Draw against Stockfish

Of course, it should be noted that I was limiting the engines think time to blitz conditions, but given my comparative strength, I was still pleased to catch the engine in this amusing little loop. 

That's what happens when you turn off its brain.

I'll bet you can outrun a car too under certain circumstances.

Yeah, because stockfish automatically drops about 2000 rating points when you limit its think time a bit. 

If you want to occasionally draw Stockfish through 3-position rep, leave it as it is. However, if you don't want it to do that, set the contempt value to a positive 20 or 25 centipawns.

Yes it does. It depends on what type of computer you are running it on and how you set it up. How much time you gave it. Whether or not it had access to an opening book. Lots of other things in the configuration you can mess with.

All sorts of things that you did not mention.

There is nothing special about that.

I could do this against Houdini easily.

Anyone can force a draw.

The opening book was on and I ran it on a mid range windows 7 laptop, calm down sport. As I said, I had it playing under blitz conditions, as I was not exactly taking a tremendous amount of time on my own moves. Either way, it's still an engine with an opening book, and it's still an engine rated over 3000.

Time limit or no, it's still a more than competent oppnonent. I never said this was a stunning victory or anything, I just said it was an amusing loop to catch an engine of this level in. 

I really don't think it's scientifically sound to say that anyone could force a draw against a 3000 something rated engine. This is the sort of claim that requires some supporting evidence. 

It really depends on how sneaky Lightyearz meant to be. You could set the contempt to a large negative number, and the engine would try to draw on every single move. In that case, Lightyearz would be completely correct.

In this case, I think there was a very good reason to ask him to defend his claim by explaining what conditions he played under.

It seems reasonable to assume that if a 1257 rated player can achieve a draw against a 3000+ rated engine, something has been done to make the computer play more weakly than it usually does.  

He admitted doing something to limit it. We just don't know what he did. We still don't know how many processors his computer has/was using, how old it is, what he means by "under blitz conditions", what he means by he "was not exactly taking a tremendous amount of time", etc.

If he achieved the draw under legitimate conditions, I will take my hat off to him.

Well, clearly there was no opening book installed, because the engine was calculating on Move 1.

A depth of 16 ply isn't THAT bad. He was probably only giving Stockfish 5 or 6 seconds per move.

Probably more important were the default settings. Stockfish defaults to one thread, so I suspect that was all the engine was running. Also, the default contempt setting for Stockfish is zero, so it's going to be very amenable to accepting perpetuals.

If drawing with Stockfish set at a certain level is out of the ordinary for him, then it's an achievement and worthy of congratulation. This whole "It's not an achievement because I'm better than you" or "Well you're still under 2000 so you can't have played a good game" attitude gets pretty lame after a while.

OP: Well done, nice game!

OK, I didn't notice that. So 1 sec per move. Still, 17 ply isn't anything to sneeze at when us patzers have to play it.

And I'm gonna beat this d*** dead horse into the ground ---The important factor in the game is the default contempt setting of zero. With that setting, Stockfish will walk into a lot of perpetuals, as evidenced by the OP's post.

Yes. I can draw Stockfish 5 everytime on ICC (they have some program allowing you to get knight odds vs stockfish.) by just letting it repeat. OOf course stockfish takes the draw because he is down 3 pawns, so even in bullet conditions, most of the time I can draw stockfish 5.

Again, these really were not legitimate conditions, and I never implied otherwise. I was giving the engine a few seconds or so on each move (still enough for it to make very sound moves, really), and it's not as if I went to great (or any) efforts to conceal as much. It was just something of an amusing little surprise for myself, as stockfish, even limited to this extent, is usually good enough to kill me on a regular basis. 

The fact that FMs can do this on a regular basis really isn't something I care about, as I am not actually an FM, thus, this is not something I regularly am able to throw together. 

A couple of days later I came across the fact that Stockfish 5 doesn't even use an opening book. (Comment from Stockfish support.) As a result, crippling it by not allowing it think for more than a second probably does knock a couple of thousand points off its rating. 

I don't know what the minimum time conditions are for expecting it to perform at a decent level.  One website show ratings for computer playing computers at 5 minutes base with a 3 second per move increment.   That site shows PB (permanent brain)/ pondering on for all engines tested.

I hope poor old Mr-Endron hasn't completely given up on the thread. He just wanted to show an unexpected three-position draw against an engine, and now he has people trying to pick apart his game.

Anyway, after doing some testing, I don't think it knocks off anywhere near 2000 rating points. I ran 10 games using Stockfish 5 x64 versus a 2200 engine (Chispa 4.0.3). Stockfish used 1 thread, no opening book, pondering off, and only 1 second per move. Chispa used 1 thread (it's not mp), an engine opening book, pondering off, and 3 minutes per move (classical time control). Amazingly enough, Stockfish still managed to win 9 games and only draw 1 game. I know 10 games is a very small sample, but I think we can safely say that Stockfish at 1 second per move is way north of 2200 elo...Maybe even north of 2500 elo.

I still believe that the draw had almost nothing to do with the engine strength and everything to do with the contempt value setting.

That's very interesting.  Does 1 thread mean you limited it to 1 processor? 

I had searched for quite a while and was surprised that I was unable to find evidence of similar experiments trying to handicap modern engines.

I agree with you that the "contempt" factor being altered can make engines more likely to accept draws.  

In a nutshell, yes. The default setting for Stockfish is one thread; If you want more threads, you have to change the engine configuration setting.

I guess it's not a question that gets asked much. Since the proof is in the pudding, I just decided to try it for myself to see what would actually happen.

It was really fascinating to watch the games (well, parts of them, anyway - at 3 min/move for one engine, the game progresses very slowly). Chispa was actually processing more nodes per second than Stockfish, so with Chispa's 3 min/move time control, Chispa was racking up a huge advantage in total nodes processed for any one move. I could only conclude that Stockfish's other search & evaluation techniques were far superior to Chispa's.

BTW, I had a hard time setting up the time handicap games. Winboard was the only GUI that I had that would allow the time handicap games, and I'm not very high on the Winboard learning curve. I couldn't get the time handicap to work for Winboard tournaments, so I had to set up each game individually, and even then I had to futz with the GUI settings quite a bit.

And in retrospect, the results probably shouldn't be too surprising. There's a rule of thumb in computer chess that states a doubling of thinking time should result in a strength increase of about 50 or 60 elo points. So, going from 1 second/move to a classical TC is between 7 and 8 time doublings. So, 7 (or 8) x 50 (or 60) elo points = 350 to 480 elo points. Therefore, you might reasonably expect the time-handicapped Stockfish strength to be about 3240 - (350 to 480) = 2890 to 2760. (Note that only testing of many games could determine whether this rule of thumb holds over many time doublings. Also, you might have to knock off some elo points for not using an opening book.)