News for hackers. Stuff that matters.

A comment on Raspberry Turk.

Re:Raspberry Turk by amelius on Saturday April 01, 2017 @03:29PM (#14013696)
Very nice documentation.

I notice that your confusion matrix for piece classification contains about 10-20% error still. Is that a problem in practice? And any ideas how to improve upon it?

[ Parent | on HN ]
Re:Raspberry Turk by joey_meyer on Saturday April 01, 2017 @03:42PM (#14013766)
The error was too high so I haven't integrated it into the actual code base yet, so it hasn't been a problem in practice yet! Ha. Yes I think the best thing I can do will be to collect more data, I am hoping that will close the gap a bit more. Also, I haven't spent that much time tuning hyperparameters.
[ Parent | on HN ]