[timeseries] Fix loading of Tabular models failing if predictor moved…

… to a different directory (#4171)

timeseries/src/autogluon/timeseries/models/autogluon_tabular/mlforecast.py

@@ -88,6 +88,27 @@ def __init__(
         self._train_target_median: Optional[float] = None
         self._non_boolean_real_covariates: List[str] = []
 
+    @property
+    def tabular_predictor_path(self) -> str:
+        return os.path.join(self.path, "tabular_predictor")
+
+    def save(self, path: str = None, verbose: bool = True) -> str:
+        assert "mean" in self._mlf.models_, "TabularPredictor must be trained before saving"
+        tabular_predictor = self._mlf.models_["mean"].predictor
+        self._mlf.models_["mean"].predictor = None
+        save_path = super().save(path=path, verbose=verbose)
+        self._mlf.models_["mean"].predictor = tabular_predictor
+        return save_path
+
+    @classmethod
+    def load(
+        cls, path: str, reset_paths: bool = True, load_oof: bool = False, verbose: bool = True
+    ) -> "AbstractTimeSeriesModel":
+        model = super().load(path=path, reset_paths=reset_paths, load_oof=load_oof, verbose=verbose)
+        assert "mean" in model._mlf.models_, "Loaded model doesn't have a trained TabularPredictor"
+        model._mlf.models_["mean"].predictor = TabularPredictor.load(model.tabular_predictor_path)
+        return model
+
     def preprocess(self, data: TimeSeriesDataFrame, is_train: bool = False, **kwargs) -> Any:
         if is_train:
             # All-NaN series are removed; partially-NaN series in train_data are handled inside _generate_train_val_dfs
@@ -295,7 +316,7 @@ def _fit(
 
         estimator = TabularEstimator(
             predictor_init_kwargs={
-                "path": os.path.join(self.path, "tabular_predictor"),
+                "path": self.tabular_predictor_path,
                 "verbosity": verbosity - 2,
                 "label": MLF_TARGET,
                 **self._get_extra_tabular_init_kwargs(),

timeseries/tests/unittests/models/test_mlforecast.py

@@ -1,3 +1,6 @@
+import os
+import shutil
+import tempfile
 from unittest import mock
 
 import numpy as np
@@ -252,3 +255,25 @@ def test_given_train_data_has_nans_when_fit_called_then_nan_rows_removed_from_tr
     model.fit(train_data=data)
     train_df, val_df = model._generate_train_val_dfs(model.preprocess(data, is_train=True))
     assert len(train_df) + len(val_df) == len(data.dropna())
+
+
+@pytest.mark.parametrize("model_type", TESTABLE_MODELS)
+@pytest.mark.parametrize("eval_metric", ["WAPE", "WQL"])
+def test_when_trained_model_moved_to_different_folder_then_loaded_model_can_predict(model_type, eval_metric):
+    data = DUMMY_TS_DATAFRAME.copy().sort_index()
+    old_model_dir = tempfile.mkdtemp()
+    model = model_type(
+        path=old_model_dir,
+        freq=data.freq,
+        eval_metric=eval_metric,
+        quantile_levels=[0.1, 0.5, 0.9],
+        prediction_length=3,
+        hyperparameters={"differences": []},
+    )
+    model.fit(train_data=data)
+    model.save()
+    new_model_dir = tempfile.mkdtemp()
+    shutil.move(model.path, new_model_dir)
+    loaded_model = model_type.load(os.path.join(new_model_dir, model.name))
+    predictions = loaded_model.predict(data)
+    assert isinstance(predictions, TimeSeriesDataFrame)