Save schema in old location for backwards compatibility (#267)

* Save schema in old location for backwards compatibility

* Restore schema remove in `test_hive_partitioned_data`

---------

Co-authored-by: Karl Higley <karlb@nvidia.com>

Author: oliverholworthy

merlin/io/dataset.py

@@ -952,6 +952,8 @@ def to_parquet(
         fs.mkdirs(str(output_path), exist_ok=True)
 
         tf_metadata = TensorflowMetadata.from_merlin_schema(schema)
+        tf_metadata.to_proto_text_file(output_path)
+
         metadata_path = fs.sep.join([str(output_path), MERLIN_METADATA_DIR_NAME])
         fs.mkdirs(metadata_path, exist_ok=True)
         tf_metadata.to_json_file(metadata_path)

tests/unit/io/test_io.py

@@ -683,6 +683,9 @@ def test_hive_partitioned_data(tmpdir, cpu):
     assert result_paths
     assert all(p.endswith(".parquet") for p in result_paths)
 
+    # reading into dask dastaframe cannot have schema in same directory
+    os.remove(os.path.join(path, "schema.pbtxt"))
+
     # Read back with dask.dataframe and check the data
     df_check = dd.read_parquet(path, engine="pyarrow").compute()
     df_check["name"] = df_check["name"].astype("object")