @preprocessing = "filter_columns_within_subset"
-# subset_rule = "pred_.+layer.+"
-# keep_matching = ".+_(mean|max|min)_.+"
+[trainer.preprocessing_pipeline.*.aggregation_selector]
+@preprocessing = "filter_columns_within_subset"
+subset_rule = "pred_.+layer.+"
+keep_matching = ".+_(mean)_.+"
 
 ## Outcomes
 [trainer.preprocessing_pipeline.*.outcome_selector]
@@ -75,7 +75,7 @@ column_expectations = [["outc_", 1], ["prediction_timestamp", 0]]
 # Estimator #
 #############
 [trainer.task.task_pipe.sklearn_pipe.*.model]
-@estimator_steps_suggesters = "lightgbm_suggester"
+@estimator_steps = "xgboost"
 
 ########
 # Task #

psycop/projects/cvd/model_training/train_cvd_layers.py

@@ -0,0 +1,39 @@
+import logging
+from pathlib import Path
+
+import confection
+
+from psycop.common.model_training_v2.config.baseline_pipeline import train_baseline_model_from_cfg
+from psycop.common.model_training_v2.config.populate_registry import populate_baseline_registry
+from psycop.projects.cvd.model_training.populate_cvd_registry import populate_with_cvd_registry
+
+if __name__ == "__main__":
+    import coloredlogs
+
+    coloredlogs.install(  # type: ignore
+        level="INFO",
+        format="%(asctime)s [%(levelname)s] %(name)s: %(message)s",
+        datefmt="%Y/%m/%d %H:%M:%S",
+    )
+
+    populate_baseline_registry()
+    populate_with_cvd_registry()
+
+    for layer in range(1, 5):
+        cfg = confection.Config().from_disk(Path(__file__).parent / "cvd_baseline.cfg")
+        layers = [str(i) for i in range(1, layer + 1)]
+        cfg["trainer"]["preprocessing_pipeline"]["*"]["layer_selector"][
+            "keep_matching"
+        ] = f".+_layer_({'|'.join(layers)}).+"
+
+        logging.info(f"Training model with layers {layers}")
+        train_baseline_model_from_cfg(cfg=cfg)
+
+        if layer == 1:
+            aggs = ".+(mean|min|max).+"
+            cfg["trainer"]["preprocessing_pipeline"]["*"]["aggregation_selector"][
+                "keep_matching"
+            ] = aggs
+
+            logging.info(f"Training model with {aggs}")
+            train_baseline_model_from_cfg(cfg=cfg)

psycop/projects/cvd/model_training/train_cvd_lookbehinds.py

@@ -0,0 +1,26 @@
+from pathlib import Path
+
+import confection
+
+from psycop.common.model_training_v2.config.baseline_pipeline import train_baseline_model_from_cfg
+from psycop.common.model_training_v2.config.populate_registry import populate_baseline_registry
+from psycop.projects.cvd.model_training.populate_cvd_registry import populate_with_cvd_registry
+
+if __name__ == "__main__":
+    import coloredlogs
+
+    coloredlogs.install(  # type: ignore
+        level="INFO",
+        format="%(asctime)s [%(levelname)s] %(name)s: %(message)s",
+        datefmt="%Y/%m/%d %H:%M:%S",
+    )
+
+    populate_baseline_registry()
+    populate_with_cvd_registry()
+
+    cfg = confection.Config().from_disk(Path(__file__).parent / "cvd_baseline.cfg")
+    cfg["trainer"]["training_data"]["paths"] = [
+        f"E:/shared_resources/cvd/feature_set/flattened_datasets/cvd_lookbehind_experiments/{split}.parquet"
+        for split in ["train", "test"]
+    ]
+    train_baseline_model_from_cfg(cfg)