Backports v0.14.3 (#3077)

* Fix Rotbaum serialization and deserialization (#3068)

* Fix Rotbaum to handle short series (#3073)

---------

Co-authored-by: Anurag Pant <anuragpant@cs.ucla.edu>

src/gluonts/ext/rotbaum/_model.py

@@ -20,7 +20,7 @@
 import gc
 from collections import defaultdict
 
-from gluonts.core.component import validated
+from gluonts.core.component import equals, validated
 
 
 class QRF:
@@ -121,6 +121,13 @@ def _create_xgboost_model(model_params: Optional[dict] = None):
             }
         return xgboost.sklearn.XGBModel(**model_params)
 
+    def __eq__(self, that):
+        """
+        Two QRX instances are considered equal if they have the same
+        constructor arguments.
+        """
+        return equals(self, that)
+
     def fit(
         self,
         x_train: Union[pd.DataFrame, List],

src/gluonts/ext/rotbaum/_predictor.py

@@ -13,12 +13,14 @@
 
 import concurrent.futures
 import logging
+import pickle
 from itertools import chain
 from typing import Iterator, List, Optional, Any, Dict
 from toolz import first
 
 import numpy as np
 import pandas as pd
+from pathlib import Path
 from itertools import compress
 
 from gluonts.core.component import validated
@@ -340,6 +342,31 @@ def predict(  # type: ignore
                 item_id=ts.get("item_id"),
             )
 
+    def serialize(self, path: Path) -> None:
+        """
+        This function calls parent class serialize() in order to serialize
+        the class name, version information and constuctor arguments. It
+        persists the tree predictor by pickling the model list that is
+        generated when pickling the TreePredictor.
+        """
+        super().serialize(path)
+        with (path / "predictor.pkl").open("wb") as f:
+            pickle.dump(self.model_list, f)
+
+    @classmethod
+    def deserialize(cls, path: Path, **kwargs: Any) -> "TreePredictor":
+        """
+        This function loads and returns the serialized model. It loads
+        the predictor class with the serialized arguments. It then loads
+        the trained model list by reading the pickle file.
+        """
+
+        predictor = super().deserialize(path)
+        assert isinstance(predictor, cls)
+        with (path / "predictor.pkl").open("rb") as f:
+            predictor.model_list = pickle.load(f)
+        return predictor
+
     def explain(
         self, importance_type: str = "gain", percentage: bool = True
     ) -> ExplanationResult:

src/gluonts/ext/rotbaum/_preprocess.py

@@ -452,9 +452,12 @@ def make_features(self, time_series: Dict, starting_index: int) -> List:
         end_index = starting_index + self.context_window_size
         if starting_index < 0:
             prefix = [None] * abs(starting_index)
+            time_series_window = time_series["target"]
         else:
             prefix = []
-        time_series_window = time_series["target"][starting_index:end_index]
+            time_series_window = time_series["target"][
+                starting_index:end_index
+            ]
         only_lag_features, transform_dict = self._pre_transform(
             time_series_window, self.subtract_mean, self.count_nans
         )
@@ -464,7 +467,10 @@ def make_features(self, time_series: Dict, starting_index: int) -> List:
             if self.use_feat_static_real
             else []
         )
-        if self.cardinality:
+        if (
+            self.cardinality
+            and time_series.get("feat_static_cat", None) is not None
+        ):
             feat_static_cat = (
                 self.encode_one_hot_all(time_series["feat_static_cat"])
                 if self.one_hot_encode
@@ -477,10 +483,10 @@ def make_features(self, time_series: Dict, starting_index: int) -> List:
             list(
                 chain(
                     *[
-                        list(ent[0]) + list(ent[1].values())
+                        prefix + list(ent[0]) + list(ent[1].values())
                         for ent in [
                             self._pre_transform(
-                                ts[starting_index:end_index],
+                                ts if prefix else ts[starting_index:end_index],
                                 self.subtract_mean,
                                 self.count_nans,
                             )

test/ext/rotbaum/test_model.py

@@ -11,10 +11,11 @@
 # express or implied. See the License for the specific language governing
 # permissions and limitations under the License.
 
