TalkChess.com

At this point I'd be happy with a single 2minute/move game from Zenmastur's computer that ends in draw. If every single game is lost then that's some kind of demonstration.

Ovyron wrote: ↑Mon Feb 10, 2020 3:53 am At this point I'd be happy with a single 2minute/move game from Zenmastur's computer that ends in draw. If every single game is lost then that's some kind of demonstration.

Of all the people on this board I think you would benefit most from new hardware. You obviously enjoy computer chess and long time control analysis. I think you're shorting yourself by not investing in what you enjoy. All the things you have learned to do WITHOUT good hardware will only be magnified with “ANY” modern hardware.

My next door neighbor knows nothing about hardware and is on a budget so I put together a list of components for her when she is ready to upgrade. The total cost was $477 US ( no monitor, keyboard, or mouse). The only thing I would change if the build was for playing chess would be to change the graphics card to an RTX so it could be used for NN engines. This would add about $170 to the total. I think a system like this would greatly increase you ability to analyze positions. I.e do what you enjoy.

As long as were on the subject of what would make us happy:

I'd be happy with a very large room full of racks of dual EPYC 7742 systems with 4TB of RAM running at CL12-12-12-25, and 10 RTX 2080 Ti's clocked at 2Ghz, plus 100 PB NVMe 4.0 raid array etc.

Another thing I would like is a book for SF that learns using MCTS/MAB algorithm. Then I could play a few hundred million games and see what the learning algorithm comes up with. But, that's just me.

As it is:

I'm still doing testing and I have more hardware to install in my machine i.e. I'm not done with it just yet. I'm also working an analyzing Tal-Koblencs 1957. I did an analysis back when I had an i5 and 16GB of ram (about 5-6 Mnps). So, I want to compare what I get now vice my previous analysis. But I'll see if I can find some time for a 1.g4 d5 2.c4 or other match. Do you have any particular variation you want to see? Or maybe I'll play a game as white. Who knows what I might do?

I do plan on doing a couple of ICCF GM norm matches (health permitting) as soon as I get my machine the way I want it. I suspect this will take up most of my time once I start. Not sure when that will be though. I don't want to be too committal.

Regards,

Zenmastur

Zenmastur wrote: ↑Mon Feb 10, 2020 5:29 am All the things you have learned to do WITHOUT good hardware will only be magnified with “ANY” modern hardware.

But I can emulate better hardware by just letting every position be analyzed for more time. If I want to see how life would be like, I can just analyze a position for 10 minutes instead of 1, and say "oh, so that's what x10 faster hardware would bring in 1 minute."

I've done all sort of experiments with this, ever since I upgraded from my Athlon 2.0Ghz 32bit 1CPU. Which was exactly 10 times slower than what I have today. Surprisingly I found out that >90% of my moves were the same after upgrading to 2.33Ghz 4CPU 64bit! Because my analysis methods carried most of the weight, my move was decided very soon and more analysis wouldn't help to change it.

The only thing I could do was play more simultaneous games at a higher level, up to +60 simultaneous games. The problem was that the more games played, the more they became a blur, to the point they were so many I was starting to forget my wins! I realized that it was a problem when I was having to timestamp my wins and my opponents' names for the lines I won, only to wonder later "huh? Who was this guy I beat? I don't remember playing this at all..."

Back when I only played 6 games at a time they were memorable, I still remember them. Ironically, the only ones that stand out from the big pile are the lost ones...

So I have toned it down to just 16 games, which are still played at the level of those +60. It'd be extremely easy for me to simulate 4 times faster hardware by doing everything I do, but leaving each position analyze x4 the time, because I do have the time. Except it didn't seem to do anything when I've tried it so the conclusion is that I don't need bigger hardware.

Taking my current game against mmt as an example, it has no clocks, I could easily do everything I do but have each position analyzed x10 the time, and take x10 the time for each move. What I have seen is that all my moves would have been the same, and if I win it turns out having x10 times faster hardware would just have made me waste 9/10s of the resources.

So let me turn things around.

Zenmastur, I'm going to claim that you'd not need hardware that big to play at your level if you had a better analysis method.

As a demonstration, get into a position where you need big hardware to see a move *that matters* (making it, or not making it changes to outcome of the game.) Now, try to analyze the position "blind", as if you didn't know about this move, but using 1/10th of the resources. Is there something you can do to find this move? Is it absolutely impossible for you to find it with 1/10th of the resources?

If you can find it, it means you could have found it in the same time with x10 times slower hardware. And that's how my analysis methods have become so sharp that I don't need big hardware, by playing against strong opponents and seeing moves that I missed with my current methods, I can develop better methods that can find those moves with minimal work. I spent 2 days opponent modeling Harvey in our game, and am now equipped with methods that find all his moves in reasonable time without needing to upgrade my hardware.

