Cast object dtype columns to strings in crosswalking

containers/crosswalking/context/main.py

@@ -104,6 +104,8 @@ def crosswalk(
 
  result = matrix.copy()
  result.obs = merged_obs
+ _fix_obs_columns_dtype(result)
+
  return result
 
 
@@ -129,6 +131,18 @@ def _set_default_match(obs: pd.DataFrame, column: str) -> None:
  obs.loc[obs[column].isna(), column] = "skos:exactMatch"
 
 
+def _fix_obs_columns_dtype(matrix: anndata.AnnData):
+ """Converts object and category columns to string to prevent errors when writing h5ad file.
+
+ Args:
+ matrix (AnnData): Matrix to update
+ """
+ for column in matrix.obs.columns:
+ array = matrix.obs[column]
+ if array.dtype in ("category", "object"):
+ matrix.obs[column] = array.astype(str)
+
+
 def _get_empty_table(args: argparse.Namespace) -> pd.DataFrame:
  """Creates an empty crosswalk table.