-
+from pathlib import Path
 import pytest
+import tempfile
 
-from gluonts.ext.rotbaum import TreeEstimator
+from gluonts.ext.rotbaum import TreeEstimator, TreePredictor
 
 
 @pytest.fixture()
@@ -33,5 +34,20 @@ def test_accuracy(accuracy_test, hyperparameters, quantiles):
     accuracy_test(TreeEstimator, hyperparameters, accuracy=0.20)
 
 
-def test_serialize(serialize_test, hyperparameters):
-    serialize_test(TreeEstimator, hyperparameters)
+def test_serialize(serialize_test, hyperparameters, dsinfo):
+    forecaster = TreeEstimator.from_hyperparameters(
+        freq=dsinfo.freq,
+        **{
+            "prediction_length": dsinfo.prediction_length,
+            "num_parallel_samples": dsinfo.num_parallel_samples,
+        },
+        **hyperparameters,
+    )
+
+    predictor_act = forecaster.train(dsinfo.train_ds)
+
+    with tempfile.TemporaryDirectory() as temp_dir:
+        predictor_act.serialize(Path(temp_dir))
+        predictor_exp = TreePredictor.deserialize(Path(temp_dir))
+        assert predictor_act == predictor_exp
+        assert predictor_act.model_list == predictor_exp.model_list

test/ext/rotbaum/test_rotbaum_smoke.py

@@ -12,10 +12,12 @@
 # permissions and limitations under the License.
 
 import pytest
+import numpy as np
 
-from gluonts.ext.rotbaum import TreeEstimator
+from gluonts.ext.rotbaum import TreeEstimator, TreePredictor
 
 from gluonts.testutil.dummy_datasets import make_dummy_datasets_with_features
+from gluonts.dataset.common import ListDataset
 
 # TODO: Add support for categorical and dynamic features.
 
@@ -59,3 +61,68 @@ def test_rotbaum_smoke(datasets):
     predictor = estimator.train(dataset_train)
     forecasts = list(predictor.predict(dataset_test))
     assert len(forecasts) == len(dataset_test)
+
+
+def test_short_history_item_pred():
+    prediction_length = 7
+    freq = "D"
+
+    dataset = ListDataset(
+        data_iter=[
+            {
+                "start": "2017-10-11",
+                "item_id": "item_1",
+                "target": np.array(
+                    [
+                        1.0,
+                        9.0,
+                        2.0,
+                        0.0,
+                        0.0,
+                        1.0,
+                        5.0,
+                        3.0,
+                        4.0,
+                        2.0,
+                        0.0,
+                        0.0,
+                        1.0,
+                        6.0,
+                    ]
+                ),
+                "feat_static_cat": np.array([0.0, 0.0], dtype=float),
+                "past_feat_dynamic_real": np.array(
+                    [
+                        [1.0222e06 for i in range(14)],
+                        [750.0 for i in range(14)],
+                    ]
+                ),
+            },
+            {
+                "start": "2017-10-11",
+                "item_id": "item_2",
+                "target": np.array([7.0, 0.0, 0.0, 23.0, 13.0]),
+                "feat_static_cat": np.array([0.0, 1.0], dtype=float),
+                "past_feat_dynamic_real": np.array(
+                    [[0 for i in range(5)], [750.0 for i in range(5)]]
+                ),
+            },
+        ],
+        freq=freq,
+    )
+
+    predictor = TreePredictor(
+        freq=freq,
+        prediction_length=prediction_length,
+        quantiles=[0.1, 0.5, 0.9],
+        max_n_datapts=50000,
+        method="QuantileRegression",
+        use_past_feat_dynamic_real=True,
+        use_feat_dynamic_real=False,
+        use_feat_dynamic_cat=False,
+        use_feat_static_real=False,
+        cardinality="auto",
+    )
+    predictor = predictor.train(dataset)
+    forecasts = list(predictor.predict(dataset))
+    assert forecasts[1].quantile(0.5).shape[0] == prediction_length