Playing the PV of the engine, letting it fill up the hash, and backtrack to see if it finds something better is extremely rudimentary, you could be doing so much more with what you have, so much more that you'd realize you can do fine with what you have and don't need more, or that you could do it with 1/10 the resources. Limiting your resources and trying to find moves that supposedly only fast hardware can find is a key to understanding this.

I hope one day we play a game, the best you have as white against my best black defense, if I can hold with my current hardware, then it suffices.

Ovyron wrote: ↑Mon Feb 10, 2020 6:06 am

Zenmastur wrote: ↑Mon Feb 10, 2020 5:29 am

What are the differences in the ways you two go about doing CC analysis? (Ignore hardware differences.) I'm not so sure, though you often seem to disagree violently about the way to do it.

I maximize the number of positions analyzed while minimizing the amount of effort needed to find the best moves. This includes aiming for positions where future effort will be minimized based on extrapolation of what is happening on the positions (say, if on a line my opponent has 10 playable lines, and after more analysis they go up to 15, and then 20, I just assume they'll go up to 30 and search for an early alternative that wouldn't need to check as many lines...)

But at the core, this is what I do, I make one of these choices:

a) Find a move to extend (usually this was found where this position first appeared in analysis, so this is my mainline)

b) Find an alternative to that main move (a move that could be best instead of a) - for this I can use exclude moves to make the analysis faster than if a) was included, using MultiPV if there's many moves close in score, or ask some other engine if it's a position that requires creativity.) If b) is found, it becomes a) And a sub "b)" would be here so I find an alternative to both a) and b), and so on.

c) Extend the mainline (I have the best moves from both sides, and assume they have been played in the game. Now I do a) but on the tail of the variation.)

d) Refute the mainline (this does b) from the opponent's side, trying to find alternative moves that are better than the best defense I've found)

e) Improve the mainline (this does b) from my side to move on the mainline, to see if I can find a better line for myself than what I have)

f) Extend b) (pretend that my mainline has fallen so I need to see if I have something better for my alternative line. This is useful in case d) refutes a) to be worse than b), and e) can't save it, so b) becomes a), and again I look for more alternative lines on here) Please note I don't try to refute b), if a) > b) then what the opponent has is good enough and I don't care if they have something better.

g) Improve b) (look for better alternatives for my secondary line in case it can be made better than a) )

h) Improve other lines (in case both a) and b) aren't best here, I try to find a third move or fourth move that is, depending on how close they are)

i) Recursion (assume I have found the best moves for a) and my opponent's reply against it, now repeat this whole process from my opponent's side, where a) is now my opponent's best reply. I only leave the recursion if my original a) falls below the alternative, otherwise I'll keep doing recursion and adding 1 to my relative depth. It's very important to see progress, either by an increasing eval or by having the opponents' lines falling down so you need to find new lines for them)

And that's it. I don't know how Zenmastur interacts with the positions, but whatever move you can find in 1 hour, you can find it in 6 minutes with the right analysis method (and the devil is in the details, it's very important what letter you choose and how much time you spend on it. Also, my method might crumble when I'm on a lost position, as I spend most of my time returning to h) and I can't ever reach i) .)

jp wrote: ↑Mon Feb 10, 2020 6:17 am

Ovyron wrote: ↑Mon Feb 10, 2020 6:06 am

Zenmastur wrote: ↑Mon Feb 10, 2020 5:29 am

What are the differences in the ways you two go about doing CC analysis? (Ignore hardware differences.) I'm not so sure, though you often seem to disagree violently about the way to do it.

I think “violent” is bit of a reach.

Ovyron wrote: ↑Mon Feb 10, 2020 7:15 am I maximize the number of positions analyzed while minimizing the amount of effort needed to find the best moves. This includes aiming for positions where future effort will be minimized based on extrapolation of what is happening on the positions (say, if on a line my opponent has 10 playable lines, and after more analysis they go up to 15, and then 20, I just assume they'll go up to 30 and search for an early alternative that wouldn't need to check as many lines...)

But at the core, this is what I do, I make one of these choices:

a) Find a move to extend (usually this was found where this position first appeared in analysis, so this is my mainline)

b) Find an alternative to that main move (a move that could be best instead of a) - for this I can use exclude moves to make the analysis faster than if a) was included, using MultiPV if there's many moves close in score, or ask some other engine if it's a position that requires creativity.) If b) is found, it becomes a) And a sub "b)" would be here so I find an alternative to both a) and b), and so on.

c) Extend the mainline (I have the best moves from both sides, and assume they have been played in the game. Now I do a) but on the tail of the variation.)

d) Refute the mainline (this does b) from the opponent's side, trying to find alternative moves that are better than the best defense I've found)

e) Improve the mainline (this does b) from my side to move on the mainline, to see if I can find a better line for myself than what I have)

f) Extend b) (pretend that my mainline has fallen so I need to see if I have something better for my alternative line. This is useful in case d) refutes a) to be worse than b), and e) can't save it, so b) becomes a), and again I look for more alternative lines on here) Please note I don't try to refute b), if a) > b) then what the opponent has is good enough and I don't care if they have something better.

