A revolutionary new tool for chess analysis that uses a chess engine such as Stockfish or Lc0 to keep calculating the best lines starting from a position using a creative 'iterative iterative deepening' approach. It can even be used to provide a more accurate evaluation of a position than the engine.
Modern chess engines are extremely good at calculating the best move in a position. However, one of its main weaknesses is evaluation. This tool is used to fix the following two main problems:
- Evaluation can be unreliable. This means that even though the best moves are usually very stable, the evaluation can fluctuate a lot in some positions. This tool solves this problem by forcing the engine to re-evaluate the position after every few best moves are played, resulting in increasingly accurate evaluation as the game progresses. For instance, PVplayer with 10 iterations and 100 million nodes per iteration usually provides a more accurate evaluation than Stockfish with 1 billion nodes, especially in complicated positions where over-pruning causes a single search run to be unreliable.
- Evaluation can be vague and meaningless. Especially after the Stockfish eval was normalized, a +1.0 eval is now tied to a "50% win probability". This causes a huge problem for human users who have no idea whether this +1.0 eval is enough to force a win. PVplayer solves this problem by running the search deep enough into the game to calculate to a tablebase position where the Win/Loss/Draw is confirmed.
python-chesslibrary- An executable chess engine (e.g. Stockfish). The path to the executable can be modified in config.yml.
cd src
python3 PV_trace.py
There are many terms in config.yml that can be modified to suit your needs.
This includes the limits for evaluation, number of PV moves to use for each iteration, etc.
Modifying these terms may increase/decrease the accuracy and speed of the analysis.