fix issue 3899, call self.save after update _time_fit_training, add t… (

autogluon#3900)

Co-authored-by: Oleksandr Shchur <shchuro@amazon.com>

full_install.bat

@@ -0,0 +1,9 @@
+
+python -m pip install -e common/[tests]
+python -m pip install -e core/[all,tests]
+python -m pip install -e features/
+python -m pip install -e tabular/[all,tests]
+python -m pip install -e multimodal/[tests]
+python -m pip install -e timeseries/[all,tests]
+python -m pip install -e eda/
+python -m pip install -e autogluon/

timeseries/src/autogluon/timeseries/learner.py

@@ -123,9 +123,9 @@ def _fit(
             excluded_model_types=kwargs.get("excluded_model_types"),
             time_limit=time_limit,
         )
-        self.save_trainer(trainer=self.trainer)
 
         self._time_fit_training = time.time() - time_start
+        self.save()
 
     def _align_covariates_with_forecast_index(
         self,

timeseries/tests/unittests/test_predictor.py

@@ -411,6 +411,18 @@ def test_when_fit_summary_is_called_then_all_keys_and_models_are_included(
             assert len(fit_summary[key]) == num_models
 
 
+EXPECTED_INFO_KEYS = [
+    "path",
+    "version",
+    "time_fit_training",
+    "time_limit",
+    "best_model",
+    "best_model_score_val",
+    "num_models_trained",
+    "model_info",
+]
+
+
 @pytest.mark.parametrize(
     "hyperparameters, num_models",
     [
@@ -421,23 +433,26 @@ def test_when_fit_summary_is_called_then_all_keys_and_models_are_included(
 def test_when_info_is_called_then_all_keys_and_models_are_included(temp_model_path, hyperparameters, num_models):
     predictor = TimeSeriesPredictor(path=temp_model_path)
     predictor.fit(DUMMY_TS_DATAFRAME, hyperparameters=hyperparameters)
-    expected_keys = [
-        "path",
-        "version",
-        "time_fit_training",
-        "time_limit",
-        "best_model",
-        "best_model_score_val",
-        "num_models_trained",
-        "model_info",
-    ]
     info = predictor.info()
-    for key in expected_keys:
+    for key in EXPECTED_INFO_KEYS:
         assert key in info
 
     assert len(info["model_info"]) == num_models
 
 
+def test_when_predictor_is_loaded_then_info_works(temp_model_path):
+    predictor = TimeSeriesPredictor(path=temp_model_path, prediction_length=2)
+    predictor.fit(train_data=DUMMY_TS_DATAFRAME, hyperparameters=DUMMY_HYPERPARAMETERS)
+    predictor.save()
+    del predictor
+    predictor = TimeSeriesPredictor.load(temp_model_path)
+    info = predictor.info()
+    for key in EXPECTED_INFO_KEYS:
+        assert key in info
+
+    assert len(info["model_info"]) == len(DUMMY_HYPERPARAMETERS) + 1  # + 1 for ensemble
+
+
 def test_when_train_data_contains_nans_then_predictor_can_fit(temp_model_path):
     predictor = TimeSeriesPredictor(path=temp_model_path)
     df = DUMMY_TS_DATAFRAME.copy()