hi, i see your Sapling v1.2.2 is now mentioned in the CCRL lists .. it has rating 3337 : congrats !!
indeed i was aware of your progress ..
i know Sapling eval is based on its NN, but i wonder : do piece values exist like 100 for a pawn, 300 for a Bishop and kNight, etc. ? Or maybe all values in a NN are based on win-lose-draw statistics ?
i have no clue about the inner working of your NN or its training, but if these basic piece values exist, could those be dynamic ? While in some "position type" the piece can be valued differently ..
at this moment i have no better way to put my idea.
i await your response, i can elaborate.
-Roelof
hi, i see your Sapling v1.2.2 is now mentioned in the CCRL lists .. it has rating 3337 : congrats !!
indeed i was aware of your progress ..
i know Sapling eval is based on its NN, but i wonder : do piece values exist like 100 for a pawn, 300 for a Bishop and kNight, etc. ? Or maybe all values in a NN are based on win-lose-draw statistics ?
i have no clue about the inner working of your NN or its training, but if these basic piece values exist, could those be dynamic ? While in some "position type" the piece can be valued differently ..
at this moment i have no better way to put my idea.
i await your response, i can elaborate.
-Roelof