Sprint 5 - v5 - 12.33 hrs

Yeah, I must have been high when thinking that it would be easy to re-write it in C++, a language that I have basically no experience in (I decided that I might as well go the whole length rather than just re-writing it into Cython).

P.S. I have decided to split this sprint into two parts, because I ended up using a different approach than I originally intended.

v5.1 - 7.33 hrs

Goals

• Re-write the whole library in C++

Logs

Ok, from my research, I had found that bitboard chess engines are the fastest, so I decided to go with that approach. Initially, I wrote a function to convert a FEN string into usable information. To start with, I wrote some code to generate Rook, Queen, and Bishop moves as they are the simplest. I used switch statements wherever possible because switch statements are faster than if statements in C++.

Soon after, I gave up on writing the chess library from scratch due to the complexity of Knight and Pawn movement and how slow it already was.

It was here that I found this library. It was a bitboard chess library written in C++, and it seemed to be blazingly fast:

• from 10-20 times faster than my C++ implementation (Max's (Disservin) library is also used to power Stockfish, one of the strongest chess engines in the world)

v5.2 - 5 hrs

Goals

• Same as above, but using Max's library, i.e. remake whatever progress I had.

Before using the chess library, I decided to run some benchmarks (PERFT against the old python library)

• First Test (starting position):

  Library	Depth	Time (seconds)	Nodes Searched	Nodes per Second (approx)
  python-chess	5	30.879215955734253	4,865,609	157,569
  chess-library	6	28.7222588	119,060,324	4,145,223

  INFO

  This is a whole move deeper and is still faster.

• Second Test (starting position):

  Library	Depth	Time (seconds)	Nodes Searched	Nodes per Second (approx)
  python-chess	7	17484.683883190155	3,195,901,860	182,783
  chess-library	7	751.4311222	3,195,901,860	4,253,087

  Therefore, we can conclude that chess-library is 23.2685 times faster.

Great! Now that the bottleneck is again my code, I spent the rest of this sprint re-writing all of my existing code to work with this library. This took a while, but I got it done.