back


30275David
2020-07-05 14:17:07
In the latest version of Komodo there are 25 levels and 5 styles suitable for playing blitz, people play in 3 '...
If you want you can also put Stockfish, but you have to pay the disc.
Pure commercial logic, but you could, on a Stockfish basis, create an engine, call it Houdini 7 and even sell it ...
Or can't you use the name Houdini 7?
30274David
2020-07-05 13:55:08
@hernandez
I mean that you can use Stockfish to study or for engines Tournaments, but, for exemple, it is not suitable to play against a human in a 3 minute game, (perhaps only the first 3 levels on 20 , but only for strong masters).
I have still to find a good engine for training, maybe the old Chessmasters Xi ( ubisoft) is still the best.
30273hernandez
2020-07-05 13:35:18
David @30271
Interesting! but would you elaborate a bit more the meaning of "...funniest and most comfortable" related to be the best program?
30272David
2020-07-05 13:05:58
but I'm not saying Komodo is comfortable.
30271David
2020-07-05 12:53:35
If, in the past the best program was the most competitive: I think about Kasparov's matches with Deep Junior or Deep Thought.
Today we are touching the tip of the parable of this trend.
Tomorrow the best program will be the funniest and most comfortable one.
Stockfish is strong but also "uncofortable".
30270how white wins?
2020-07-05 12:15:10
why tcec change links?
30269David
2020-07-05 12:08:50
@queen13
The best program is not necessarily the strongest one. If 99.9% of the world's population is unable to beat Fritz 1 on dos, then perhaps a weaker program is better.
30268how white wins?
2020-07-05 12:08:26
if k so bad
none buy they no money
if k #1 ahead stockfish
i not buy they no money
if release Komodo making it public
none buy they no money

no money wins?
no more business?
30267how white wins?
2020-07-05 11:54:23
not only engine develop
is business
again :)
:-p
30266how white wins?
2020-07-05 11:48:43
'why they wouldn't release Komodo.'
money?
'making it public'
no money?
30265queen13
2020-07-05 11:13:23
I don't understand why they wouldn't release Komodo.

Make it open source and see how that goes. They aren't going to compete with stockfish right?

30263 makes it seem impossible for them to compete.

Maybe making it public would get many people to look into it and improve it.
30264how white wins?
2020-07-05 08:26:13
stockfish c 50
komodo c -8
30263Mrert
2020-07-05 06:15:15
Komodo has not made much progress in the last 2 years roughly. And honestly it's not surprising.

A bit of history. Komodo was created by GM Larry Kaufmann and Don Dailey. Don sadly passed away in 2013 (leukemia). If you look at SF releases, one of the versions is SF DD in honor of Don Dailey. Then Mark Lefler joined Komodo the same year.
So there's only 2 guys mainly developping. Larry is 72 and Mark is about 62-63 (according to his graduation year), so not is their prime years.

Meanwhile SF's level skyrocketed (special thanks to ChessDBCN for providing such ludicrous computing power to test many ideas in a short timeframe).

@ken: they used C=50
30262ken
2020-07-05 03:43:00
what contempt was used by stockfish against komodo?
30261how white wins?
2020-07-04 20:46:40
komodo team must be shocked
30260Spook
2020-07-04 20:25:35
I claim that even 8 or more RTX 2080ti lost the SUFI18 to Stockfish. Why? 8 RTX2080 would never have seen more trained positions than 2 RTX2080 in 90 minutes. Lzero reaches its "intellectual" limits. Man, (stockfish developer) is superior to the nn.
30259Mrert
2020-07-04 18:10:39
@queen13 : If it was not new it wouldn't have been accepted for publication (or the reviewers are garbage).
I think you got it, it was a new way to apply neural networks to a class of problems (adversial games like chess for instance)
30258how white wins?
2020-07-04 16:43:39
http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=33583
lol
30257how white wins?
2020-07-04 14:14:13
i look for tothfish build on latest stockfish
not find yet
30256David
2020-07-04 12:53:52
...and I say this respecting Leela. Simply, its programmers will rethink and restructure its architecture.
She does her best.
30255David
2020-07-04 12:17:38
Undoubtedly a match between S12 and Alphazero, today, would be very popular, it would be "the match of the century".
But some companies move for money and not for sport nor to satisfy the public,nor to science.
Some companies ...
30254queen13
2020-07-04 10:50:16
@Mrert

So if I understand correctly, there's a new way (revolutionary?) to apply NN.

I know that NNs aren't new but the alphazero algorithm is and with this algorithm

a program can "learn" a variety of different tasks starting from "zero" just

with the knowledge of rules(if it is a game).

This wasn't possible previously even though NNs could learn in other ways.

It's true that I don't completely understand how the NNs work so maybe

the whole alphazero is just nothing new and just an application of what was already

know but with stronger hardware. I thought it was a pretty good breakthrough.

Deepmind created alphago which is better than all humans at go, something which

was expected to happen 10 years later.

I think NNs have huge potential but they need a very detailed structure and way to learn that

is yet to be discovered. After all the human brain is a neural network. A more complicated

and more nuanced, one which is very difficult to replicate.

So, is alphazero not a new algorithm to be used with NNs?

I think NNs have improved since they were first used because of better ways to use them

both because of structure and algorithms/ways to learn.
30253Lalit Mohan Bankoti
2020-07-03 22:46:08
STOCKFISH be aware for tcec season 19 also.
30252Mrert
2020-07-03 19:39:31
@queen13: Neural networks are an old concept aimed at mimicking our brain. Every neuron has an array of inputs. If those input are very small, nothing really happens. Then if the inputs are sufficient, the neuron will emit a signal depending on the input.
In an artificial NNs the output the weighted sum of the inputs (the famous tunable weights) go through the neuron's activation function (look at ReLu, sigmoid and heaviside). This output will be sent to the next layer of neurons.

But there are many ways to build such network (linear leayers, convolutive network, recurrent networks, ladder network...), and ways to learn (supervised learning, reinforcement learning...). So when a research team finds a new way to apply NN they publish a paper
30251WCN
2020-07-03 18:13:20
Congratulations to the Stockfish team! for the amazing Deep Tactics version of Stockfish :) for Winning TCEC 18 :)

I saw this coming some time ago as most people know and it's the End! of NN engines,as I wrote NN engines do a Wide search which is OK ? but it misses important Deep Tactics,whereas Stockfish does a Deep search which cuts through the Wide searches and finds the Truth! namely everything aims for the King and Checkmate! everything else does not matter!

So this leads to the Stockfish mottos...


{STOCKFISH's DEPTH DESTROYS (NN's) WIDTH}

{WHEN STOCKFISH's TACTICS SAYS MATE THEN IT'S MATE!}


https://groups.google.com/forum/#!topic/fishcooking/pEA6XRiDWHg%5B26-50%5D





A.R.B :)

next page >
single page