At the moment all the defaults are on Float64, to have this in line with other deep learning frameworks it should be changed to Float32