Extra inputs
AtacWorks can now take as input any number of additional bigWig files ('layers') aside from the noisy ATAC-seq data. All of these will be concatenated as additional channels in the model input ('input'). The number of input tracks is supplied as an argument ('in_channels') to main.py.
h5 files now contain multiple datasets ('input' for inputs, 'label_reg' and 'label_cla' for labels) instead of combining everything into one dataset called 'data'.
Closes #13 .
Always using --nolabel for inference
I deleted some lines in DatasetInfer (dataset.py) because I don't understand their function. As far as I can see we never need DatasetInfer to supply labels - only the input.
Deleted unused DatasetEval.
run.sh included two commands to encode inference data, with and without labels. Since we never need labels to be encoded in the .h5 file for inference, I removed the command for encoding with labels.
Extra inputs AtacWorks can now take as input any number of additional bigWig files ('layers') aside from the noisy ATAC-seq data. All of these will be concatenated as additional channels in the model input ('input'). The number of input tracks is supplied as an argument ('in_channels') to
main.py
.h5 files now contain multiple datasets ('input' for inputs, 'label_reg' and 'label_cla' for labels) instead of combining everything into one dataset called 'data'.
Closes #13 .
Always using --nolabel for inference
I deleted some lines in DatasetInfer (dataset.py) because I don't understand their function. As far as I can see we never need DatasetInfer to supply labels - only the input.
Deleted unused DatasetEval.
run.sh included two commands to encode inference data, with and without labels. Since we never need labels to be encoded in the .h5 file for inference, I removed the command for encoding with labels.