custom_target and UMCG lesion_PIRADS (#8)

### Data Processing Enhancements:
*
[`src/dragon_prep/prepare_umcg.py`](diffhunk://#diff-317657660713de29f27d6c002d61cdce86787751e6188f62c09cf05f28e7fc59R53):
Added a new field `lesion_PIRADS` to the `umcg_entries` dictionary and
cleaned up the notation by replacing "NA" with "N/A".
[[1]](diffhunk://#diff-317657660713de29f27d6c002d61cdce86787751e6188f62c09cf05f28e7fc59R53)
[[2]](diffhunk://#diff-317657660713de29f27d6c002d61cdce86787751e6188f62c09cf05f28e7fc59R68-R70)

### Classification Target Enhancements:
*
[`src/dragon_prep/utils.py`](diffhunk://#diff-65eab55e9c448f1d168962379b7c7ca3fcb3021ffdfb45023c6bed36edd41c8dR46):
Added a new `custom_target` type to the list of classification targets
and updated the `split_and_save_data` function to include this new
target type.
[[1]](diffhunk://#diff-65eab55e9c448f1d168962379b7c7ca3fcb3021ffdfb45023c6bed36edd41c8dR46)
[[2]](diffhunk://#diff-65eab55e9c448f1d168962379b7c7ca3fcb3021ffdfb45023c6bed36edd41c8dR310)

build.sh

@@ -1 +1 @@
-docker build . --tag dragon_prep:latest --tag dragon_prep:v0.2.5
+docker build . --tag dragon_prep:latest --tag dragon_prep:v0.2.6

setup.py

@@ -5,7 +5,7 @@
         long_description = fh.read()
 
     setuptools.setup(
-        version='0.2.5',
+        version='0.2.6',
         author_email='[email protected]',
         long_description=long_description,
         long_description_content_type="text/markdown",

src/dragon_prep/prepare_umcg.py

@@ -50,6 +50,7 @@ def prepare_umcg_radiology_reports(input_dir: Path) -> pd.DataFrame:
 
         umcg_entries.append({
             "num_pirads_345": sum([score >= 3 for score in scores]),
+            "lesion_PIRADS": ",".join(study_df.pirads.values),
             **study_df.iloc[0].to_dict(),
         })
 
@@ -64,6 +65,9 @@ def prepare_umcg_radiology_reports(input_dir: Path) -> pd.DataFrame:
     # merge with radiology reports
     df["text"] = df["uid"].map(df_reports.text.to_dict())
 
+    # clean up notation
+    df["lesion_PIRADS"] = df["lesion_PIRADS"].str.replace("NA", "N/A")
+
     return df
 
 

src/dragon_prep/utils.py

@@ -43,6 +43,7 @@
     "single_label_binary_classification_target",
     "multi_label_binary_classification_target",
     "multi_label_named_entity_recognition_target",
+    "custom_target",
 ]
 
 
@@ -306,6 +307,7 @@ def split_and_save_data(
             - "multi_label_multi_class_classification_target": must be an array of strings for each case.
             - "single_label_binary_classification_target": must be an int for each case.
             - "multi_label_binary_classification_target": must be an array of ints (each either 0 or 1) for each case.
+            - "custom_target": custom target column (experimental).
         task_name (str): The name of the task. The splits will be saved in a directory with this name.
         output_dir (Path or str, optional): The output directory. Defaults to None (don't save anything).
         seed (int, optional): The random seed. Defaults to 42.

tests/development-README.md

@@ -33,4 +33,4 @@ AutoPEP8 for formatting (this can be done automatically on save, see e.g. https:
 # Push release to PyPI
 1. Increase version in setup.py, and set below
 2. Build: `python -m build`
-3. Distribute package to PyPI: `python -m twine upload dist/*0.2.5*`
+3. Distribute package to PyPI: `python -m twine upload dist/*0.2.6*`