Open gokceneraslan opened 1 year ago
Sorry for the late response. We initially optimized the model on DNase data (classification not regression). We "just" used it here becasue it had reasonable correlation (of course this can be improved) but basically fullfilled the task to inspect important motifs.
Thank you for releasing the code. One thing I noticed in the model code is the use of
softmax
in convlayers andsigmoid
in the hidden dense layer (https://github.com/visze/sequence_cnn_models/blob/master/workflow/scripts/model.py#L11). I was wondering what the reasoning is for these activations.