14 November 2016

TCEC Season 9 Superfinal Underway

Let's put Korchnoi's Events aside for a few months and look at the two World Championship matches currently being held. For the more important of the two matches, the human version, I'll refer to my World Championship Blog, where the most recent post was World Championship Bullying, and where I'll be tracking the event on at least a weekly basis.

For the stronger of the two matches, the engine version, I'll mention my previous post Carlsen, TCEC, Karjakin, Korchnoi (October 2016), where the TCEC acronym is the bit I'm interested in now. By coincidence, the human and engine matches both started at the same time, and the next day Chessdom.com (the official site for the engine match) reported Carlsen – Karjakin game 1 draw:-

Stockfish 8 and Houdini 5 also draw in game 1 • Together with the highest individual event in the chess calendar -- the World Championship -- started also the highest event for computer chess, the Top Chess Engine Championship (TCEC). In the Superfinal of Season 9 Stockfish and Houdini started their 100 game match with a draw.

In the nearly three days since the event started, the engines have played 12 complete games non-stop with Stockfish leading Houdini by +3-0=9. How does one engine with a rating slightly over 3200 Elo beat an engine with a rating slightly under 3200? Like this...

tcec.chessdom.com/live -> Archive -> Season 9 Superfinal game 5

...The graph in the top left ('Depth') shows that Stockfish (playing White) consistently searches variations to a greater depth than does Houdini. This means that Stockfish finds any winning path first ('Evaluation', top right), partly because it goes into endgame lookups faster than its opponent ('Tablebase hits', bottom right). As for the graph in the bottom left ('Speed'), Houdini appears to be faster, but that doesn't matter (whatever it means).

The two other games that Stockfish has won -- games 7 and 9 -- show similar patterns. The bottom line is that a deeper search translates into an advantage. Does Houdini have a chance to win the match? The available evidence indicates that, no, it doesn't. We're looking at something like a +25-0=75 final score, where isolated wins for Houdini will be the exception.

No comments: