args.split_type = 'random' args.split_sizes = (0.05, 0.95) # metric args.metric = 'mae' # run seeds args.pytorch_seeds = [0,1,2,3,4] ################################################ # names and directories args.results_dir = '/home/willlamb/results_pdts/swag_greedy2' args.save_dir = '/home/willlamb/checkpoints_pdts/swag_greedy2' args.checkpoint_path = '/home/willlamb/checkpoints_pdts/map_greedy' args.wandb_proj = 'lanterne_swag2' args.wandb_name = 'swag_greedy' args.thompson = False ### swag ### args.swag = True args.samples = 50 args.pdts = True args.pdts_batches = 30 args.epochs_init_map = 0 args.epochs = 0 args.lr_swag = 2e-5
args.data_seeds = [0, 1, 2, 3, 4] args.split_type = 'random' args.split_sizes = (0.05, 0.95) # metric args.metric = 'mae' # run seeds args.pytorch_seeds = [0, 1, 2, 3, 4] ################################################ # names and directories args.results_dir = '/home/willlamb/results_pdts/dropR_thom' args.save_dir = '/home/willlamb/checkpoints_pdts/dropR_thom' args.checkpoint_path = '/home/willlamb/checkpoints_pdts/dropR_thom' args.wandb_proj = 'lanterne_dropR' args.wandb_name = 'dropR_thom' args.thompson = True ### dropR ### args.samples = 50 args.pdts = True args.pdts_batches = 30 args.epochs_init_map = 500 args.epochs = 200 args.lr = 1e-4
args.max_data_size = 150000 args.seed = 0 # seed for data splits args.split_type = 'scaffold_balanced' args.split_sizes = (0.64, 0.16, 0.2) # metric args.metric = 'mae' ################################################ # names and directories args.save_dir = '/home/willlamb/checkpoints/bbp' args.results_dir = '/home/willlamb/results/bbp' args.wandb_proj = 'official3' args.wandb_name = 'bbp' args.checkpoint_path = '/home/willlamb/checkpoints/map' # ensembling and samples args.ensemble_size = 3 args.ensemble_start_idx = 0 args.pytorch_seeds = [0, 1, 2, 3, 4] args.samples = 30 ### bbp ### args.bbp = True args.epochs = 0 args.epochs_bbp = 100 args.batch_size_bbp = 50 args.lr_bbp = 1e-4 args.prior_sig_bbp = 0.05
args.max_data_size = 150000 args.seed = 0 # seed for data splits args.split_type = 'scaffold_balanced' args.split_sizes = (0.64, 0.16, 0.2) # metric args.metric = 'mae' ################################################ # names and directories args.save_dir = '/home/willlamb/checkpoints/dun' args.results_dir = '/home/willlamb/results/dun' args.wandb_proj = 'official_dun' args.wandb_name = 'dun' args.checkpoint_path = None # ensembling and samples args.ensemble_size = 1 args.ensemble_start_idx = 0 args.pytorch_seeds = [0, 1, 2, 3, 4] args.samples = 100 ### dun ### args.dun = True args.depth_min = 1 args.depth_max = 5 args.epochs = 0 args.epochs_dun = 350
args.bias = False # data args.max_data_size = 150000 # full data set args.seed = 0 # seed for data splits args.split_type = 'scaffold_balanced' args.split_sizes = (0.64, 0.16, 0.2) # metric args.metric = 'mae' ################################################ # names and directories #args.save_dir = '/home/willlamb/checkpoints/map' #args.results_dir = '/home/willlamb/results/map' #args.wandb_proj = 'official1' #args.wandb_name = 'map' args.method = 'map' args.checkpoint_path = '/Users/georgelamb/Documents/checkpoints/map' # SET THIS TO MAP FOR SWAG AND SGLD args.results_dir = '/Users/georgelamb/Documents/results/map' # ensembling and samples args.ensemble_size = 5 args.samples = 1 ################################################ # run new_noise(args)