| 101 | |
| 102 | |
| 103 | def fit_models(args): |
| 104 | dataset = args.dataset |
| 105 | # ["spherec", "spherecplus", "wrap", "sphericalharmonics", "theory", "grid", "direct", "cartesian3d"] |
| 106 | # ["linear", "siren", "fcnet", "mlp"] |
| 107 | |
| 108 | if not isinstance(args.pe, list): |
| 109 | args.pe = list(args.pe) |
| 110 | |
| 111 | if not isinstance(args.nn, list): |
| 112 | args.nn = list(args.nn) |
| 113 | |
| 114 | for pe in args.pe: |
| 115 | for nn in args.nn: # |
| 116 | for seed in np.arange(args.num_seeds): |
| 117 | args_fit = Namespace(dataset=dataset, |
| 118 | pe=pe, |
| 119 | nn=nn, |
| 120 | save_model=True if dataset == "inat2018" else False, # inat needs checkpoints for evaluation |
| 121 | log_wandb=False, |
| 122 | hparams='hparams.yaml', |
| 123 | results_dir='results/train', |
| 124 | expname=f'exp_quantiative/{seed}', |
| 125 | legendre_polys=None, # take from hparams file |
| 126 | seed=seed, |
| 127 | harmonics_calculation="analytic", |
| 128 | resume_ckpt_from_results_dir=args.resume_ckpt_from_results_dir, |
| 129 | gpus=args.gpus, |
| 130 | use_expnamehps=False, |
| 131 | matplotlib=False, |
| 132 | matplotlib_show=False, |
| 133 | accelerator=args.accelerator, |
| 134 | checkerboard_scale=1, |
| 135 | max_epochs=None, |
| 136 | min_radius=None) # take from hparams file |
| 137 | |
| 138 | fit(args_fit) |
| 139 | return args |
| 140 | |
| 141 | if __name__ == '__main__': |
| 142 | main() |