g) Improve b) (look for better alternatives for my secondary line in case it can be made better than a) )

h) Improve other lines (in case both a) and b) aren't best here, I try to find a third move or fourth move that is, depending on how close they are)

i) Recursion (assume I have found the best moves for a) and my opponent's reply against it, now repeat this whole process from my opponent's side, where a) is now my opponent's best reply. I only leave the recursion if my original a) falls below the alternative, otherwise I'll keep doing recursion and adding 1 to my relative depth. It's very important to see progress, either by an increasing eval or by having the opponents' lines falling down so you need to find new lines for them)

And that's it. I don't know how Zenmastur interacts with the positions, but whatever move you can find in 1 hour, you can find it in 6 minutes with the right analysis method (and the devil is in the details, it's very important what letter you choose and how much time you spend on it. Also, my method might crumble when I'm on a lost position, as I spend most of my time returning to h) and I can't ever reach i) .)

Well, that's quite the algorithm. It will take me a while to process it fully.

Could you do a small test analysis for us? Something not too difficult like analyze this position to mate:

[d]2b3k1/r2r2p1/p1N1q3/1B6/3p4/2b1B2R/PPP5/2K4R w - - 0 36

No outside help please!

No, it's not some trick position if that's what you are wondering. If I recall, when I had a machine that was about twice as fast as yours I could do mates in 30 no problem. I hit more resistance when I got to a mate in 35. And Mates in 40 were difficult and time consuming to produce. So, I figure somewhere in this range should stress your hardware a fair amount. If your analysis algorithm is as good as you claim this position shouldn't be much of a problem for you. AB engines excel at this type of positions, especially programs like SF. If on the other hand it's too much for your hardware (it's mainly a hardware test) and your analysis is less than stellar then you may have problems calculating it all the way to mate. So, it's not meant to be so hard that you can't do it just hard enough so that your hardware by itself probably won't get you there. Or so my theory says. Then depending on how much or little you struggle with it will give you an idea of if you need new hardware or a new analysis algorithm or neither or both. This shouldn't take that much time if you are as efficient as you say you are.

When I'm analyzing, I don't try to be efficient for the most part. I like chasing positions all the way to mate if it's within the capabilities of the hardware I'm running on. Even if it takes a lot more time to do that than just finding a move that is “good enough”. A failing of mine that I fully recognize. I try to avoid it when playing CC games but sometimes the urge to find the best move is too strong.

And, you are right my algorithm for finding the best move is much simpler than yours.

Regards,

Zenmastur

Zenmastur wrote: ↑Mon Feb 10, 2020 9:52 am Could you do a small test analysis for us? Something not too difficult like analyze this position to mate:

[d]2b3k1/r2r2p1/p1N1q3/1B6/3p4/2b1B2R/PPP5/2K4R w - - 0 36

You mean some mate score as shown by the engine, right? Not the shortest mate, which I don't see the point of doing...

So for this I'll be using 2 different engines, instead of some boring Engine 1 and Engine 2 let's nickname them:

Femme Fatale: This engine is really good at finding mates at low depth. However, it'll show some insane lowerbounds, like "+M160", which stands for "I can guarantee to win this position, and I can guarantee to do it on 160 moves or less." That she does, she'll decrease their counter with every move, but not with more depth. But I don't think you'd be happy if I show a mate score of 200 for this, would you?

Harem Girl: This one is the opposite, failing to see mate until it's on her face, choosing to report scores of +132.00 instead. Her problem is that these +132 scores may stay the same after a move, or they may go backwards (like from 132.71 to 132.70), so it's not reliable on finding them. HOWEVER, once she sees the mate she'll show a much better lowerbound, such as +M23 for the same position that Femme Fatale showed +M160. This would be the score that you'd want to backsolve.

For these purposes I'll use Femme Fatale for the white side exclusively. The idea is that she'll be leading the position to a mate score. I'll be using Harem Girl for the black side at the tail of the variations, so she tells us better mate scores (which could be lower than what she shows, but we'd be happy if those are brought to the root.)

SPOILER ALERT - I'm not doing it to completion, just for enough time so it's clear how the procedure works.

Initially, I use Femme Fatale at Depth 32 for the main move, and Depth 22 to get an alternative move. This is important in case her mainline falls below an alternative move, so we're not following the winning line. So I start (I'm snipping the end of the variations for clarity):

32/58 1:47 +9.45 1.Rh8+ Kf7 2.Rf1+ Qf5 3.Rxf5+ Kg6 4.Rg5+ Kf6 5.Bd3 dxe3 6.Rg6+
22/38 0:04 +0.32 1.Bc4 Qxc4 2.Rh8+ Kf7 3.Ne5+ Ke6 4.Nxc4 Rac7 5.bxc3 dxe3 6.Nxe3

I'll be making these first white and black moves that it shows in her PV, so 1.Rh8+ Kf7 are moved.

