Fix Pre-trained DimeNet++ performance on QM9

[xnuohz]

Fix #4698

Target: 00, MAE: 0.02975 ± 0.05869
Target: 01, MAE: 0.04322 ± 0.15740
Target: 02, MAE: 24.43286 ± 35.56775
Target: 03, MAE: 19.42164 ± 31.41084
Target: 05, MAE: 0.28941 ± 0.66366
Target: 06, MAE: 1.21997 ± 2.13234
Target: 07, MAE: 6.15220 ± 13.26018
Target: 08, MAE: 6.20371 ± 12.61268
Target: 09, MAE: 6.48553 ± 14.85413
Target: 10, MAE: 7.41481 ± 14.50145
Target: 11, MAE: 0.02268 ± 0.02604

[codecov]

Codecov Report

Merging #8239 (dfe371e) into master (cd6d1f2) will decrease coverage by 0.69%.
Report is 2 commits behind head on master.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #8239      +/-   ##
==========================================
- Coverage   88.07%   87.39%   -0.69%     
==========================================
  Files         473      473              
  Lines       28615    28617       +2     
==========================================
- Hits        25203    25010     -193     
- Misses       3412     3607     +195     

Files	Coverage Δ
torch_geometric/nn/models/dimenet.py	98.18% <100.00%> (+0.02%)	⬆️

... and 37 files with indirect coverage changes

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

[rusty1s]

Thank you. I would assume this change will break DimeNet due to the difference in their implementation?

[xnuohz]

Actually, it shouldn't be. I'll continue to fix this.

[xnuohz]

@rusty1s Checked the origin code again(See dimenet and dimenet++), and the author indeed used different angle calculations for these 2 models. Seems it's not consistent with the paper, anyway we just align with the author's code.

[rusty1s]

Yeah, can we somehow condition that on which DimeNet version we are using?