Let ImbalancedSampler accept torch.Tensor as input

[EdisonLeeeee]

Hi, I modified torch_geometric.loader.ImbalancedSampler to accept torch.Tensor object, i.e., the class distribution as input. Now it is supported to use the class labels to initialize the sampler, as like:

from torch_geometric.loader import NeighborLoader, ImbalancedSampler

sampler = ImbalancedSampler(data.y, input_nodes=data.train_mask)
loader = NeighborLoader(data.y, input_nodes=data.train_mask,
                        batch_size=64, num_neighbors=[-1, -1],
                        sampler=sampler, ...)    
                                sampler=sampler, ...)  

And the issue in #5108 should be fixed with minimal adaptation:

from torch_geometric.datasets import DBLP
from torch_geometric.loader import ImbalancedSampler, NeighborLoader

dataset = DBLP(root='~/data/pygdata/dblp')
data = dataset[0]

# NOTE: input the node labels instead of the dataset `data['author']`
sampler = ImbalancedSampler(data['author'].y, input_nodes=data['author'].train_mask) 

train_loader = NeighborLoader(data, input_nodes=('author', data['author'].train_mask),
                              num_neighbors=[10, 10], 
                              sampler=sampler, batch_size=20)

[codecov]

Codecov Report

Merging #5138 (f8de132) into master (7692969) will increase coverage by 0.00%.
The diff coverage is 100.00%.

❗ Current head f8de132 differs from pull request most recent head 59099ee. Consider uploading reports for the commit 59099ee to get more accurate results

@@           Coverage Diff           @@
##           master    #5138   +/-   ##
=======================================
  Coverage   82.98%   82.98%           
=======================================
  Files         333      333           
  Lines       18368    18371    +3     
=======================================
+ Hits        15243    15246    +3     
  Misses       3125     3125           

Impacted Files	Coverage Δ
torch_geometric/loader/imbalanced_sampler.py	88.46% <100.00%> (+1.50%)	⬆️

📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more

[EdisonLeeeee]

Fix several typos

[rusty1s]

Thanks!