32/54 1:10 +9.07 2.Rf1+ Qf5 3.Rxf5+ Ke6 4.Re5+ Kd6 5.Bf4 axb5 6.Nxa7
22/39 0:04 +0.35 2.Bc4 dxe3 3.bxc3 Qxc4 4.Ne5+ Ke6 5.Nxc4 Rac7 6.Nxe3

2.Rf1+ Qf5 are moved.

32/57 2:14 +10.66 3.Rxf5+ Ke6 4.Re5+ Kd6 5.Bf4 axb5 6.Nxa7 Rxa7 7.Rxc8

Here 2 minutes / move are deemed too long per node, so I decrease main depth to 31, and the depth of the alternatives to 21.

21/38 0:04 +5.15 3.Bc4+ Kf6 4.Rxf5+ Kxf5 5.Rf8+ Rf7 6.Bxf7 dxe3 7.Bc4+

3.Rxf5 Ke6 are moved.

31/02 0:01 +10.66 4.Re5+ Kd6 5.Bf4 axb5 6.Nxa7 Rxa7 7.Rxc8 Ba5 8.Rxb5+
21/37 0:04 +6.74 4.Rf3 axb5 5.Nxa7 Rxa7 6.Rxc8 Rxa2 7.bxc3 dxe3 8.Rxe3+

And so on...

31/02 0:02 +10.66 5.Bf4 axb5 6.Nxa7 Rxa7 7.Rxc8 Ba5 8.Rxb5+ Kd7 9.Rc4
21/36 0:03 +6.74 5.Rxc8 dxe3 6.Rxe3 Bf6 7.Rd3+ Ke6 8.Nxa7 Rxa7 9.Rc6+

31/49 0:59 +10.01 6.Nxa7 Rxa7 7.Rxc8 Ba5 8.Rxb5+ Kd7 9.Rc4 d3 10.cxd3
21/31 0:00 +8.50 6.Rxc8 Rf7 7.Rf5+ Ke6 8.Rxf7 Rxf7 9.Nd8+ Kd7 10.Rc7+

31/49 1:01 +13.01 7.Rxc8 Ba5 8.Rxb5+ Kd7 9.Rcb8 Be1 10.Rd5+ Ke7 11.Re5+
21/02 0:00 +8.40 7.Rg5+ Ke6 8.Rxc8

31/39 0:01 +13.01 8.Rxb5+ Kd7 9.Rcb8 Be1 10.Rd5+ Ke7 11.Re5+ Kf7 12.Rxe1
21/39 0:01 +8.70 8.Rc4 Rf7 9.Re4+ Kd7 10.Rexd4+ Ke7 11.Rc5 Bb6 12.Bg5+

31/60 1:13 +16.01 9.Rcb8 Be1 10.Rd5+ Kc6 11.Rd6+ Kc5 12.Rc8+

I consider a score of 16 as "winning" so I stop checking alternative moves at low depth

31/54 0:08 +49.45 10.Rd5+ Kc6 11.Rd6+ Kc5 12.Rc8+ Kb5 13.Re8

31/40 0:00 +49.45 11.Rd6+ Kc5 12.Rc8+ Kb5 13.Re8 Bb4 14.Rxd4

31/40 0:00 +49.53 12.Rc8+ Kb5 13.Re8 Bb4 14.Rxd4 Be7 15.Rb8+

31/40 0:00 +49.56 13.Re8 Bb4 14.Rxd4 Be7 15.Rb8+ Kc5 16.Be3

31/42 0:09 +M161 14.Rxd4 Be7 15.Rb8+ Kc5 16.Be3 Kc6 17.Rc8+

There we go! Engine showing a mate score. Let's see if it goes up with higher depth?

40/02 0:13 +M161 14.Rxd4 Be7 15.Rb8+ Kc5 16.Be3 Kc6 17.Rc8+ Rc7 18.Rc4+ Kd5 19.R4xc7 Bd6 20.Rxg7 Ke5 21.Rg5+ Ke4 22.Re8+ Kf3 23.Rg6 Bc7 24.Rf6+ Kg2 25.Rg8+ Bg3 26.Rxg3+ Kxg3 27.Rg6+ (44.489.997) 3341

What did I tell you?

So enter Harem Girl... (you'll be able to recognize her as the Depth 40 engine that always plays black moves)

40/44 0:59 +M21 14...Bc5 15.Rb8+ Kc6 16.Rc8+ Kb6 17.Rd3 Ra4 18.Rb3+ Rb4 19.Rxb4+ Bxb4 20.Rb8+ Ka5 21.Bc7+ Ka6 22.Rxb4 g5 23.Bb6 Kb7 24.Be3+ Kc7 25.Rg4 Kc8 (272.048.134) 4597 TB:1.109.873

After 14.Rxd4 she does show a better bound of mate in 21 or better. Now I'll go back a move and ask her if she can find a better alternative, or if the PV of Femme Fatale played the best move.

