[Info] Another test, with killer moves

Killer moves implemented with a reasonable gain:

1: Mediocre v0.34 622,5/1000
2: Mediocre 1.0 377,5/1000

Still some ways to go, but certainly getting closer.

I have some suspicion the new version is forfeiting some games due to not returning a best move. I have a safety measure for this in v0.34 but not having the root search in place for v1.0 it's missing there.

There is also something fishy with mate scores in the new version, will have to look into that.

I'm hoping root move ordering will bring v1.0 up to par with v0.34. An even score at that stage would certainly be thrilling as there is still extensions, futility pruning and internal iterative deepening left to do.

[Info] A little test

Getting closer to v0.34:

1: Mediocre v0.34  711,0/1000 
2: Mediocre 1.0    289,0/1000

What's left to implement is internal iterative deepening, killer moves, extensions and a more elaborate root move ordering. I should try history moves for ordering as well, though from what I've read they seem to have little relevance anymore. But trying doesn't hurt.

[Info] Starting on basic move ordering

I've spent the last day trying to identify a bug that made Mediocre go haywire (dropping pieces, missing mates in one etc.) when the hash move was used for ordering, which is of course really strange since it shouldn't add or remove any moves, just search them in a different order.

Of course it turned out that one of the few new things I've added was the culprit.

Instead of keeping two integer arrays, one for the moves and one for the ordering values, I've mashed them together into a Move-object. Pretty much only for clarity (it shouldn't cause any performance degradation).

Since I didn't use any move ordering at all, ordering part of the Move-object was left untouched. That is until I started to use hash moves which I order as 10,000 (search first in any circumstance) and then -10,000 to mark it already used.

As no other moves received a score, that -10,000 was lingering in the move array and eventually caused no moves to be searched. :)

-

Well, that's over and done with and using the hash move for ordering gave the following result in my standard mini-test:

1: Mediocre 1.0+  14,0/20
2: Mediocre 1.0   6,0/20

Barely any statistical relevance really, but I'm happy.

Steady going, on to more move ordering.

[Info] Check evasions

I'm taking baby steps and try to get everything right from the beginning.

Having considered Jon Dart's suggestion and examined a few open source engines (GreKo http://greko.110mb.com/ is a favorite) I decided to add the notion of not standing pat in the quiescence search if in check.

That required continuing for one step and a need for check evasions.

My new check evasion code takes about a fourth of the time compared to the old one, which seems nice. And there might still be a few more optimizations to do.

Last bug a squashed concerning it was evading a check by capturing to the last rank and promote to a non-queen.

How I've missed these ultra-specific bugs... :)

[Info] A new approach

After having spent about 20 hours the last week trying to tweak Mediocre into becoming atleast a tiny bit better I suddenly had an epiphany and gave up.

It's been far too long since I last worked with this and there are so many things going on that I have no idea if they help or hurt.

So I scrapped the entire search function and started over...

Currently Mediocre 1.0 beta searches about 4-5 ply on average and gets beaten to scraps by v0.34. :) So far I've implemented quiescence search and transposition tables, and pretty much nothing else, not even rudimentary move ordering.

In doing so I've noticed that something is not well with how I handle transposition scores. For example I've been returning alpha/beta scores from the table rather than the actual alpha/beta in the node. This seems very wrong.

There's also something fishy going on with the hash move, but I'm not sure if this is true for v0.34 (there might be something fixing it along the way).

-

Anyway, I'm having fun, and the improvements come in 100 rating chunks again. My favorite kind of improvement. :)

Let's see if it ends up surpassing v0.34 eventually. I certainly hope so.

Edit:

First non-loss for the new engine! :)

   Engine            Score
1: Mediocre v0.34    19,5/20
2: Mediocre 1.0 beta  0,5/20

Sad day

Sometimes when you innovate, you make mistakes. It is best to admit them quickly, and get on with improving your other innovations. - Steve Jobs

[Info] Some info please and a TODO

I've implemented Jon Dart's suggested changes and I think they bring some improvement to playing strength. However to be honest, I'm not sure. My small tests so far point in all kinds of directions.

I need to set up a new testing environment and have taken a look at cutechess-cli which seems really nice. But I'm having problems with Mediocre crashing (or rather stalling) while running tournaments. Especially the new version. It also leaves really expensive Mediocre processes running, completely clogging up the computer.

I have no idea why.

This brings me to an important point that's been missing in Mediocre since the beginning. A decent logging framework. This should shed some light on those pesky stalls.

-

So a TODO for the next couple of days:

General

Set up a testing environment using cutechess-cli, some creative scripting (to be able to run gauntlet matchs) and a number of well-chosen engines. Most likely using the YATS testing again (even though I can't get ProDeo to work anymore).

Get some decent logging going. This should help immensely in tracking down bugs and getting a feel for where improvements might be needed.

Engine improvement

Apart from Jon Dart's improvements:

Pondering - It's time to get it done, and shouldn't take too long.

Endgame knowledge and evaluation - Take a look at what I started with two years ago. I might be able to shake out some improvements.

Reboot

What happened to the repositories?

It has been a long long time since I took a look at the source code at https://sourceforge.net/projects/mediocrechess/. I have a firm memory of using CVS, but there's no trace of CVS ever being used.

Well, I'm planning for a new version of Mediocre. Thanks to Jon Dart (http://www.arasanchess.org/) for coming with some suggested improvements that sparked my interest.

So in light of that I'll set up everything again. Mediocre v0.34 has been uploaded to the SVN on Sourceforge and I'll be using it "correctly" this time.

Trunk will be used for ongoing development, branches for specific ongoing changes, and tags for released versions.

Look forward to a new version in the near future.

Long time no see

In my last post I promised to update Mediocre before 6 months. It has now been almost 16 months. Well well.

On a personal note I now work with Java development in a investment bank, and while it's a super job it leaves me little time for such time consuming activities as chess programming.

Mediocre is as mentioned not abandonded, just resting, and while I won't make any promises this time, I will sooner or later get around to updating again.

This post is mostly here to let you know I'm still around, and if you have questions I'm still reading your mails (mediocrechess@gmail.com, but I will most likely answer from zlaire@gmail.com).

Until later.

Update

I am happy there are so many people asking for new updates, so here is a small update on what is going on.

When I get to work on Mediocre I always do it 100% of my time, which means many updates in a short time. This is the only way I know how to work on a project, since I need to keep everything in my head (I guess I am not a very organized person in general).

As I am a full time computer science student working on my last year of my education there is little time for anything but studies.

However, this project will never be abandoned, atleast not in a foreseeable future. There will just be times where it is resting for longer or shorter periods.

Upcoming updates

I have a version that is (and has been for some time) basically ready for release, including ponder and some other nifty things.

Also people have mentioned that the memory usage is a bit sketchy and I will have to take a look at that.

So in summary, there will be updates for Mediocre in the future, but I really can't tell you when. But it will not be another 6 months, that is for sure. :)