add pytorch and python version to NN model; remove model metadata functionality (for now)

Author: osma

annif/backend/nn_ensemble.py

@@ -3,11 +3,9 @@
 
 from __future__ import annotations
 
-import importlib
-import json
 import os.path
 import shutil
-import zipfile
+import sys
 from io import BytesIO
 from typing import TYPE_CHECKING, Any
 
@@ -117,8 +115,10 @@ def save(self, filepath):
         torch.save(
             {
                 "model_state_dict": self.state_dict(),
-                "model_config": self.model_config,
                 "model_class": self.__class__.__name__,
+                "model_config": self.model_config,
+                "pytorch_version": str(torch.__version__),
+                "python_version": sys.version,
             },
             filepath,
         )
@@ -173,11 +173,9 @@ def initialize(self, parallel: bool = False) -> None:
         try:
             self._model = NNEnsembleModel.load(model_filename)
         except Exception as err:
-            metadata = self.get_model_metadata(model_filename)
             message = (
                 f"loading model from {model_filename}; "
-                f"model metadata: {metadata}; "
-                f'Original error message: "{err}"'
+                f'original error message: "{err}"'
             )
             raise OperationFailedException(message, backend_id=self.backend_id)
 
@@ -343,16 +341,3 @@ def _learn(
         self._fit_model(
             corpus, int(params["learn-epochs"]), int(params["lmdb_map_size"])
         )
-
-    def get_model_metadata(self, model_filename: str) -> dict | None:
-        """Read metadata from Keras model files."""
-
-        try:
-            with zipfile.ZipFile(model_filename, "r") as zip:
-                with zip.open("metadata.json") as metadata_file:
-                    metadata_str = metadata_file.read().decode("utf-8")
-                    metadata = json.loads(metadata_str)
-                    return metadata
-        except Exception:
-            self.warning(f"Failed to read metadata from {model_filename}")
-            return None

tests/test_backend_nn_ensemble.py

@@ -1,7 +1,5 @@
 """Unit tests for the nn_ensemble backend in Annif"""
 
-import importlib
-import os.path
 import time
 from datetime import datetime, timedelta, timezone
 from unittest import mock
@@ -193,44 +191,8 @@ def test_nn_ensemble_modification_time(app_project):
     assert datetime.now(timezone.utc) - nn_ensemble.modification_time < timedelta(1)
 
 
-@pytest.mark.skip
-def test_nn_ensemble_get_model_metadata(app_project):
-    nn_ensemble_type = annif.backend.get_backend("nn_ensemble")
-    nn_ensemble = nn_ensemble_type(
-        backend_id="nn_ensemble",
-        config_params={"sources": "dummy-en"},
-        project=app_project,
-    )
-    model_filename = os.path.join(nn_ensemble.datadir, nn_ensemble.MODEL_FILE)
-
-    expected_version = importlib.metadata.version("torch")
-    expected_date_saved = datetime.now(timezone.utc)
-    actual_metadata = nn_ensemble.get_model_metadata(model_filename)
-
-    assert actual_metadata["torch_version"] == expected_version
-    datetime_format = "%Y-%m-%d@%H:%M:%S"
-    actual_datetime = datetime.strptime(actual_metadata["date_saved"], datetime_format)
-    assert expected_date_saved - actual_datetime.astimezone(
-        tz=timezone.utc
-    ) < timedelta(1)
-
-
-def test_nn_ensemble_get_model_metadata_nonexistent_file(app_project):
-    nn_ensemble_type = annif.backend.get_backend("nn_ensemble")
-    nn_ensemble = nn_ensemble_type(
-        backend_id="nn_ensemble",
-        config_params={"sources": "dummy-en"},
-        project=app_project,
-    )
-    nonexistent_model_file = "nonexistent.zip"
-    model_filename = os.path.join(nn_ensemble.datadir, nonexistent_model_file)
-
-    actual_metadata = nn_ensemble.get_model_metadata(model_filename)
-    assert actual_metadata is None
-
-
-@mock.patch("annif.backend.nn_ensemble.load_model", side_effect=Exception)
-def test_nn_ensemble_initialize_error(load_model, app_project):
+@mock.patch("annif.backend.nn_ensemble.NNEnsembleModel.load", side_effect=Exception)
+def test_nn_ensemble_initialize_error(load, app_project):
     nn_ensemble_type = annif.backend.get_backend("nn_ensemble")
     nn_ensemble = nn_ensemble_type(
         backend_id="nn_ensemble",
@@ -240,10 +202,10 @@ def test_nn_ensemble_initialize_error(load_model, app_project):
     assert nn_ensemble._model is None
     with pytest.raises(
         OperationFailedException,
-        match=r"loading Keras model from .*; model metadata: .*",
+        match=r"loading model from .*; original error message: .*",
     ):
         nn_ensemble.initialize()
-    assert load_model.called
+    assert load.called
 
 
 def test_nn_ensemble_initialize(app_project):