Avoid modifying mode_kwargs in MultiAggregation

CHANGELOG.md

@@ -34,6 +34,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Added `BaseStorage.get()` functionality ([#5240](https://github.com/pyg-team/pytorch_geometric/pull/5240))
 - Added a test to confirm that `to_hetero` works with `SparseTensor` ([#5222](https://github.com/pyg-team/pytorch_geometric/pull/5222))
 ### Changed
+- Avoid modifying `mode_kwargs` in `MultiAggregation` ([#5601](https://github.com/pyg-team/pytorch_geometric/pull/5601))
 - Changed `BatchNorm` to allow for batches of size one during training ([#5530](https://github.com/pyg-team/pytorch_geometric/pull/5530))
 - Integrated better temporal sampling support by requiring that local neighborhoods are sorted according to time ([#5516](https://github.com/pyg-team/pytorch_geometric/issues/5516))
 - Fixed a bug when applying several scalers with `PNAConv` ([#5514](https://github.com/pyg-team/pytorch_geometric/issues/5514))

torch_geometric/nn/aggr/multi.py

@@ -1,3 +1,4 @@
+import copy
 from typing import Any, Dict, List, Optional, Union
 
 import torch
@@ -68,7 +69,7 @@ def __init__(
         ])
 
         self.mode = mode
-        mode_kwargs = mode_kwargs or {}
+        mode_kwargs = copy.copy(mode_kwargs or {})
         self.in_channels = mode_kwargs.pop('in_channels', None)
         self.out_channels = mode_kwargs.pop('out_channels', None)
         if mode == 'proj' or mode == 'attn':