40/03 0:08 +M22 13...Bb4 14.Rxd4 Bc5 (28.074.911) 3365 TB:371.104

13...Bb4 was best, moving back to move 12...

40/03 0:01 +M23 12...Kb5 13.Re8 Bb4 (4.229.736) 3266 TB:39.141

12...Kb5 was best, moving back to move 11...

40/05 0:08 +M24 11...Kc5 12.Rc8+ Kb5 13.Re8 Bb4 (37.716.820) 4211 TB:239.667

12...Kb5 was best, moving back to move 10...

40/55 1:03 +132.70 10...Ke6 11.Re5+ Kf7 12.Rxe1 Ra5 13.Rb7+ Kf6 14.Be5+ Rxe5 15.Rxe5 Kxe5 16.Rxg7 Kd5 17.Rg6 Kc5 18.b4+ Kb5 19.Rd6 Kc4 20.a4 Kc3 21.Rxd4 Kxd4 (284.805.025) 4484 TB:2.185.544

Bingo! Harem Girl can't see a mate from here, so Ke6 is an improvement. What does Femme Fatale sees after it is played?

32/45 0:15 +M167 11.Re5+ Kf7 12.Rxe1 Rd7 13.Rb6 Ra7 14.Be5 Re7 15.Rf1+ Kg8 16.Bxd4

She does see a mate, poor lowerbound.

I'll be playing the moves one against the other until Harem Girl shows a better bound.

40/55 1:03 +132.69 11...Kf7 12.Rxe1 g5 13.Be5 (289.581.638) 4531 TB:3.170.358

32/02 0:00 +M166 12.Rxe1 Rd7 13.Rb6 Ra7 14.Be5 Re7 15.Rf1+ Kg8 16.Bxd4 Rd7 17.Rb8+

40/11 0:54 +132.71 12...Ra5 13.Rb7+ Kf6 14.Be5+ Rxe5 15.Rxe5 Kxe5 16.Rxg7 Kd6 17.Rc7 Kd5 (225.792.115) 4125 TB:1.582.087

32/48 0:07 +M146 13.Rb7+ Kf6 14.Be5+ Kf5 15.Bxg7 Kg6 16.Be5 Rxa2 17.Rg7+

40/55 0:16 +M16 13...Kg6 14.Be5 Kf5 15.Rxg7 Rxa2 16.Bxd4 Ra1+ 17.Kd2 Rxe1 18.Kxe1 Ke4 19.Bc3 Kf5 20.Kd2 Ke6 (82.675.441) 5152 TB:1.330.495

Harem Girl does see the mate, let's see if she can find a better move for 12th black move...

40/40 0:49 +132.70 12...Rxa2 13.Rb7+ Kf6 14.Kb1 Ra8 15.Be5+ Ke6 16.Bxd4+ Kd6 17.Rxg7 Rd8 18.Rg6+ Kd5 19.b3 Rd6 (226.028.673) 4541 TB:1.741.898

She does, what does Femme Fatale say?

32/36 0:02 +M142 13.Rb7+ Kf6 14.Be5+ Ke6 15.Ra7 Rxa7 16.Bxd4+ Kd5 17.Bxa7 g6 18.Rg1

Still mating...

40/13 0:06 +132.71 13...Kf6 14.Be5+ Ke6 15.Bxd4+ Kd6 16.Ra7 Rxa7 17.Bxa7 g6 18.Rg1 Ke7 19.Rxg6 Kd7 (27.582.835) 4368 TB:457.303

32/02 0:00 +M141 14.Be5+ Ke6 15.Ra7 Rxa7 16.Bxd4+ Kd5 17.Bxa7 g6 18.Rg1 g5 19.Rxg5+

40/25 0:21 +M12 14...Kf5 15.Kd2 g5 16.Rf7+ Kg6 17.Rf6+ Kh5 18.Rh1+ Kg4 19.Rg1+ Kh5 20.Rf5 Ra1 21.Rgxg5+ Kh4 22.Rg7 Ra6 23.Bf6+ Rxf6 24.Rxf6 Kh5 25.Kd3 Kh4 26.Rh6+ (95.385.390) 4410 TB:2.142.665

Harem Girl sees the mate, I'm going back to see if she finds a better 13th move...

40/40 0:04 +132.74 13...Kg6 14.Be5 Kf5 15.Rxg7 (18.840.756) 4211 TB:577.563

And Femme Fatale says...

