Set l_rate correctly and ensure no overlap between subjects in dataset splits in CovidHierarchicalModel config

InnerEye/ML/configs/classification/CovidHierarchicalModel.py

@@ -1,6 +1,8 @@
 import codecs
 import logging
 import pickle
+import random
+import math
 from pathlib import Path
 
 from typing import Any, Callable
@@ -62,7 +64,6 @@ class CovidHierarchicalModel(ScalarModelBase):
                                         "is assumed to contain unique ids.")
 
     def __init__(self, covid_dataset_id: str = COVID_DATASET_ID, **kwargs: Any):
-        learning_rate = 1e-5 if self.use_pretrained_model else 1e-4
         super().__init__(target_names=['CVX03vs12', 'CVX0vs3', 'CVX1vs2'],
                          loss_type=ScalarLoss.CustomClassification,
                          class_names=['CVX0', 'CVX1', 'CVX2', 'CVX3'],
@@ -81,10 +82,13 @@ def __init__(self, covid_dataset_id: str = COVID_DATASET_ID, **kwargs: Any):
                          num_epochs=50,
                          l_rate_scheduler=LRSchedulerType.Step,
                          l_rate_step_gamma=1.0,
-                         l_rate=learning_rate,
                          l_rate_multi_step_milestones=None,
                          **kwargs)
         self.num_classes = 3
+
+    def validate(self) -> None:
+        self.l_rate = 1e-5 if self.use_pretrained_model else 1e-4
+        super().validate()
         if not self.use_pretrained_model and self.freeze_encoder:
             raise ValueError("No encoder to freeze when training from scratch. You requested training from scratch and"
                              "encoder freezing.")
@@ -94,15 +98,30 @@ def should_generate_multilabel_report(self) -> bool:
 
     def get_model_train_test_dataset_splits(self, dataset_df: pd.DataFrame) -> DatasetSplits:
         if self.test_set_ids_csv:
-            test_df = pd.read_csv(self.local_dataset / self.test_set_ids_csv)
-            in_test_set = dataset_df.series.isin(test_df.series)
-            train_ids = dataset_df.series[~in_test_set].values
-            test_ids = dataset_df.series[in_test_set].values
-            num_val_samples = 400
-            val_ids = train_ids[:num_val_samples]
-            train_ids = train_ids[num_val_samples:]
-            return DatasetSplits.from_subject_ids(dataset_df, train_ids=train_ids, val_ids=val_ids, test_ids=test_ids,
-                                                  subject_column="series", group_column="subject")
+            test_set_ids_csv = self.local_dataset / self.test_set_ids_csv
+            test_series = pd.read_csv(test_set_ids_csv).series
+
+            all_series = dataset_df.series.values
+            check_all_test_series = all(test_series.isin(all_series))
+            if not check_all_test_series:
+                raise ValueError(f"Not all test series from {test_set_ids_csv} were found in the dataset.")
+
+            test_set_subjects = dataset_df[dataset_df.series.isin(test_series)].subject.values
+            train_and_val_series = dataset_df[~dataset_df.subject.isin(test_set_subjects)].series.values
+            random.seed(42)
+            random.shuffle(train_and_val_series)
+            num_val_samples = math.floor(len(train_and_val_series) / 9)
+            val_series = train_and_val_series[:num_val_samples]
+            train_series = train_and_val_series[num_val_samples:]
+
+            logging.info(f"Dropped {len(all_series) - (len(test_series) + len(train_and_val_series))} series "
+                         f"due to subject overlap with test set.")
+            return DatasetSplits.from_subject_ids(dataset_df,
+                                                  train_ids=train_series,
+                                                  val_ids=val_series,
+                                                  test_ids=test_series,
+                                                  subject_column="series",
+                                                  group_column="subject")
         else:
             return DatasetSplits.from_proportions(dataset_df,
                                                   proportion_train=0.8,