Certainly a fun thing to ruminate over, but the model Matt uses here is way too simplistic to draw any conclusions from about the sufficiency of using a NN.
A short stack raising >11% when he shouldn't be is a pretty severe leak ;)
Just a thought: I don't see the implementation of the algorithm here, but if you're using the values of individual cards scaled to 1, then you are saying that v(A) - v(K) = v(8) - v(7) which seems absurd. Imagine that there were (fantasize for the sake of argument) a few shoves in front of you, then should be that v(AK) - v(KQ) >> v(87) - v(76).
I'd need to give more thought to this problem in general, but I have to ask Matt...why did you choose short stacking full ring NLHE? It seems mechanical for someone very familiar with the game, but it's so situational. The first area I'd think you'd be most effective would be LHE-HU-SNG's.
Hey jp, great points. I would have posted the code, but for the life of me I can't find it anywhere. I had the Excel sheet stored in a different folder, which is why I was able to post it.
Anyway, I think I was thinking that including the difference from the average value would help the neural network identify pairs. In retrospect, it should have been able to recognize that simply from the card values, no?
Limit holdem had been done a lot and No Limit was my area of expertise, which is why I chose it. I didn't have that much experience with shortstacking, but I had read pretty much everything there was to read on it and it seemed like a good start because of its simplicity. Little did I know that the profits from shortstacking come from a very few carefully timed moves that are close to impossible for a rules-based bot to emulate (more on that in another post). Eventually I decided to build a no limit heads up bot, which is where my success wound up being.
I didn't mention it in the post, but the key was moving away from rules based to a value based system, where the bot calculated the profitability of its options and decided based on that. It was a lot easier to debug, but there were a whole set of challenges to that too, like estimating your opponent's range.
Just a thought: I don't see the implementation of the algorithm here, but if you're using the values of individual cards scaled to 1, then you are saying that v(A) - v(K) = v(8) - v(7) which seems absurd
Instantiating the features correctly is crucial to getting good neural network accuracy.
It might be the case that a thermometer representation for the card value is more appropriate.
The thermometer representation is that there are thirteen bit-values, and the first n values are on for the value n.
Using this sort of input representation, the network can model the non-linear relationship between different card values, instead of assuming a strictly linear correlation.
It's a neat idea, but I doubt it would work well because the hand value in this context is so qualitative (whereas in HU LHE it would be more quantitative.)
A short stack raising >11% when he shouldn't be is a pretty severe leak ;)
Just a thought: I don't see the implementation of the algorithm here, but if you're using the values of individual cards scaled to 1, then you are saying that v(A) - v(K) = v(8) - v(7) which seems absurd. Imagine that there were (fantasize for the sake of argument) a few shoves in front of you, then should be that v(AK) - v(KQ) >> v(87) - v(76).
I'd need to give more thought to this problem in general, but I have to ask Matt...why did you choose short stacking full ring NLHE? It seems mechanical for someone very familiar with the game, but it's so situational. The first area I'd think you'd be most effective would be LHE-HU-SNG's.