returnn 1.20250417.191918__tar.gz → 1.20250418.115249__tar.gz

{returnn-1.20250417.191918/returnn.egg-info → returnn-1.20250418.115249}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250417.191918
+Version: 1.20250418.115249
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

returnn-1.20250418.115249/_setup_info_generated.py

@@ -0,0 +1,2 @@
+version = '1.20250418.115249'
+long_version = '1.20250418.115249+git.3cef372'

{returnn-1.20250417.191918 → returnn-1.20250418.115249}/returnn/datasets/generating.py

@@ -792,10 +792,10 @@ class DummyDataset(GeneratingDataset):
         i1 = seq_idx
         i2 = i1 + seq_len * self.num_inputs
         features = numpy.array(
-            [((i % self.input_max_value) + self.input_shift) * self.input_scale for i in range(i1, i2)]
+            [((i % self.input_max_value) + self.input_shift) * self.input_scale for i in range(i1, i2)], dtype="float32"
         ).reshape((seq_len, self.num_inputs))
         i1, i2 = i2, i2 + seq_len
-        targets = numpy.array([i % self.num_outputs["classes"][0] for i in range(i1, i2)])
+        targets = numpy.array([i % self.num_outputs["classes"][0] for i in range(i1, i2)], dtype="int32")
         return DatasetSeq(seq_idx=seq_idx, features=features, targets=targets)
 
 

{returnn-1.20250417.191918 → returnn-1.20250418.115249}/returnn/datasets/postprocessing.py

@@ -141,13 +141,14 @@ class PostprocessingDataset(CachedDataset2):
         self._map_seq_stream = map_seq_stream
         if map_seq_stream_preserves_num_seqs is None and map_seq_stream is not None:
             map_seq_stream_preserves_num_seqs = getattr(map_seq_stream, "preserves_num_seqs", None)
-        self._map_seq_stream_preserves_num_seqs = map_seq_stream_preserves_num_seqs or False
+        assert map_seq_stream_preserves_num_seqs is None or isinstance(map_seq_stream_preserves_num_seqs, bool)
+        self._map_seq_stream_preserves_num_seqs = map_seq_stream_preserves_num_seqs
         self._map_outputs = map_outputs
         self._rng = RandomState(self._get_random_seed_for_epoch(0))
         self._seq_list_for_validation: Optional[List[str]] = None
 
         self._dataset = init_dataset(self._dataset_def, parent_dataset=self)
-        if self._map_seq_stream is None or self._map_seq_stream_preserves_num_seqs:
+        if self._map_seq_stream is None or self._map_seq_stream_preserves_num_seqs is True:
             # if the stream mapper is set, the num_seqs may change and the estimation is less accurate
             self._estimated_num_seqs = self._dataset.estimated_num_seqs
         self._data_iter: Optional[Iterator[Tuple[int, TensorDict]]] = None
@@ -210,7 +211,7 @@ class PostprocessingDataset(CachedDataset2):
         self._data_iter = enumerate(self._build_mapping_iter())
         self._data_iter_produced_num_seqs = 0
         self._seq_list_for_validation = seq_list
-        if self._map_seq_stream is None or self._map_seq_stream_preserves_num_seqs:
+        if self._map_seq_stream is None or self._map_seq_stream_preserves_num_seqs is True:
             # If we don't have an iterable mapper (or the user explicitly specifies this),
             # we know the number of segments exactly equals the number of segments in the wrapped dataset
             try:

{returnn-1.20250417.191918 → returnn-1.20250418.115249/returnn.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: returnn
-Version: 1.20250417.191918
+Version: 1.20250418.115249
 Summary: The RWTH extensible training framework for universal recurrent neural networks
 Home-page: https://github.com/rwth-i6/returnn/
 Author: Albert Zeyer

{returnn-1.20250417.191918 → returnn-1.20250418.115249}/tests/test_Dataset.py

@@ -2,7 +2,7 @@
 
 from __future__ import annotations
 
-from typing import Any, Iterator, List, Dict, Optional
+from typing import Any, Iterator, TYPE_CHECKING, List, Dict, Optional
 import os
 import sys
 import _setup_test_env  # noqa
@@ -16,10 +16,14 @@ from returnn.datasets.basic import Dataset, DatasetSeq, init_dataset
 from returnn.util.basic import NumbersDict
 from returnn.util import better_exchook
 
+if TYPE_CHECKING:
+    from returnn.tensor import TensorDict
+
 
 def dummy_iter_dataset(dataset: Dataset, *, epoch: int = 1) -> List[DatasetSeq]:
     """
-    :param Dataset dataset:
+    :param dataset:
+    :param epoch:
     :return: seqs
     """
     dataset.init_seq_order(epoch=epoch)
@@ -1159,6 +1163,36 @@ def test_PostprocessingDataset():
     assert func(2) == 21
 
 
+def _post_process_map_seq_no_op(tdict: TensorDict, **_other) -> TensorDict:
+    return tdict
+
+
+def test_PostprocessingDataset_pickle():
+    from returnn.datasets.postprocessing import PostprocessingDataset
+    import pickle
+
+    ds0_opts = {"class": "DummyDataset", "input_dim": 2, "output_dim": 3, "num_seqs": 20}
+    ds0 = init_dataset(ds0_opts)
+    ds0_seqs = dummy_iter_dataset(ds0)
+
+    ds_opts = {
+        "class": "PostprocessingDataset",
+        "dataset": ds0_opts,
+        "map_seq": _post_process_map_seq_no_op,
+    }
+    ds = init_dataset(ds_opts)
+    assert isinstance(ds, PostprocessingDataset)
+    ds_seqs = dummy_iter_dataset(ds)
+    compare_dataset_seqs(ds0_seqs, ds_seqs)
+
+    s = pickle.dumps(ds)
+    ds2 = pickle.loads(s)
+    assert isinstance(ds2, PostprocessingDataset)
+    assert ds2 is not ds
+    ds2_seqs = dummy_iter_dataset(ds2)
+    compare_dataset_seqs(ds0_seqs, ds2_seqs)
+
+
 def test_MultiEpochDataset():
     from returnn.datasets.meta import MultiEpochDataset
     from returnn.datasets.cached2 import CachedDataset2

returnn-1.20250417.191918/_setup_info_generated.py

@@ -1,2 +0,0 @@
-version = '1.20250417.191918'
-long_version = '1.20250417.191918+git.4ce0802'