New Directions

Russell Sherwood  Sunday, March 31, 2019

This I write this feeling probably the best I have done for around a month – it's quite a strange virus that removes almost all energy and motivation and gives mild flu symptoms which last for weeks. Even more so when it affects more than one person! I can tell you its not fun when is coincides with a surge at work as well!

 Still, time to crack on and clear all the backlog!

Anyway back to Correspondence Chess. The last few months have generated quite a few interesting developments with regard to CC and Engines. These developments are, I believe, quite positive as well as having the potential to be quite disruptive to use Marketing speak.

I believe most people will have heard of Alpha-Zero and the advances this has delivered. This is covered in the excellent book “Game Changer” (In an in-depth review from a CC perspective will follow in April – let me clear the backlog first!).

Whilst AZ is very interesting and as I will cover next month, gives a number of insights very useful to the CC player, it is in itself not going to have a direct impact on CC for a number of years, it at all – why: because until the software and hardware become readily available it will not be within the realms of the ordinary CC player.

So discounting AZ, what is changing that is going to disrupt CC? If we consider a year or two ago the situation was, in a nut-shell a choice of Stockfish, Houdini or Komodo, with Stockfish the choice of the majority and the main difference between players being a combination of Hardware, Engine-wrangling skills and Chess knowledge, pessimistically probably in that order.

However, the situation has changed over the last year…..

• LCZero: The development of the AZ inspired open-source engine and crowd-powered Networks has been truly amazing, with LCZero now being on par with Stockfish in typical engine tests.
• Komodo MCTS: Whilst not as strong as Stockish, the Komodo teams efforts to bring an effective Monte-Carlo Tree Search to an AB engine have generated an excellent tool for the CC player.
• Monte Carlo Tree Search: For many years the AB engine has ruled the roost (and still does in very rapid time controls) but the use of MCTS has flourished in the last year giving a previously unseen level of variety in move possibilities
• New Engines such as Ethereal
• Stockfish derivatives: A number of developments have taken place over the last year which are of interest to the CC player. Examples of this are:
  • The excellent Raubfisch , which continues to tweak its results upwards
  • Stockfish Cluster – an attempt to allow the running on massive hardware
  • ShashChess – A very interesting offshoot which adapts itself based on the type of position.
  • Sugar MCTS – a powerful implementation utilising a version of MCTS but also including a learning function.
  • Thothfish – a very new kid on the block but bringing two new very interesting features of a “Magic Tactic Solver” and more interestingly a way to manipulate analysis to swap or not swap certain pieces.

 

• 7 Piece TB: These are now coming available for download. In reality, I don’t see anyone downloading the full set, due to speed and memory limitations but the downloading of specific ones can aid enormously.

Putting this together (and other advances I have not mentioned) we are now seeing a situation develop where there is more than “one horse in town” in the form of Stockfish, where these different approaches all have very similar performance in traditional testing (The flaws in this for CC I will cover another day!) but often give different suggestions in the same position.

The outcome of this is two-fold: Human input will become more important, the decision to take path A instead of path B, C or D and that this will lead to a reduction of the draw rate or rather more hard-fought draws at a minimum!

So a positive picture for CC? I believe so.

Anyway back to the backlog clearance, then onto launching EARG, writing the Engine and Introduction to CC Guidebooks!