The size of tensor does not match

[gefeiZ]

HI Yang,
I keep getting this errors reported when using the sample dataset to run step2.

inputs = inputs * torch.exp(seqcons)
RuntimeError: The size of tensor a (24) must match the size of tensor b (98304) at non-singleton dimension 2

I check the data['seq'].shape, data['seqcons'].shape, data['signal'].shape
It shows
torch.Size([2, 4, 24]) torch.Size([2, 1, 98304]) torch.Size([2, 1, 768])
I'm just getting into deep learning, so I just follow your instruction steps and I don't really understand what's going on.
Could you please help with this issue? Thanks a lot.

[yal054]

Hi, thank you for your interest!
Could you share some details about your input and output dimensions?

[gefeiZ]

Hi, thank you for your interest! Could you share some details about your input and output dimensions?

Hi Yang,Thank you for your reply.
I realized that the problem may not with the training model, I checked the h5 file generated by generate_datasets.py file.

h5_path = '~/epiformer/datasets/human.phastCons.0.h5' dataset = HDF5Dataset(h5_path, preprocess=True) data_loader = DataLoader(dataset, batch_size=1, shuffle=True) for batch_idx, data in enumerate(data_loader, 0): print(data['seq'].shape) print(data['seqcons'].shape) print(data['signal'].shape)

The outputs is:
torch.Size([1, 4, 24]) #Should be (batch_size, 4, 98304)
torch.Size([1, 1, 98304])
torch.Size([1, 4, 768])
So I'm wondering if the problem is in the step of generating the h5 file.

[gefeiZ]

Hi Yang,
I think the possible cause of this error may be due to this line in the generate_datasets.py

                '''get sequence'''
                index = peak_dict[ ':'.join([line[0],line[1],line[2]]) ]
                seq = seqfile[index].rstrip().split(None, 1)[0]

to

                '''get sequence'''
                index = peak_dict[ ':'.join([line[0],line[1],line[2]]) ]
               #
                seq = seqfile[index].rstrip().split(None, 1)[1] 

If possible, please tell me if this inference is reasonable. Thanks a lot.