elijahcole / sinr

Line 76 of the code produces a "shape mismatch" message. This appears to caused by preds.shape being too large (499385,256) to be pasted into op_im (shape: (499385,)). Which slice of preds is the correct one to use?

Hi there.

Can you confirm that you have not modified this line:

sinr/viz_map.py

Line 65 in df676be

    
           preds = model(locs_enc, return_feats=False, class_of_interest=class_of_interest).cpu().numpy()

preds should have the same number of entries as rows in locs_enc and have a single value for each entry.

Hello, I didn't modify any of the code. After replacing the faulty libdeflate package I opened a terminal from Anaconda and ran viz_map code in the pdb debugger. Here's the output - with irrelevant lines removed. (GeoModel) D:\Data\PyCharm\sinr-main>python -m pdb viz_map.py --taxa_id=870009

d:\data\pycharm\sinr-main\viz_map.py(1)<module>()

-> """ (Pdb) n

d:\data\pycharm\sinr-main\viz_map.py(6)<module>()

-> import torch .... -> main(eval_params) (Pdb) s --Call--

d:\data\pycharm\sinr-main\viz_map.py(18)main()

-> def main(eval_params): .... Loading taxa: 870009 ...

d:\data\pycharm\sinr-main\viz_map.py(61)main()

-> locs_enc = enc.encode(locs).to(eval_params['device']) (Pdb) n

d:\data\pycharm\sinr-main\viz_map.py(64)main()

-> with torch.no_grad(): (Pdb) locs_enc.shape torch.Size([499385, 4]) (Pdb) n

d:\data\pycharm\sinr-main\viz_map.py(65)main()

-> preds = model(locs_enc, return_feats=False, class_of_interest=class_of_interest).cpu().numpy() (Pdb) n

d:\data\pycharm\sinr-main\viz_map.py(68)main()

-> if eval_params['threshold'] > 0: (Pdb) preds.shape (499385, 256) (Pdb) n

d:\data\pycharm\sinr-main\viz_map.py(74)main()

-> if not eval_params['disable_ocean_mask']: (Pdb) n

d:\data\pycharm\sinr-main\viz_map.py(75)main()

-> op_im = np.ones((mask.shape[0] * mask.shape[1])) * np.nan # set to NaN (Pdb) n

d:\data\pycharm\sinr-main\viz_map.py(76)main()

-> op_im[mask_inds] = preds (Pdb) n ValueError: shape mismatch: value array of shape (499385,256) could not be broadcast to indexing result of shape (499385,) (Pdb) q Cheers, Martin

…

On Sun, 24 Sept 2023 at 18:53, Oisin Mac Aodha ***@***.***> wrote: Hi there. Can you confirm that you have not modified this line: https://github.com/elijahcole/sinr/blob/df676beece3234382d9623530d6bfef5ee9c699a/viz_map.py#L65 preds should have the same number of entries as rows in locs_enc and have a single value for each entry. — Reply to this email directly, view it on GitHub <#9 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A5UY5LWTPOPDX2LPXZIWY7TX4BXP3ANCNFSM6AAAAAA5FCKB5U> . You are receiving this because you authored the thread.Message ID: ***@***.***>

The code here

https://github.com/macaodha/geo_prior/blob/b0c761712a0aeced56eba86dd19eda657fbe895a/geo_prior/models.py#L58

uses "torch.matmul(x, ...)" rather than the equivalent line in this package which uses "x * ...". Is that significant?

Thanks for flagging this. Yes, it was an issue with how "*" is treated in the version of torch specified in the requirements. Should be fixed with the latest commit.

ValueError in viz_map.py