Merge pull request #5 from jdb78/test/models

Add more tests - testing multiple normalizers and settings for temporal fusion transformer

Author: jdb78

.github/workflows/code_quality.yml

@@ -30,4 +30,4 @@ jobs:
           # Enable linters
           black: true
           flake8: true
-          mypy: true
+          # mypy: true

pytorch_forecasting/data.py

@@ -1191,6 +1191,22 @@ def to_dataloader(self, train: bool = True, batch_size: int = 64, **kwargs) -> D
         Args:
             train (bool, optional): if dataloader is used for training or prediction
                 Will shuffle and drop last batch if True. Defaults to True.
+            batch_size (int): batch size for training model. Defaults to 64.
+            **kwargs: additional arguments to ``DataLoader()``
+
+
+        Examples:
+
+            To samples for training:
+
+            .. code-block:: python
+
+                from torch.utils.data import WeightedRandomSampler
+
+                # length of probabilties for sampler have to be equal to the length of the index
+                probabilities = np.sqrt(1 + data.loc[dataset.index, "target"])
+                sampler = WeightedRandomSampler(probabilities, len(probabilities))
+                dataset.to_dataloader(train=True, sampler=sampler, shuffle=False)
 
         Returns:
             DataLoader: dataloader that returns Tuple.
@@ -1208,15 +1224,16 @@ def to_dataloader(self, train: bool = True, batch_size: int = 64, **kwargs) -> D
                 Second entry is target
         )
         """
-        return DataLoader(
-            self,
+        default_kwargs = dict(
             shuffle=train,
             drop_last=train and len(self) > batch_size,
             collate_fn=self._collate_fn,
             batch_size=batch_size,
-            **kwargs,
         )
 
+        default_kwargs.update(kwargs)
+        return DataLoader(self, **default_kwargs,)
+
     def get_index(self) -> pd.DataFrame:
         """
         Data index / order in which items are returned in train=False mode by dataloader.

tests/test_models/conftest.py

@@ -2,7 +2,7 @@
 import numpy as np
 from data import get_stallion_data, generate_ar_data
 from pytorch_forecasting import TimeSeriesDataSet
-from pytorch_forecasting.data import NaNLabelEncoder, EncoderNormalizer
+from pytorch_forecasting.data import GroupNormalizer, NaNLabelEncoder, EncoderNormalizer
 
 
 @pytest.fixture
@@ -35,8 +35,42 @@ def data_with_covariates():
     return data
 
 
-@pytest.fixture
-def dataloaders_with_coveratiates(data_with_covariates):
+@pytest.fixture(
+    params=[
+        dict(),
+        dict(
+            static_categoricals=["agency", "sku"],
+            static_reals=["avg_population_2017", "avg_yearly_household_income_2017"],
+            time_varying_known_categoricals=["special_days", "month"],
+            variable_groups=dict(
+                special_days=[
+                    "easter_day",
+                    "good_friday",
+                    "new_year",
+                    "christmas",
+                    "labor_day",
+                    "independence_day",
+                    "revolution_day_memorial",
+                    "regional_games",
+                    "fifa_u_17_world_cup",
+                    "football_gold_cup",
+                    "beer_capital",
+                    "music_fest",
+                ]
+            ),
+            time_varying_known_reals=["time_idx", "price_regular", "price_actual", "discount", "discount_in_percent"],
+            time_varying_unknown_categoricals=[],
+            time_varying_unknown_reals=["volume", "log_volume", "industry_volume", "soda_volume", "avg_max_temp"],
+            constant_fill_strategy={"volume": 0},
+            dropout_categoricals=["sku"],
+        ),
+        dict(static_categoricals=["agency", "sku"]),
+        dict(target_normalizer=EncoderNormalizer(), min_encoder_length=2),
+        dict(target_normalizer=GroupNormalizer(log_scale=True)),
+        dict(target_normalizer=GroupNormalizer(groups=["agency", "sku"], coerce_positive=1.0)),
+    ]
+)
+def dataloaders_with_coveratiates(data_with_covariates, request):
     training_cutoff = "2016-09-01"
     max_encoder_length = 36
     max_prediction_length = 6
@@ -49,30 +83,7 @@ def dataloaders_with_coveratiates(data_with_covariates):
         group_ids=["agency", "sku"],
         max_encoder_length=max_encoder_length,
         max_prediction_length=max_prediction_length,
-        static_categoricals=["agency", "sku"],
-        static_reals=["avg_population_2017", "avg_yearly_household_income_2017"],
-        time_varying_known_categoricals=["special_days", "month"],
-        variable_groups=dict(
-            special_days=[
-                "easter_day",
-                "good_friday",
-                "new_year",
-                "christmas",
-                "labor_day",
-                "independence_day",
-                "revolution_day_memorial",
-                "regional_games",
-                "fifa_u_17_world_cup",
-                "football_gold_cup",
-                "beer_capital",
-                "music_fest",
-            ]
-        ),
-        time_varying_known_reals=["time_idx", "price_regular", "price_actual", "discount", "discount_in_percent"],
-        time_varying_unknown_categoricals=[],
-        time_varying_unknown_reals=["volume", "log_volume", "industry_volume", "soda_volume", "avg_max_temp"],
-        constant_fill_strategy={"volume": 0},
-        dropout_categoricals=["sku"],
+        **request.param  # fixture parametrization
     )
 
     validation = TimeSeriesDataSet.from_dataset(
@@ -85,7 +96,7 @@ def dataloaders_with_coveratiates(data_with_covariates):
     return dict(train=train_dataloader, val=val_dataloader)
 
 
-@pytest.fixture
+@pytest.fixture()
 def dataloaders_fixed_window_without_coveratiates(data_with_covariates):
     data = generate_ar_data(seasonality=10.0, timesteps=400, n_series=10)
     data["static"] = "2"

tests/test_models/test_temporal_fusion_transformer.py

@@ -1,3 +1,5 @@
+from pytorch_forecasting.data import TimeSeriesDataSet
+import pytest
 import shutil
 import pytorch_lightning as pl
 from pytorch_lightning.loggers import TensorBoardLogger
@@ -7,8 +9,6 @@
 
 
 # todo: run with multiple normalizers
-# todo: run with muliple datasets and normalizers: ...
-# todo: monotonicity
 # todo: test different parameters
 def test_integration(dataloaders_with_coveratiates, tmp_path):
     train_dataloader = dataloaders_with_coveratiates["train"]
@@ -28,7 +28,11 @@ def test_integration(dataloaders_with_coveratiates, tmp_path):
         fast_dev_run=True,
         logger=logger,
     )
-
+    # test monotone constraints automatically
+    if "discount_in_percent" in dataloaders_with_coveratiates["train"].dataset.reals:
+        monotone_constaints = {"discount_in_percent": +1}
+    else:
+        monotone_constaints = {}
     net = TemporalFusionTransformer.from_dataset(
         train_dataloader.dataset,
         learning_rate=0.15,
@@ -40,7 +44,7 @@ def test_integration(dataloaders_with_coveratiates, tmp_path):
         log_interval=5,
         log_val_interval=1,
         log_gradient_flow=True,
-        monotone_constaints={"discount_in_percent": +1},
+        monotone_constaints=monotone_constaints,
     )
     net.size()
     try:
@@ -56,7 +60,3 @@ def test_integration(dataloaders_with_coveratiates, tmp_path):
         net.predict(val_dataloader, fast_dev_run=True, return_index=True, return_decoder_lengths=True)
     finally:
         shutil.rmtree(tmp_path, ignore_errors=True)
-
-
-def test_monotinicity():
-    pass