[ENH] EXPERIMENTAL PR: make the data_module dataclass-like (#1832)

This PR makes the `data_modulel` dataclass-like
See discussion in #1829

Author: phoeenniixx

pytorch_forecasting/data/data_module.py

@@ -163,14 +163,6 @@ def __init__(
             else:
                 self.continuous_indices.append(idx)
 
-        # overwrite __init__ params for upwards compatibility with AS PRs
-        # todo: should we avoid this and ensure classes are dataclass-like?
-        self.min_prediction_length = self._min_prediction_length
-        self.min_encoder_length = self._min_encoder_length
-        self.categorical_encoders = self._categorical_encoders
-        self.scalers = self._scalers
-        self.target_normalizer = self._target_normalizer
-
     def _prepare_metadata(self):
         """Prepare metadata for model initialisation.
 

tests/test_data/test_data_module.py

@@ -86,8 +86,8 @@ def test_init(sample_timeseries_data):
 
     assert dm.max_encoder_length == 24
     assert dm.max_prediction_length == 12
-    assert dm.min_encoder_length == 24
-    assert dm.min_prediction_length == 12
+    assert dm._min_encoder_length == 24
+    assert dm._min_prediction_length == 12
     assert dm.batch_size == 8
     assert dm.train_val_test_split == (0.7, 0.15, 0.15)