Recently I read some paper about Libratus, learn of it based on 3 main components, the most important parts is Nested Subgame Solving——useing real-time computer base supercomputer in Pittsburgh, Libratus can response different bet size on the turn and river bet round.
Then we know Snowie use only 4 bet size with fix range and one size includes the whole range. The AI find the best counter-strategy through board texture. If opponent change the bet size, Snowie doesn't change its strategy, So I curious how Snowie made its range if I don't use the optimal size in his recommended since it not has the ability to resolve, then I find its Advise range doesn't match the Hand range, even I use the optimal bet size, that means: Snowie will change its hand combos after opponent's move.
That's impossible in Solver style software since every hand had fixed on preflop, So in my opinion: I think Snowie has little cheating, maybe that's why It will give some wired advice on the river with the error accumulate. Also I don't know how Snowie designs its range in the challenge.
On Snowie's official website, it has an
explanation about this:
what do you think about that? Does this strategy really robust like its described?
Last edited by jasonhau; 06-06-2018 at 10:13 PM.