32/02 0:00 +M140 14.Be5 d3 15.cxd3 Kf5 16.Rxg7 Ke6 17.Kd2 Ra8 18.Bc3+ Kd6 19.Rh1

40/47 0:11 +M15 14...Kf5 15.Bxg7 Ra1+ 16.Kd2 Rxe1 17.Kxe1 d3 18.cxd3 Ke6 19.Rb6+ Ke7 20.Bd4 Kd7 21.Rf6 Ke7 22.Ke2 Kd7 23.Ke3 Kc7 24.Ke4 Kd8 25.Kd5 Kd7 26.Be5 Kc8 27.Kc6 Kd8 (50.293.754) 4481 TB:1.574.128

So I ask Harem Girl for another 13th move...

40/40 0:01 +M16 13...Kg6 14.Be5 Kf5 (4.245.182) 4157 TB:73.321

So I ask Harem Girl for another 12th move...

40/57 0:51 +M22 12...Rd7 13.Be5 g6 14.Rf1+ (240.999.882) 4698 TB:1.621.832

40/40 0:20 +132.69 11...Kd7 12.Rd5+ Ke6 (93.231.706) 4637 TB:419.453

This one for Femme Fatale...

32/41 0:04 +M148 12.Rxe1 Kc6 13.Rc8+ Kd7 14.Rce8 Rb7 15.R1e7+ Kc6 16.Rc8+ Kb6 17.Rxb7+

40/03 0:21 +M11 12...g5 13.Rg8 Rb7 (106.049.435) 4859 TB:589.530

We need a better 11th one...

40/40 0:42 +M24 11...Kf6 12.Rb6+ Kf7 13.Rxe1 Rd7 14.Be5 (211.930.617) 5027 TB:1.227.327

We need a better 10th one...

40/40 0:22 +M25 10...Ke6 11.Re5+ Kf6 (109.621.373) 4825 TB:1.494.348

We need a better 9th one...

30/03 1:30 +57.84 9...Bd8 10.Rd5+ Ke6 (419.762.209) 4630 TB:523.565

I stop the analysis at Depth 30, it's clear depth 40 would take unreasonable time.

32/51 0:23 +M163 10.Rd5+ Ke6 11.Rbxd8 Ra4 12.R8d6+ Ke7 13.Rxd4 Ra8 14.Rd8 Ra7 15.R4d6

Femme Fatale is not impressed...

31/04 0:12 +132.65 10...Ke7 11.Rbxd8 Ra4 12.Rxd4 (58.639.789) 4790 TB:178.926

32/36 0:00 +M160 11.Rbxd8 Ra4 12.Rxd4 Ra7 13.R4d6 Ra4 14.R8d7+ Ke8 15.Rd4 Ra6 16.Rd8+ Kf7 17.R4d7+ Ke6 18.Rd6+ Rxd6 19.Rxd6+ (589.132) 1275

40/57 1:26 +M18 11...Ra4 12.Be5 d3 13.cxd3 Ra7 14.Bxg7 Rc7+ 15.Kd2 Rb7 16.Bf8+ Kf6 17.R8d6+ Kf7 18.Rd7+ Rxd7 19.Rxd7+ Ke6 20.Rd6+ Kf7 21.Rb6 Ke8 22.Rb7 Kd8 23.a4 Kc8 24.Rg7 Kd8 (508.113.334) 5855 TB:3.373.157

We need a better 10th...

40/40 0:25 +M21 10...Ke6 11.Rbxd8 Kf6 12.R8d6+ (126.773.429) 4995 TB:1.830.374

We need a better 9th...

27/49 1:37 +62.26 9...Ra6 10.b4 Bd8 11.Rd5+ Ke7 12.Rbxd8 Rxa2 13.R8d7+ Ke6 14.Rxd4 g5 15.R7d6+ Kf7 16.Bxg5 Kg7 17.Re6 Ra1+ 18.Kb2 Ra7 19.Rd7+ Rxd7 20.Re7+ Kg6 21.Rxd7 Kf5 22.Be3 Ke4 (456.401.481) 4671 TB:603.931

Again, Harem Girl slows down to a crawl on move 9.

32/54 2:16 +48.37 10.Rd5+ Ke7 11.b4 Bc7 12.Bxc7 g6 13.b5 Rxa2 14.Rxd4 Ra1+ 15.Kd2 g5 16.Bd8+

Aha! This is a position where Femme Fatale doesn't see a mate anymore. Here I go back to the strategy of letting her play its top white and black move choices on the board until she announced mate...

10.Rd5+ Ke7 played...

31/02 0:00 +48.37 11.b4 Bc7 12.Bxc7 g6 13.b5 Rxa2 14.Rxd4 Ra1+ 15.Kd2 g5 16.Bd8+ Ke6 17.Bxg5

11.b4 Bc7 played...

31/56 0:43 +25.23 12.Bxc7 Ra7 13.Bg3 Kf7 14.b5 Kg6 15.b6 Rxa2 16.Rxd4 Ra1+ 17.Kb2 Ra5 18.c4

12.Bxc7 Ra7 played...

