Runtime Error in graph.py

[jannisborn]

During property optimization, if all molecules are invalid, the following line fails:

torchdrug/torchdrug/data/graph.py

Line 1222 in 119002a

if count.max() > 1:

with the error:

RuntimeError: max(): Expected reduction dim to be specified for input.numel() == 0. Specify the reduction dim with the 'dim' argument.

I thought I can fix this by increasing the batch size so that always at least one molecule is valid in the batch but due to the resampling this does not work either:

19:18:53   2 / 16 molecules are invalid even after 20 resampling
19:18:53   2 / 14 molecules are invalid even after 20 resampling
19:18:53   2 / 12 molecules are invalid even after 20 resampling
19:18:54   5 / 10 molecules are invalid even after 20 resampling
19:18:54   1 / 5 molecules are invalid even after 20 resampling
19:18:54   2 / 4 molecules are invalid even after 20 resampling
19:18:54   1 / 1 molecules are invalid even after 20 resampling

Then, I would like to increase the max_resample parameter but it's hardcoded, so there is no way to control it.

versions:

• torch: 1.11.0
• torchdrug: 0.1.2

Fix:
The following should be more robust but do the same job:

if len(count) > 0 and max(count) > 1:

[KiddoZhu]

Good point! We will fix it.