Let ImbalancedSampler accept torch.Tensor as input (pyg-team#5138)

* Fix ImbalancedSampler to support torch.Tensor as input

* Update test

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* typo

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* update

* changelog

* update

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

Author: 3 people

CHANGELOG.md

@@ -5,6 +5,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ## [2.0.5] - 2022-MM-DD
 ### Added
+- Let `ImbalancedSampler` accept `torch.Tensor` as input ([#5138](https://github.com/pyg-team/pytorch_geometric/pull/5138))
 - Added `flow` argument to `gcn_norm` to correctly normalize the adjacency matrix in `GCNConv` ([#5149](https://github.com/pyg-team/pytorch_geometric/pull/5149))
 - `NeighborSampler` supports graphs without edges ([#5072](https://github.com/pyg-team/pytorch_geometric/pull/5072))
 - Added the `MeanSubtractionNorm` layer ([#5068](https://github.com/pyg-team/pytorch_geometric/pull/5068))

test/loader/test_imbalanced_sampler.py

@@ -1,7 +1,6 @@
 from typing import List
 
 import torch
-from torch import Tensor
 
 from torch_geometric.data import Data
 from torch_geometric.loader import (
@@ -22,16 +21,21 @@ def test_dataloader_with_imbalanced_sampler():
     sampler = ImbalancedSampler(data_list)
     loader = DataLoader(data_list, batch_size=10, sampler=sampler)
 
-    ys: List[Tensor] = []
-    for batch in loader:
-        ys.append(batch.y)
+    y = torch.cat([batch.y for batch in loader])
 
-    histogram = torch.cat(ys).bincount()
+    histogram = y.bincount()
     prob = histogram / histogram.sum()
 
     assert histogram.sum() == len(data_list)
     assert prob.min() > 0.4 and prob.max() < 0.6
 
+    # Test with label tensor as input:
+    torch.manual_seed(12345)
+    sampler = ImbalancedSampler(torch.tensor([data.y for data in data_list]))
+    loader = DataLoader(data_list, batch_size=10, sampler=sampler)
+
+    assert torch.allclose(y, torch.cat([batch.y for batch in loader]))
+
 
 def test_neighbor_loader_with_imbalanced_sampler():
     zeros = torch.zeros(10, dtype=torch.long)
@@ -46,12 +50,18 @@ def test_neighbor_loader_with_imbalanced_sampler():
     loader = NeighborLoader(data, batch_size=10, sampler=sampler,
                             num_neighbors=[-1])
 
-    ys: List[Tensor] = []
-    for batch in loader:
-        ys.append(batch.y)
+    y = torch.cat([batch.y for batch in loader])
 
-    histogram = torch.cat(ys).bincount()
+    histogram = y.bincount()
     prob = histogram / histogram.sum()
 
     assert histogram.sum() == data.num_nodes
     assert prob.min() > 0.4 and prob.max() < 0.6
+
+    # Test with label tensor as input:
+    torch.manual_seed(12345)
+    sampler = ImbalancedSampler(data.y)
+    loader = NeighborLoader(data, batch_size=10, sampler=sampler,
+                            num_neighbors=[-1])
+
+    assert torch.allclose(y, torch.cat([batch.y for batch in loader]))

torch_geometric/loader/imbalanced_sampler.py

@@ -33,10 +33,23 @@ class distribution.
                                 batch_size=64, num_neighbors=[-1, -1],
                                 sampler=sampler, ...)
 
+    You can also pass in the class labels directly as a :class:`torch.Tensor`:
+
+    .. code-block:: python
+
+        from torch_geometric.loader import NeighborLoader, ImbalancedSampler
+
+        sampler = ImbalancedSampler(data.y)
+        loader = NeighborLoader(data, input_nodes=data.train_mask,
+                                batch_size=64, num_neighbors=[-1, -1],
+                                sampler=sampler, ...)
+
     Args:
-        dataset (Dataset or Data): The dataset from which to sample the data,
-            either given as a :class:`~torch_geometric.data.Dataset` or
-            :class:`~torch_geometric.data.Data` object.
+        dataset (Dataset or Data or Tensor): The dataset or class distribution
+            from which to sample the data, given either as a
+            :class:`~torch_geometric.data.Dataset`,
+            :class:`~torch_geometric.data.Data`, or :class:`torch.Tensor`
+            object.
         input_nodes (Tensor, optional): The indices of nodes that are used by
             the corresponding loader, *e.g.*, by
             :class:`~torch_geometric.loader.NeighborLoader`.
@@ -50,7 +63,7 @@ class distribution.
     """
     def __init__(
         self,
-        dataset: Union[Data, Dataset, List[Data]],
+        dataset: Union[Dataset, Data, List[Data], Tensor],
         input_nodes: Optional[Tensor] = None,
         num_samples: Optional[int] = None,
     ):
@@ -60,6 +73,10 @@ def __init__(
             assert dataset.num_nodes == y.numel()
             y = y[input_nodes] if input_nodes is not None else y
 
+        elif isinstance(dataset, Tensor):
+            y = dataset.view(-1)
+            y = y[input_nodes] if input_nodes is not None else y
+
         elif isinstance(dataset, InMemoryDataset):
             y = dataset.data.y.view(-1)
             assert len(dataset) == y.numel()