refactor(model/training): remove custom (de)serialization for FitModelReturn

Author: ruancomelli

boiling_learning/model/training.py

@@ -2,7 +2,6 @@
 
 from contextlib import contextmanager, nullcontext
 from datetime import timedelta
-from pathlib import Path
 from typing import Any, Callable, Dict, Iterator, List, Optional, Tuple, TypeVar, Union
 
 import tensorflow as tf
@@ -16,11 +15,9 @@
 from boiling_learning.describe.described import Described
 from boiling_learning.describe.describers import describe
 from boiling_learning.io import json
-from boiling_learning.io.storage import Metadata, dataclass, deserialize, load, save, serialize
+from boiling_learning.io.storage import dataclass, load
 from boiling_learning.model.callbacks import RegisterEpoch, SaveHistory
 from boiling_learning.model.model import Evaluation, ModelArchitecture
-from boiling_learning.utils.dataclasses import fields, shallow_asdict
-from boiling_learning.utils.pathutils import resolve
 from boiling_learning.utils.timing import Timer
 from boiling_learning.utils.typeutils import typename
 
@@ -91,22 +88,6 @@ class FitModelReturn:
     evaluation: Evaluation
 
 
-# NOTE: after v0.34.23, dataclasses are automatically serializable and de-serializable.
-# however, we are keeping this overload for backwards compatibility.
-@serialize.instance(FitModelReturn)
-def _serialize_fit_model_return(instance: FitModelReturn, path: Path) -> None:
-    path = resolve(path, dir=True)
-    for field_name, field in shallow_asdict(instance).items():
-        save(field, path / field_name)
-
-
-@deserialize.dispatch(FitModelReturn)
-def _deserialize_fit_model_return(path: Path, _metadata: Metadata) -> FitModelReturn:
-    return FitModelReturn(
-        **{field.name: load(path / field.name) for field in fields(FitModelReturn)}
-    )
-
-
 def get_fit_model(
     compiled_model: CompiledModel,
     datasets: Described[DatasetTriplet[tf.data.Dataset], json.JSONDataType],