Merge pull request #3 from BiomedSciAI/remove_one_hot

Removed the change from string binary to int based

.github/workflows/python-package.yml

@@ -36,7 +36,7 @@ jobs:
  python scripts/tasks_retrival/HLA_task_creation.py --allow-downloads True
  python scripts/tasks_retrival/HPA_tasks_creation.py --allow-downloads True
  python scripts/tasks_retrival/humantfs_task_creation.py --allow-downloads True
- python scripts/tasks_retrival/Reactome_tasks_creation.py --use-local-files False
+ python scripts/tasks_retrival/Reactome_tasks_creation.py  --allow-downloads True
 
  - name: Test with pytest
  run: |

.gitignore

@@ -160,3 +160,5 @@ cython_debug/
 # and can be added to the global gitignore or merged into this file. For a more nuclear
 # option (not recommended) you can uncomment the following to ignore the entire idea folder.
 #.idea/
+.vscode/settings.json
+.DS_Store

gene_benchmark/tasks.py

@@ -16,25 +16,6 @@
 )
 
 
-def is_binary_outcomes(outcomes: pd.Series | pd.DataFrame):
- """
- Checks if a vector represents a binary prediction task.
-
- Args:
- ----
- outcomes (pd.series): a series containing the labels for prediction
-
- Returns:
- -------
- bool: True if the series represents binary classification
-
- """
- if isinstance(outcomes, pd.Series):
- return outcomes.nunique() == 2
- else:
- return False
-
-
 def convert_to_mat(data: pd.Series | pd.DataFrame):
  """
  Convert a 1d series or df with np arrays as values to a 2D/3D np array.
@@ -238,11 +219,7 @@ def run(self, error_score=np.nan):
  descriptions_df = self._create_encoding()
  encodings_df = self.encoder.encode(descriptions_df)
  encodings = self._post_processing_mat(encodings_df)
-
- if is_binary_outcomes(self.task_definitions.outcomes):
- outcomes = pd.get_dummies(self.task_definitions.outcomes).iloc[:, 0]
- else:
- outcomes = self.task_definitions.outcomes
+ outcomes = self.task_definitions.outcomes
 
  cs_val = cross_validate(
  self.base_prediction_model,
@@ -272,7 +249,12 @@ def summary(self):
  summary_dict["sub_task"] = self.task_definitions.sub_task
  summary_dict["base_prediction_model"] = str(self.base_prediction_model)
  summary_dict["sample_size"] = self.task_definitions.outcomes.shape[0]
- if is_binary_outcomes(self.task_definitions.outcomes):
+ is_bin = (
+ self.task_definitions.outcomes.nunique() == 2
+ if isinstance(self.task_definitions.outcomes, pd.Series)
+ else False
+ )
+ if is_bin:
  summary_dict["class_sizes"] = ",".join(
  [str(v) for v in self.task_definitions.outcomes.value_counts().values]
  )

gene_benchmark/tests/test_tasks.py

@@ -21,7 +21,6 @@
  convert_to_mat,
  filter_exclusion,
  get_tasks_definition_names,
- is_binary_outcomes,
  load_task_definition,
  sub_sample_task_frames,
 )
@@ -352,16 +351,6 @@ def test_task_modification_3D_to_2D_concat(self):
  )
  assert full_entity_task._post_processing_mat(threeDmat).shape == (3, 6)
 
- def test_is_binary_outcomes(self):
- assert is_binary_outcomes(pd.Series(["Blue", "Blue", "Blue", "Green", "Green"]))
- assert is_binary_outcomes(pd.Series([0, 0, 1, 1, 0, 1]))
- assert is_binary_outcomes(pd.Series([True, False, True, False]))
- assert not is_binary_outcomes(
- pd.Series(["Blue", "Blue", "Blue", "Green", "Green", "yellow"])
- )
- assert not is_binary_outcomes(pd.Series([0, 0, 1, 1, 0, 1, 2, 2]))
- assert not is_binary_outcomes(pd.Series([True, False, True, False, 5]))
-
  def test_load_multilabel(self):
  task_name = "Protein class"
  mpnet_name = "sentence-transformers/all-mpnet-base-v2"