Add missing test labels in HGBDataset

[EdisonLeeeee]

Hi, I found a bug when using torch_geometric.dataset.HGBDataset. Here are my codes to reproduce the bug:

import torch
from torch_geometric.datasets import HGBDataset
from collections import Counter

datasets = ['acm', 'dblp', 'freebase', 'imdb']
labeled_types = ['paper', 'author', 'book', 'movie']

for name, types in zip(datasets, labeled_types):
    data = HGBDataset("~/data/pygdata/HGB", name)[0]
    if name != 'imdb':
        print(name, Counter(data[types].y[data[types].test_mask].tolist()))
    else:
        print(name, torch.any(data[types].y[data[types].test_mask]))
    

And the outputs are:

acm Counter({-1: 1059})
dblp Counter({-1: 1420})
freebase Counter({-1: 2784})
imdb tensor(False)

This indicates that all test labels are missing on four datasets.

I checked the code in process and found that it seems to have forgotten to assign values to the test labels. After adding these codes, the bug has been fixed. The output becomes:

acm Counter({2: 364, 0: 355, 1: 340})
dblp Counter({2: 404, 0: 404, 3: 354, 1: 258})
freebase Counter({1: 1372, 2: 591, 4: 347, 0: 146, 6: 141, 5: 127, 3: 60})
imdb tensor(True)

[codecov]

Codecov Report

Merging #5233 (591ba57) into master (cd72c22) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #5233   +/-   ##
=======================================
  Coverage   83.13%   83.13%           
=======================================
  Files         336      336           
  Lines       18523    18523           
=======================================
  Hits        15399    15399           
  Misses       3124     3124           

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[rusty1s]

Thank you!