31/40 0:03 +M140 13.Bb6 Ra3 14.Bxd4 Ra6 15.Rg5 Rd6 16.Rb7+ Ke6 17.Rg6+ Kd5 18.Rxd6+ Kxd6 19.Rxg7 Kd5 20.b5 Kc4 21.b6 Kxd4 22.b7 Ke5 23.b8Q+ Kf6 24.Qe5+ Kxe5 (15.011.028) 3932

And it sees a mate again. What does Harem Girl say?

40/03 0:23 +M13 13...Rd7 14.Rxd7+ Kxd7 (117.993.596) 4999 TB:1.116.009

Much better bound. We need a new 12th...

40/40 1:16 +M19 12...Rc6 13.Bd8+ Ke6 14.Rxd4 Rc3 (376.917.589) 4938 TB:2.451.222

We need a new 11th...

40/03 1:46 +132.72 11...Bxb4 12.Rxb4 Rxa2 (497.532.322) 4687 TB:4.321.092

What does Femme Fatale say?

31/44 0:10 +M147 12.Rxb4 Ra7 13.Rbxd4 Ke6 14.Rd6+ Kf5 15.Rd7 Ra3 16.Rxg7 Ke6 17.Rg5

Still mate...

40/03 0:19 +132.72 12...Rxa2 13.Rb7+ Ke6 (83.947.351) 4257 TB:2.014.163

31/16 0:00 +M146 13.Rb7+ Kf6 14.Rd6+ Kf5 15.Rxd4 Ra1+ 16.Kd2 Rf1 17.Rb5+ Kg6 18.Rg5+

40/06 0:04 +132.73 13...Ke6 14.Rg5 Ra1+ 15.Kb2 Ra8 16.Rb6+ (17.178.864) 4127 TB:691.899

31/29 0:00 +M137 14.Rg5 Ra1+ 15.Rb1 Ra8 16.Rb6+ Kf7 17.Be5 g6 18.Rgxg6 Re8 19.Rg7+

40/31 0:11 +M16 14...Ra1+ 15.Kd2 Ra8 16.Rb6+ Kf7 17.Be5 Re8 18.Bxg7 Re6 19.Rxe6 Kxe6 20.Bxd4 Kf7 21.Rb5 Ke7 22.Rb6 (47.675.463) 4275 TB:2.356.277

And now Harem Girl sees it as well. We need better 13th...

40/40 0:01 +M17 13...Ke6 14.Rg5 Ra4 (8.402.447) 4286 TB:139.514

We need better 12th...

40/40 0:20 +132.73 12...d3 13.Rb7+ Kf6 (90.993.847) 4411 TB:1.680.255

What does Femme Fatale say?

31/29 0:00 +M141 13.Rb7+ Kf6 14.Rd6+ Rxd6 15.Bxd6 g6 16.cxd3 g5 17.Rb2 g4 18.Rg2 g3 19.Rxg3

Still mating...

40/05 0:16 +M15 13...Ke6 14.Rd6+ Rxd6 15.Bxd6 Kxd6 (73.716.682) 4588 TB:3.568.518

We need better 12th...

40/40 0:43 +132.72 12...Kf7 13.Rb7+ Kf6 (197.090.128) 4516 TB:2.596.604

Let's play Femme Fatale's moves into Harem Girl until Harem Girl shows a good bound...

31/20 0:00 +M146 13.Rd7+ Ke6 14.Rd6+ Rxd6 15.Bxd6 (625.838) 1381

40/09 0:04 +132.70 13...Kf6 14.Rd6+ Rxd6 15.Bxd6 g6 16.Rxd4 Kf5 17.Rd3 Ke6 (20.752.554) 4466 TB:459.109

31/30 0:00 +M140 14.Rd6+ Rxd6 15.Bxd6 g6 16.Rxd4 Kg5 17.Rd3 Kf5 18.Rg3 g5 19.Rxg5+

40/07 0:05 +132.73 14...Rxd6 15.Bxd6 Ke6 16.Rxd4 Kf5 17.Rd3 g5 (26.798.680) 4563 TB:619.010

31/02 0:00 +M139 15.Bxd6 g6 16.Rxd4 Kg5 17.Rd3 Kf5 18.Rg3 g5 19.Rxg5+ Ke6 20.Rg6+

40/51 2:25 +M12 15...Ke6 16.a4 Kxd6 17.a5 Kc6 18.a6 g6 19.a7 Kc5 20.Rb3 Kd6 (846.536.881) 5826 TB:21.719.732

And, there it is. We need a new 14th.

40/40 0:00 +M13 14...Rxd6 15.Bxd6 Ke6 (2.363.128) 3658 TB:43.703

We need a new 13th.

