How to use double precision


#1

I am using cpp-package. My server runs on intel cpu. I have compiled mxnet with mkl dnn support.

By default mxnet uses float32, but i would like to perform all calculations in double precision. Couldn’t figure out how to do it. Any help appreciated.

Thank you.


#2

When creating your input NDArrays, use dtype='float64'. For example:

x = nd.random.uniform(0, 1, shape=(16, 3, 224, 224), dtype='float64')