Set x_ref_preprocessed=True during legacy loading (#732)

For patch release v0.10.5

Author: Ashley Scillitoe

CHANGELOG.md

@@ -5,10 +5,9 @@
 [Full Changelog](https://github.com/SeldonIO/alibi-detect/compare/v0.10.4...v0.10.5)
 
 ### Fixed
-- Fixed a bug preventing backward compatibility when loading detectors (containing TensorFlow models) saved with `<v0.10.0`
-([#729](https://github.com/SeldonIO/alibi-detect/pull/729)). This bug also meant that detectors (containing TensorFlow models) saved with 
-`save_detector(..., legacy=True)` in `>=v0.10.0` did not properly obey the legacy file format. The `config.toml` file format used by 
-default in `>=v0.10.0` is not affected. 
+- Fixed two bugs preventing backward compatibility when loading detectors saved with `<v0.10.0`
+([#729](https://github.com/SeldonIO/alibi-detect/pull/729) and [#732](https://github.com/SeldonIO/alibi-detect/pull/732)). This bug also meant that detectors
+saved with `save_detector(..., legacy=True)` in `>=v0.10.0` did not properly obey the legacy file format. The `config.toml` file format used by default in `>=v0.10.0` is unaffected. 
 
 ## v0.10.4
 ## [v0.10.4](https://github.com/SeldonIO/alibi-detect/tree/v0.10.4) (2022-10-21)

alibi_detect/saving/tensorflow/_loading.py

@@ -233,6 +233,15 @@ def load_detector_legacy(filepath: Union[str, os.PathLike], suffix: str, **kwarg
     # load outlier detector specific parameters
     state_dict = dill.load(open(filepath.joinpath(detector_name + suffix), 'rb'))
 
+    # Update the drift detector preprocess kwargs if state_dict is from an old alibi-detect version (<v0.10).
+    # See https://github.com/SeldonIO/alibi-detect/pull/732
+    if 'kwargs' in state_dict and 'other' in state_dict:  # A drift detector if both of these exist
+        if 'x_ref_preprocessed' not in state_dict['kwargs']:  # if already exists then must have been saved w/ >=v0.10
+            # Set x_ref_preprocessed to True
+            state_dict['kwargs']['x_ref_preprocessed'] = True
+            # Move `preprocess_x_ref` from `other` to `kwargs`
+            state_dict['kwargs']['preprocess_x_ref'] = state_dict['other']['preprocess_x_ref']
+
     # initialize detector
     model_dir = filepath.joinpath('model')
     detector: Optional[Detector] = None  # to avoid mypy errors