Merge pull request #104 from arnor-sigurdsson/improve/prediction-hand…

…ling Improve prediction handling
arnor-sigurdsson · Dec 11, 2024 · 3023c45 · 3023c45
2 parents 255bb6e + 0d675d8
commit 3023c45
Show file tree

Hide file tree

Showing 9 changed files with 86 additions and 94 deletions.
diff --git a/docs/tutorials/h_survival_analysis/01_survival_flchain.rst b/docs/tutorials/h_survival_analysis/01_survival_flchain.rst
@@ -59,6 +59,10 @@ Let's configure and train a model on the FLChain data. Here are the key configur
     :language: yaml
     :caption: input.yaml
 
+.. literalinclude:: ../tutorial_files/h_survival_analysis/01_flchain/fusion.yaml
+    :language: yaml
+    :caption: fusion.yaml
+
 .. literalinclude:: ../tutorial_files/h_survival_analysis/01_flchain/output.yaml
     :language: yaml
     :caption: output.yaml

diff --git a/docs/tutorials/h_survival_analysis/02_survival_flchain_cox.rst b/docs/tutorials/h_survival_analysis/02_survival_flchain_cox.rst
@@ -46,6 +46,10 @@ which is also supported for survival models
     :language: yaml
     :caption: input.yaml
 
+.. literalinclude:: ../tutorial_files/h_survival_analysis/02_flchain_cox/fusion.yaml
+    :language: yaml
+    :caption: fusion.yaml
+
 .. literalinclude:: ../tutorial_files/h_survival_analysis/02_flchain_cox/output.yaml
     :language: yaml
     :caption: output.yaml

diff --git a/eir/data_load/data_preparation_modules/imputation.py b/eir/data_load/data_preparation_modules/imputation.py
@@ -299,7 +299,7 @@ def impute_missing_output_modalities(
                 case ComputedArrayOutputInfo():
                     assert output_type == "array"
                     shape = output_object.data_dimensions.full_shape()
-                    approach = "random"
+                    approach = "constant"
 
                 case ComputedTabularOutputInfo() | ComputedSurvivalOutputInfo():
                     assert output_type in ("tabular", "survival")

diff --git a/eir/predict_modules/predict_output_modules/predict_tabular_output.py b/eir/predict_modules/predict_output_modules/predict_tabular_output.py
@@ -82,6 +82,7 @@ def predict_tabular_wrapper_with_labels(
             ids=cur_ids,
             predictions=predictions_np,
             labels=target_labels_np,
+            tabular_output_type=target_head_name,
             prediction_classes=classes,
             cat_loss_name=cat_loss_name,
         )
@@ -119,6 +120,7 @@ def _merge_ids_predictions_and_labels(
     ids: Sequence[str],
     predictions: np.ndarray,
     labels: Optional[np.ndarray],
+    tabular_output_type: str,
     cat_loss_name: str,
     prediction_classes: Union[Sequence[str], None] = None,
     label_column_name: str = "True Label",
@@ -134,7 +136,7 @@ def _merge_ids_predictions_and_labels(
     if prediction_classes is None:
         prediction_classes = [f"Score Class {i}" for i in range(predictions.shape[1])]
 
-    if cat_loss_name == "BCEWithLogitsLoss":
+    if tabular_output_type == "cat" and cat_loss_name == "BCEWithLogitsLoss":
         assert predictions.shape[1] == 1, predictions.shape
         assert len(prediction_classes) == 2, len(prediction_classes)
         prediction_classes = prediction_classes[1]
@@ -184,6 +186,7 @@ def predict_tabular_wrapper_no_labels(
         df_predictions = _merge_ids_and_predictions(
             ids=cur_ids,
             predictions=predictions,
+            tabular_output_type=target_head_name,
             prediction_classes=classes,
             cat_loss_name=cat_loss_name,
         )
@@ -206,6 +209,7 @@ def _merge_ids_and_predictions(
     ids: Sequence[str],
     predictions: np.ndarray,
     cat_loss_name: str,
+    tabular_output_type: str,
     prediction_classes: Optional[Sequence[str]] = None,
 ) -> pd.DataFrame:
     df = pd.DataFrame()
@@ -216,7 +220,7 @@ def _merge_ids_and_predictions(
     if prediction_classes is None:
         prediction_classes = [f"Score Class {i}" for i in range(predictions.shape[1])]
 
-    if cat_loss_name == "BCEWithLogitsLoss":
+    if tabular_output_type == "cat" and cat_loss_name == "BCEWithLogitsLoss":
         assert predictions.shape[1] == 1, predictions.shape
         assert len(prediction_classes) == 2, len(prediction_classes)
         prediction_classes = prediction_classes[1]

diff --git a/eir/train_utils/evaluation.py b/eir/train_utils/evaluation.py
@@ -642,7 +642,14 @@ def save_classification_predictions(
     output_folder: Path,
 ) -> None:
 
