1

I have trained a model previously in CNTK. I used Brainscript to do it. I have not used Python with CNTK ever (yet). Now I see in release 2.3 the interesting NetOpt that I would like to try to use in order to reduce the size of my model. It looks like to use NetOpt, one is required to use Python -- is that correct? If so (or even if not) could I load in my existing model, use NetOpt to reduce its size, and then run it as I have been (eval with C++) ?

Tullhead
  • 565
  • 2
  • 7
  • 17

1 Answers1

0

Netopt functionality only exists in Python. Netopt basically wraps a bunch of low level APIs in an easy to use way.

If you are using C++ to evaluate I'm guessing you are already using CNTK 2.x from C++. Which means that when you load your Brainscript model you are already using the same cntk bits that netopt will use.

The only difference is that after you use netopt, you can only save the model in V2 format, meaning you won't be able to load it in brainscript. Evaluation from C++ should work just fine though.

Nikos Karampatziakis
  • 2,050
  • 9
  • 15