Sherwood, Russell Sunday, July 30, 2017
The Similarity test was created a few years ago to be able to compare engines, mainly to be able to attempt to detect cloning (The process of taking someone else’s engine, making a few minor changes and renaming it as your own, without any acknowledgement of the original. There were many deep, nasty, arguments of the type “this engine is a copy of that engine”. These do not concern us here!)
An important area of generating winning chances is developing moves which are not the First choice or Principal Variation of our main engine. A number of analysis methods do exist which can do this, such as Subtractive or Monte Carlos Analysis but a starting point before all of this is to consider which Engines tend to give “Original” moves.
A note of caution here – “Original” moves are not necessarily “Good” moves but often the suggestion of a different (weaker) engine can help us find the best path forwards from a position. In some cases simply showing the move to an engine will help it find a better line. In other cases much more work is required.
So for the trial to determine originality I consulted the FastGM rating list http://www.fastgm.de/ on its slowest time control and ran all the engines I had from this list (plus some which I consider to be the best older ones: Gandalf, Junior, Hiarcs…….)
The results are interesting – As you would expect most of the Stockfish variants have a very high score………………………..
So what we are looking at is a heat map – it works by looking down the left-hand column to find the engine we are interested in – we then read across – the similarity in move choice is shown, the higher the score the more similar the moves selected. In general it is better to ignore the numbers and look at the colours – The more red the more similar the choices, the more green the less similar. The Engines are ordered in a rough ordering of strength – with the ones on the left/at the top being stronger and the ones on the right/bottom being weaker. The difference between the top and bottom engines is between 300 and 500 elo, depending on which rating list you believe! There were also a few engines which failed to run in the trials – Fizbo and Giraffe. Of these Giraffe is the more interesting as its evaluation is based on a very different method, however the project is currently shelved!
We can divide the map into four basic zones:
Top Left (The Red Zone)
The strongest engines – many different variants of Stockfish, so it is unsurprising that these engines are fairly similar in outlook. In general engines in this area make poor bedfellows in terms of bringing a different point of view to the table.
The Middle (Central Red Zone)
Here we have a lot of engines with similar outlooks. Without going into much detail, a number of these were/are accused of being copies of other engines.
The Bottom Corner (Green Zone along the Right and Bottom sides)
These engines tend to come up with different moves from the Top Engines – the downside often being that they’re simply weak rather than original moves and care should be taken in any suggestions.
The Interesting Zone (Green-Yellow along the top and down the left hand side)
These are the engines which tend to give a different slant of a position compared to the Top Engines. Particularly interesting are Houdini 4, Boot and Deep Shredder 13 and Andscacs. These engines are not the strongest but are not weak either. Houdini 4 is the oddity of the group as it is “obsolete”, whereas the others are all current.
Anyway If you want a different opinion to HockmodoFish, these engines represent probably the current best mix of originality and playing strength.
Updated Sunday, July 30, 2017 by Russell Sherwood