Commit History
-
Best model using 1kbp seqs with K562 data
Verified signatureDownsampling was unbiased, loss function was MSE.f4d6fa4 -
Best model using DeepSTARR layout with K562 data
Verified signatureDownsampling was unbiased, loss function was MSE.3491596 -
Best model using transformer-encoder layers with K562 data
Verified signatureDownsampling was unbiased, loss function was MSE.b890c37 -
Best model for K562 data downsampled to 1M
Verified signatureDownsampling was unbiased, loss function was MSE.4398a0e -
Best model for K562 data downsampled to 1.5M
Verified signatureDownsampling was unbiased, loss function was MSE.893458d -
Best model for K562 data downsampled to 1.3M
Verified signatureDownsampling was unbiased, loss function was MSE.d7221ce -
Best model for K562 data downsampled to 0.8M
Verified signatureDownsampling was unbiased, loss function was MSE.ede6926 -
Best model for K562 data, using custom loss
Verified signatureThe custom loss function is uses negative log likelihood for a negative binomial.c0edf56 -
Best model for K562 data from biased downsampling, and using MSE loss
Verified signaturebdd408f -
Best model for K562 data and MSE loss
Verified signaturec0e4bd6
