ehrenbrav79

4 ms·

1. 1.▲
   by ehrenbrav79 10y ago

   I think that's the epsilon parameter. That's the chance that, instead of playing the move Mario thinks has the highest value, he plays a random one instead. That in theory should help prevent him from falling into a local minimum.
2. 2.▲
   by ehrenbrav79 10y ago

   Definitely use a GPU if you can - it's just so much faster. You could also try training it on a powerful machine in the cloud if you don't mind spending the money for it... I'm running a 8 core machine, 12+ GB RAM, and a 980
3. 3.▲
   by ehrenbrav79 10y ago

   Yeah, you'd need to "punish" it somehow for running stop signs and lights :) If there's a good driving sim game on Nintendo, it should in theory take minimal modification of my existing code to make it work...
4. 4.▲
   by ehrenbrav79 10y ago

   Glad you think so! It was a lot of fun to put together and I'm contemplating working on something like Zelda next now that I have a computer-controllable NES emulator working...
5. 5.▲
   by ehrenbrav79 10y ago

   The video was made after 4M training steps. At that level, he routinely clears the first level and makes it about halfway through the second. The improvement was still going strong at that point, but I ran out of patience. I'm doing an
6. 6.▲
   by ehrenbrav79 10y ago

   No - I actually didn't use a random start. But when Mario dies far enough along in a level, he starts from near that point rather than from all the way back at the beginning, which kind of serves that purpose. I agree that what's