can the defualt paramters reproduce the paper results

[aaronlyt]

as the title, if i use the fetal dataset

[jsxlei]

Thanks for your interest. The results are obtained using the exact code in this repo except translation matching part not released, which will align the embedding on the umap since there will be a gap although the similarity is pretty good. I am working on end-to-end version. Stay tuned.

[aaronlyt]

using the default paramters except setting backed=True and make it take effect, after train 30000 steps/44 epochs, is the output below is ok? :

loss=4.29, v_num=, acc/val=0.0273, matchscore/val=0.0262, foscttm/val=0.0823, loss/val=4.610, acc/train=0.0349, matchscore/train=0.0345, foscttm/train=0.0505, loss/train=4.300

the code make backed take effected:

        dm = MixDataModule(
            data_dir=args.data_dir,
            modality=args.mod,
            batch_size=args.batch_size,
            linked=args.dist,
            split=args.split,
            n_top_genes=args.n_top_genes,
            binary=args.binary,
            use_seq=args.use_seq,
            mask=args.mask,
            backed=args.backed,
            num_workers=args.num_workers,
            eval_batch_size=args.eval_batch_size
        )