eyalk11 / chess_stability

Provide a new method for assessing stability of chess positions

Chess Stability

Provide a new statistical tool for evaluting chess postions (in addition to standard score). This tool is stability score (0-100). And describes how easy it is to break position / how stable the position is.

Usage

Installation

1. Download stockfish
2. Do:

****
pip install jupter
pip install python-chess
git clone [e-mail]:eyalk11/chess_stability.git

3. Update STOCKFISHPATH in the notebook.

Running

cd chess_stability 
jupyter lab -> select notebook

You can view your own games by adding a cell display_game(pgn)

(This shit is interactive!)

or evaluate a position by display_board(fen,iswhite).

By default, takes iswhite from fen.

See examples.

Method

We define reasonable move (or not blunder) , as a move that does less damage than 200 centipawns.

We define a function $D(x)$ that returns $0$ if $x \ge 0$ and it is white's turn or $x \le 0$ and it is black turn. Otherwise, it is the identity. This function is meant to filter out improving moves.

$C_1$ and $C_2$ some constants.

The method is the following:

1. Evaluate all reasonable moves of certain depth (for both players) $R$
2. Evalute current position $\mu$
3. If the position is worse than 200 centipawns (to each direction):

$$V= \sum_{r \in R } e^{-C_1 (D(eval(r)-\mu))^2} $$

4. Otherwise:

$$V= \sum_{r \in R } e^{-C_2 (\frac{ D(eval(r)-\mu)}{\mu})^2} $$

5. Denote stability by $\frac{V}{|R|}$

This is a twist on gaussian distrubtion, taking current position as $\mu$.

Notice

It is licensed under by-nc-sa

Contact me for questions

[e-mail]

Eyal Karni