40/52 0:16 +132.75 13...Kg6 14.Rd6+ Rxd6 15.Bxd6 Kf6 16.Be5+ Kxe5 17.Rxd4 Kxd4 (71.963.520) 4436 TB:2.306.248

31/36 0:09 +M14 14.Rd6+ Rxd6 15.Bxd6 d3 16.cxd3 Kf5 17.Rd4 Ke6 18.Bg3 Kf6 19.a4 Kf5 20.a5 Kg5 21.a6 g6 22.a7 Kf6 23.a8Q Kg7 (36.750.460) 3923

Femme Fatale manages to find a better lowerbound by herself! But we need the one from Harem Girl to backsolve it...

40/03 0:11 +M12 14...Rxd6 15.Bxd6 Kf6 (68.896.369) 6073 TB:993.121

Like this one. We need a better 13th.

40/40 0:17 +M17 13...Ke6 14.Rd6+ Rxd6 15.Bxd6 Kxd6 16.Rxd4+ Kc5 17.Rg4 (101.581.955) 5737 TB:3.150.889

We need a better 12th.

40/54 0:58 +132.72 12...Ra7 13.Rg5 Kf7 14.Be5 (265.310.352) 4521 TB:2.770.218

31/02 0:00 +M146 13.Rbxd4 Ke6 14.Rd6+ Kf5 15.Rd7 (1.293.266) 2288

So now I'd play Femme Fatale's move into Harem Girl until she shows a better bound, then go back and forward until she shows a mate score for the 9th move, and I go back to the 8th, and so on.

By this point it should be clear what I'm doing. After I'm done Harem Girl would be able to show a mate score from the root position, and since she'd store it on her learning file, she'd show it immediately even after I unload and reload the engine in the GUI.

The reason I have to reach depths below 2 minutes is that to backsolve a score I'd need to at least reach this depth everywhere, which becomes a waste of time with more depth.

Ovyron wrote: ↑Mon Feb 10, 2020 1:41 pm

Zenmastur wrote: ↑Mon Feb 10, 2020 9:52 am Could you do a small test analysis for us? Something not too difficult like analyze this position to mate:

[d]2b3k1/r2r2p1/p1N1q3/1B6/3p4/2b1B2R/PPP5/2K4R w - - 0 36

You mean some mate score as shown by the engine, right? Not the shortest mate, which I don't see the point of doing...

I assume he wants a full mating line, preferably one that is at least arguably minimal.

Ovyron wrote: ↑Mon Feb 10, 2020 1:41 pm

Zenmastur wrote: ↑Mon Feb 10, 2020 9:52 am Could you do a small test analysis for us? Something not too difficult like analyze this position to mate:

[d]2b3k1/r2r2p1/p1N1q3/1B6/3p4/2b1B2R/PPP5/2K4R w - - 0 36

You mean some mate score as shown by the engine, right? Not the shortest mate, which I don't see the point of doing...

So for this I'll be using 2 different engines, instead of some boring Engine 1 and Engine 2 let's nickname them:

Femme Fatale: This engine is really good at finding mates at low depth. However, it'll show some insane lowerbounds, like "+M160", which stands for "I can guarantee to win this position, and I can guarantee to do it on 160 moves or less." That she does, she'll decrease their counter with every move, but not with more depth. But I don't think you'd be happy if I show a mate score of 200 for this, would you?

No I wouldn't, as that defeats it's purpose. I could have given some other type of problem but a mate seem best because to get it close to it's actual depth I know about how deep you have to search. Think of it like this, if this were a non-mate position and you had an advantage of 1.20 pawns and this wasn't enough to secure a win, just a draw, you would be looking for every centi-pawn you could find in the position with the idea of driving the position to a win. In this position the win isn't in doubt. Ok, maybe in JP's mind there is some doubt, but for the rest of us it's pretty much a sure thing. Lets say if you miss the length by 5 moves or more you get to defend 1.g4 against me and if you are within 4 moves I'll defend 1.g4 against you! But you have to give a complete line to mate so we can check it. Deal?

I looked at your analysis briefly but it's late (6:15 am) so it will have to wait until tomorrow.

Regards,

Zenmastur

zullil wrote: ↑Mon Feb 10, 2020 2:31 pm

Ovyron wrote: ↑Mon Feb 10, 2020 1:41 pm

Zenmastur wrote: ↑Mon Feb 10, 2020 9:52 am Could you do a small test analysis for us? Something not too difficult like analyze this position to mate:

[d]2b3k1/r2r2p1/p1N1q3/1B6/3p4/2b1B2R/PPP5/2K4R w - - 0 36

You mean some mate score as shown by the engine, right? Not the shortest mate, which I don't see the point of doing...

I assume he wants a full mating line, preferably one that is at least arguably minimal.

It doesn't have to be exact. Within 4 moves is good enough. You shouldn't have any problems with this as it was meant to test his hardware and analysis skills in combination. Big hardware should have no problems with it.