diff --git a/crossfit/op/base.py b/crossfit/op/base.py index 14e213f..d9ea28d 100644 --- a/crossfit/op/base.py +++ b/crossfit/op/base.py @@ -70,7 +70,9 @@ def add_keep_cols(self, data, output): output[col] = data[col] columns = list(output.columns) - output = output[self.keep_cols + columns] + # we use dict.fromkeys to remove duplicates and preserve order + # (to match _build_dask_meta behavior) + output = output[list(dict.fromkeys(self.keep_cols + columns))] return output