Some additions to AnnLoader and AnnCollection (#704)

* add automatic bacthing

* release note

anndata/experimental/multi_files/_anncollection.py

@@ -466,6 +466,16 @@ def shape(self):
         """Shape of the lazily concatenated subset of the data matrix."""
         return len(self.obs_names), len(self.var_names)
 
+    @property
+    def n_obs(self):
+        """Number of observations."""
+        return self.shape[0]
+
+    @property
+    def n_vars(self):
+        """Number of variables/features."""
+        return self.shape[1]
+
     @property
     def convert(self):
         """On the fly converters for keys of attributes and data matrix.

anndata/experimental/pytorch/_annloader.py

@@ -13,10 +13,10 @@
 
 try:
     import torch
-    from torch.utils.data import Sampler, Dataset, DataLoader
+    from torch.utils.data import Sampler, BatchSampler, Dataset, DataLoader
 except ImportError:
     warnings.warn("Сould not load pytorch.")
-    Sampler, Dataset, DataLoader = object, object, object
+    Sampler, BatchSampler, Dataset, DataLoader = object, object, object, object
 
 
 # Custom sampler to get proper batches instead of joined separate indices
@@ -188,17 +188,21 @@ def __init__(
         if (
             batch_size is not None
             and batch_size > 1
-            and not has_sampler
             and not has_batch_sampler
             and not use_parallel
         ):
             drop_last = kwargs.pop("drop_last", False)
-            default_sampler = BatchIndexSampler(
-                len(dataset), batch_size, shuffle, drop_last
-            )
 
-            super().__init__(
-                dataset, batch_size=None, sampler=default_sampler, **kwargs
-            )
+            if has_sampler:
+                sampler = kwargs.pop("sampler")
+                sampler = BatchSampler(
+                    sampler, batch_size=batch_size, drop_last=drop_last
+                )
+            else:
+                sampler = BatchIndexSampler(
+                    len(dataset), batch_size, shuffle, drop_last
+                )
+
+            super().__init__(dataset, batch_size=None, sampler=sampler, **kwargs)
         else:
             super().__init__(dataset, batch_size=batch_size, shuffle=shuffle, **kwargs)

docs/release-latest.rst

@@ -25,6 +25,9 @@ This should make it much easier to support new datatypes, use partial access, an
 - In many cases :attr:`~anndata.AnnData.X` can now be `None` :pr:`463` :smaller:`R Cannoodt` :pr:`677` :smaller:`I Virshup`. Remaining work is documented in :issue:`467`.
 - Removed hard `xlrd` dependency :smaller:`I Virshup`
 - `obs` and `var` dataframes are no longer copied by default on `AnnData` instantiation :issue:`371` :smaller:`I Virshup`
+- Added PyTorch dataloader :class:`~anndata.experimental.AnnLoader` and lazy concatenation object :class:`~anndata.experimental.AnnCollection`. See the `tutorials`_ :pr:`416` :smaller:`S Rybakov`
+
+.. _tutorials: https://anndata-tutorials.readthedocs.io/en/latest/index.html
 
 .. rubric:: Bug fixes