elif args.optimizer =='Adam': model = CoxPH(net, optimizer=tt.optim.Adam(lr=args.lr, weight_decay=args.weight_decay),device=device) wandb.init(project='icml_new_'+args.dataset, group=f'fold{fold}_'+args.loss+args.optimizer, name=f'L{args.num_layers}N{args.num_nodes}D{args.dropout}W{args.weight_decay}B{args.batch_size}', config=args) wandb.watch(net) # Loss configuration ============================================================ patience=5 if args.loss =='rank': model.loss = DSAFTRankLoss(alpha=args.alpha, beta=args.beta) elif args.loss == 'mae': model.loss = DSAFTMAELoss() elif args.loss == 'rmse': model.loss = DSAFTRMSELoss() elif args.loss =='kspl': model.loss = DSAFTNKSPLLoss(args.an, args.sigma) elif args.loss =='kspl_new': model.loss = DSAFTNKSPLLossNew(args.an, args.sigma) # Training ====================================================================== batch_size = args.batch_size lrfinder = model.lr_finder(x_train, y_train, batch_size, tolerance=10) best = lrfinder.get_best_lr() model.optimizer.set_lr(best)
device=device) wandb.init( project=args.dataset, group=args.loss + '_' + args.optimizer, name= f'L{args.num_layers}N{args.num_nodes}D{args.dropout}W{args.weight_decay}B{args.batch_size}', config=args) wandb.watch(net) # Loss configuration ============================================================ patience = 10 if args.loss == 'rank': model.loss = DSAFTRankLoss() elif args.loss == 'mae': model.loss = DSAFTMAELoss() elif args.loss == 'rmse': model.loss = DSAFTRMSELoss() elif args.loss == 'kspl': model.loss = DSAFTNKSPLLoss(args.an, args.sigma) elif args.loss == 'kspl_new': model.loss = DSAFTNKSPLLossNew(args.an, args.sigma) # Training ====================================================================== batch_size = args.batch_size lrfinder = model.lr_finder(x_train, y_train, batch_size, tolerance=10) best = lrfinder.get_best_lr() # model.optimizer.set_lr(args.lr)