Add Donk Sizings for turn and river

[dpmaloney]

Hi,

In an effort to compare this solver with Pio and Jesolver (A third party solver that is much faster than pio) I decided to implement the donk sizing options that pio uses. If you want an example of this in action, check out this page where oskari tammelin (a member of the alberta research group that solved limit holdem) compares his commercial solver Jesolver to piosolver. He uses this file to create his tests, which uses donk sizes of 50% and 1000% on the turn, and only 1000%.

Adding this to your solver decreases the memory needed and provides a more accurate comparison to Jesolver and pio, as well as removing donk bets that most players (and solvers) will never use, but would take up lots of memory when using them in the solve.

Feel free to change anything about the code style/implementation to fit your style as I know that this is not the best way to achieve what I want.

Thanks!

[b-inary]

Thanks for your pull request!

I wasn't sure whether I should add the feature that allows users to specify the size of donk bets, but if there is a demand, then let's add it. I would be happy if the implementation policy would be as follows:

• Add the following DonkSizeCandidates struct to bet_size.rs:

#[derive(Debug, Clone, Default, PartialEq)]
#[cfg_attr(feature = "bincode", derive(Decode, Encode))]
pub struct DonkSizeCandidates {
    pub donk: Vec<BetSize>,
}

• Add impl TryFrom<&str> for DonkSizeCandidates to bet_size.rs, like BetSizeCandidates.
• Add the following fields to the GameConfig struct, and set the default values to None:

pub turn_donk_sizes: Option<DonkSizeCandidates>,
pub river_donk_sizes: Option<DonkSizeCandidates>,

• Apply the donk sizes if (turn)|(river)_donk_sizes.is_some() && matches!(info.last_action, Action::Chance(_)) && info.last_aggressor == player_opponent as i32.

Also, I think the current implementaion is buggy in the following situation:

• Flop: Check -> Bet -> Call (last_aggressor = IP)
• Turn: Check -> Check (last_aggressor = IP, but should be -1)

In this situation, the OOP's bet action in river is called a probe bet, not a donk bet. I prefer the following implementation:

• If someone checks, set last_aggressor = -1.
• If someone bets or raises, set last_aggressor = node.player.

I could make these changes, but I would be happy if you would modify the pull request. Thanks again!

[dpmaloney]

Sounds good, I can do that! Thank you for fixing it up for me, I made a rought attempt and didn't bug test it but good catch! I think it might be useful for people to use to compare with pio as it is a commonly used feature there.

[dpmaloney]

@b-inary How would you like me to deal with the way let (candidates, is_river) = if node.turn == NOT_DEALT { in push_actions works, as candidates would have a different type if we are using DonkSizeCandidates. I am thinking that I should just create a new donk_candidates so as not to muddy up the code.

[dpmaloney]

Hi @b-inary let me know how the changes look!

[b-inary]

I am very sorry for this very late response. Anyway, I'd like to merge this PR. Thanks so much for your contribution!