-    validation_predictions_total = val_outputs.argmax(axis=1)
+    def sigmoid(x):
+        return 1 / (1 + np.exp(-x))
+
+    # BCEWithLogitsLoss case
+    if val_outputs.shape[1] == 1:
+        validation_predictions_total = sigmoid(val_outputs).round().astype(int)
+    else:
+        validation_predictions_total = val_outputs.argmax(axis=1)
 
     df_predictions = _parse_valid_classification_predictions(
         val_true=val_labels,

diff --git a/poetry.lock b/poetry.lock
diff --git a/requirements-dev.txt b/requirements-dev.txt
@@ -42,7 +42,7 @@ contourpy==1.3.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
 coverage==7.6.9 ; python_full_version >= "3.12.0" and python_version < "3.13"
 coverage[toml]==7.6.9 ; python_full_version >= "3.12.0" and python_version < "3.13"
 cycler==0.12.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
-debugpy==1.8.9 ; python_full_version >= "3.12.0" and python_version < "3.13"
+debugpy==1.8.10 ; python_full_version >= "3.12.0" and python_version < "3.13"
 decorator==5.1.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
 deeplake==4.0.3 ; python_full_version >= "3.12.0" and python_version < "3.13"
 defusedxml==0.7.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
@@ -55,7 +55,7 @@ fastapi==0.115.6 ; python_full_version >= "3.12.0" and python_version < "3.13"
 fastjsonschema==2.21.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
 filelock==3.16.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
 flake8==7.1.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
-fonttools==4.55.2 ; python_full_version >= "3.12.0" and python_version < "3.13"
+fonttools==4.55.3 ; python_full_version >= "3.12.0" and python_version < "3.13"
 formulaic==1.0.2 ; python_full_version >= "3.12.0" and python_version < "3.13"
 fqdn==1.5.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
 frozenlist==1.5.0 ; python_full_version >= "3.12.0" and python_version < "3.13"
@@ -99,7 +99,7 @@ jupyter==1.1.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
 jupyterlab-pygments==0.3.0 ; python_full_version >= "3.12.0" and python_version < "3.13"
 jupyterlab-server==2.27.3 ; python_full_version >= "3.12.0" and python_version < "3.13"
 jupyterlab-widgets==3.0.13 ; python_full_version >= "3.12.0" and python_version < "3.13"
-jupyterlab==4.3.2 ; python_full_version >= "3.12.0" and python_version < "3.13"
+jupyterlab==4.3.3 ; python_full_version >= "3.12.0" and python_version < "3.13"
 kiwisolver==1.4.7 ; python_full_version >= "3.12.0" and python_version < "3.13"
 lifelines==0.30.0 ; python_full_version >= "3.12.0" and python_version < "3.13"
 markupsafe==3.0.2 ; python_full_version >= "3.12.0" and python_version < "3.13"

diff --git a/requirements.txt b/requirements.txt
@@ -29,7 +29,7 @@ einops==0.8.0 ; python_full_version >= "3.12.0" and python_version < "3.13"
 executing==2.1.0 ; python_full_version >= "3.12.0" and python_version < "3.13"
 fastapi==0.115.6 ; python_full_version >= "3.12.0" and python_version < "3.13"
 filelock==3.16.1 ; python_full_version >= "3.12.0" and python_version < "3.13"
-fonttools==4.55.2 ; python_full_version >= "3.12.0" and python_version < "3.13"
+fonttools==4.55.3 ; python_full_version >= "3.12.0" and python_version < "3.13"
 formulaic==1.0.2 ; python_full_version >= "3.12.0" and python_version < "3.13"
 frozenlist==1.5.0 ; python_full_version >= "3.12.0" and python_version < "3.13"
 fsspec==2024.10.0 ; python_full_version >= "3.12.0" and python_version < "3.13"

diff --git a/tests/test_modelling/test_tabular_modelling/test_omics_modelling.py b/tests/test_modelling/test_tabular_modelling/test_omics_modelling.py
@@ -869,11 +869,8 @@ def _should_compile():
                         "lr": 1e-03 * 4,
                         "gradient_accumulation_steps": 4,
                     },
-                    "training_control": {
-                        "mixing_alpha": 0.2,
-                    },
                     "attribution_analysis": {
-                        "max_attributions_per_class": 100,
+                        "max_attributions_per_class": 200,
                     },
                 },
                 "input_configs": [