I won against the trained TicTacToe AI (Q7)

Fraunhofer-IMS / AIfES_for_Arduino

This is the Arduino® compatible port of the AIfES machine learning framework, developed and maintained by Fraunhofer Institute for Microelectronic Circuits and Systems.

GNU Affero General Public License v3.0

216 stars 45 forks source link

I won against the trained TicTacToe AI (Q7) #21

Open Makrine opened 1 year ago

Makrine commented 1 year ago

Hello,

I uploaded the example code to Arduino UNO board and managed to win.

This is the win patter:

Me: b2 AI: a1 Me: a2 AI: c2 Me: b1

here AI plays c1 instead of b3 because I can win at b3.

I tested the same thing for F32 and the AI made the correct move there.

PierreGembaczka commented 1 year ago

Hi,

that may well be possible with the quantized version. Quantizing the weights reduces the accuracy of the model.

Best, Pierre