Fix indexing in convert_tables_to_dicts() for col_groups != None

Also updated the unittest to pass after the fix and not before.

Signed-off-by: Bartosz Grabowski <[email protected]>

Author: bartosz-grabowski

monai/data/utils.py

@@ -1500,7 +1500,7 @@ def convert_tables_to_dicts(
     if col_groups is not None:
         groups: dict[str, list] = {}
         for name, cols in col_groups.items():
-            groups[name] = df.loc[rows, cols].values
+            groups[name] = df.iloc[rows][cols].values
         # invert items of groups to every row of data
         data = [dict(d, **{k: v[i] for k, v in groups.items()}) for i, d in enumerate(data)]
 

tests/data/test_csv_dataset.py

@@ -182,19 +182,12 @@ def prepare_csv_file(data, filepath):
             # test pre-loaded DataFrame subset
             df = pd.read_csv(filepath1)
             df_subset = df.iloc[[1, 3, 4]]
-            dataset = CSVDataset(src=df_subset)
-            self.assertDictEqual(
-                {k: round(v, 4) if not isinstance(v, str) else v for k, v in dataset[2].items()},
-                {
-                    "subject_id": "s000004",
-                    "label": 9,
-                    "image": "./imgs/s000004.png",
-                    "ehr_0": 6.4275,
-                    "ehr_1": 6.2549,
-                    "ehr_2": 5.9765,
-                },
-            )
+            dataset = CSVDataset(src=df_subset, col_groups={"ehr": [f"ehr_{i}" for i in range(3)]})
             self.assertEqual(len(dataset), 3)
+            np.testing.assert_allclose(
+                [round(i, 4) for i in dataset[1]["ehr"]],
+                [3.3333, 3.2353, 3.4000],
+            )
 
             # test pre-loaded multiple DataFrames, join tables with kwargs
             dfs = [pd.read_csv(i) for i in filepaths]