mattragoza / LiGAN

Deep generative models of 3D grids for structure-based drug discovery
GNU General Public License v2.0
225 stars 44 forks source link

Changes required to train the CGAN model #66

Open dikshantsagar opened 1 year ago

dikshantsagar commented 1 year ago

Hello, I am trying to train a CGAN model. Please confirm what changes need to be made either in the config file or in the code to get this working correctly. Currently, on the train.config file, I have changed the model_type to CGAN and under disc_model, changed the parameters of the model architecture to match the generator, changed the loss weights as follows gan_loss to 1.0 steric_loss to 1.0 and rest 0.0, then finally match the disc_optim learning rate to match the gen_optim