Open Oceania2018 opened 5 years ago
@Oceania2018, I was going to suggest the same thing. We don't want to be writing cuda .cu files. It's not easy. However, my first suggestion would be to parallelize the code where possible. i.e. on IEnumerables
we could use .AsParallel()
, on for loops we could use Parallel.For
and Parallel.Foreach
. BenchmarkDotNet will be helpful here.
This function is very important in some occasions where accelerated computing is required but neural networks (and torch) are not suitable.
We switched to Tensorflow.Numpy.
We might use Campy to accelerate NumSharp.