Calculating corrected intersection losses at inference

[J-Salo]

Hi,

When performing inference (from files), the untuned losses are calculated and reported in all_intersection_losses_untuned. However, the optimized losses never seem to be calculated and reported.

In inference.py can we use something a long the lines of:

all_intersection_losses_untuned.append(
                    compute_revised_intersection_loss(lig_coords_pred_untuned.detach().cpu(), rec_graph.ndata['x'],
                                                        alpha=0.2, beta=8, aggression=0))
# After calculating coords_pred_optimized:
all_intersection_losses.append(compute_revised_intersection_loss(coords_pred_optimized, rec_graph.ndata['x'], alpha=0.2, beta=8, aggression=0))

Many thanks!

[HannesStark]

Feel free to make these changes in your repo. I do not think this is important. Let me know if you disagree!

[hmms117]

Thanks Hannes - will do :) I had hoped the loss of the optimized ligand would be a good rough proxy for fit (as an initial criteria). I guess the untuned loss will be good enough as well.