Hi,
It is very nice work!! There is a minor question about data preprocessing of the Hi-C matrix:
In the paper it is said that the data were normalized, and your previous repo (Theano version) mentioned it was KRnormalization. I downloaded the raw data and normalization vector from the original paper (Rao et al., 2014), and found that some values of the KR vectors are NAN. The missing values resulted from the sparsity of the matrix, as Rao et al. described.
I am curious how did you handle the NAN values in the KRnormalization vector? Did you throw out the positions with NAN?
Thank you very much!