Fritz 15.0

[Rebel]

It seems to me fair Vasik Rajlich decides to continue with its work in Rybka.
But Fritz loses his personality.
Fritz 15 seems clearly a continuation of Rybka.

Pretty much what you may expect (the 63%) from an update.

[Laskos]

That is a problem for all rating lists* with pairs of close engines and one of them is no longer tested. On the other hand we are talking about very low 1 digit Elo numbers.

I don't know about that and since you speak for all rating lists it's certainly not true for old ProDeo versions. From a textfile I made 3 years ago, 3 years ago vs now gives:

Code: Select all

CEGT 40/20
ProDeo 1.0   2384 - 2349 = loss of 35 elo during 3 years
ProDeo 1.1   2453 - 2420 = loss of 33 elo during 3 years
ProDeo 1.2   2454 - 2423 = loss of 31 elo during 3 years
ProDeo 1.6   2462 - 2432 = loss of 30 elo during 3 years

No new games were played. Those are not one digit numbers

On the other hand CCRL doesn't seem to have such an aging problem.

Code: Select all

CCRL 40/40
ProDeo 1.1   2586 - 2589 = gain of 3 elo points during 3 years
ProDeo 1.2   2616 - 2617 = gain of 1 elo point during 3 years
ProDeo 1.6   2616 - 2617 = gain of 1 elo point during 3 years

No new games were played.

Hasn't CEGT moved from Elostat to Ordo? OTOH, CCRL stayed with Bayeselo. Also, sometimes the rating lists change calibration.

[IWB]

I don't know about that and since you speak for all rating lists it's certainly not true for old ProDeo versions. From a textfile I made 3 years ago, 3 years ago vs now gives:

Code: Select all

CEGT 40/20
ProDeo 1.0   2384 - 2349 = loss of 35 elo during 3 years
ProDeo 1.1   2453 - 2420 = loss of 33 elo during 3 years
ProDeo 1.2   2454 - 2423 = loss of 31 elo during 3 years
ProDeo 1.6   2462 - 2432 = loss of 30 elo during 3 years

No new games were played. Those are not one digit numbers

If there are no games played you of yourse cant see what I was talking about! I was not talking about the general tendency to go down in rating over time ... that tis something different.
*and as Kai mentioned, this might be because of Elostat to Ordo and some elo adjustments in general.

But I am pretty sure it it would be true for them as well - given my above conditions.
In your case this would be only for PD1.1 to 1.2 (as they are close together) then stop playing 1.1 after 4000 games as you determinded that 1.2 is ~2-5 Elo better and then play 12000 more games against mainly much stronger engines. At the end 1.2 might end up below 1.1 even if it is better in reality. I could actually watch it over the years in my list with Rybka, just no one cares. Now the topic comes up because of F15.

You you can see something similar in the CEGT with R4.0 4.1 on one core. But it is harder to see as you do not play the same openings with the same opponents ...

Anhow, it is marginal with a few 1 digit elos.


Btw: I give all individual results in my downloads. You can extract the same opponents for R4.0 and 4.1 and see by yourself.

[Graham Banks]

...
However, IPON is rating it only 28 elo above Rybka 4.0

This already come up at the german CSS fora but it is a bit missleading.
Some years ago I tested R4.0 and R4.1 with a couple of thousand games vs the same opponents and the same openings ...(IPON conditions) and R4.1 come out slightly ahead.
At that point the testing of R4.0 was stopped and only R4.1 was continued in my testing.
Now a lot of better engines came and decreased the rating of R4.1 while R4.0 rating was somehow "frozen". After a while R4.0 seemed to be better than R4.1 which in reality is not the case!

That is a problem for all rating lists* with pairs of close engines and one of them is no longer tested. On the other hand we are talking about very low 1 digit Elo numbers. No human can possibly "feel" that. I consider that for any practical reason as irrelevant. For all I know R4.1 as better than R4.0!

Bye
Ingo

*Maybe even more for some lists were not only the opponents are different, but although the openings and number of games against certain opponents, maybe hardare with a different time factor included ... various things which might add variability. In short, forget about a +/- of 10 Elo in rating lists and dont mix it with engine development ...

In CCRL 40/40, Rybka 4.1 and 4.0 are so close to call, that both still get tested.
Whereas in the 1CPU list, 4.1 is slightly ahead, in the 4CPU list it is the reverse.

[ThatsIt]

[...snip...]
Hasn't CEGT moved from Elostat to Ordo? OTOH, CCRL stayed with Bayeselo. Also, sometimes the rating lists change calibration.

That's right.
Here you can find our lists before we switched to Ordo:

http://www.cegt.net/

Best wishes,
G.S.
(CEGT team)

[MikeB]

Hi all,

Some have wondered if Fritz 15.0 would be a UCI engine that can work in any interface, or if it will just be an .eng file that works only in fritz gui. I asked that question to chessbase, and they replied simply that Fritz 15.0 "will be" a UCI engine.

I replied back to them asking if they'd just sell the UCI engine to the public for a lesser price (or at least have an option for that) and I hope they will respond back to me about that, although they haven't yet.

So at least we know it will be a UCI engine.

Sincerely,

Tim.

With current exchange rates - you can get Fritz 15 ( which allegedly supports up to 2048 cores ) via download from Chessbase without VAT in the US for ~ $62. That price seems pretty reasonable to me in light of all the "other " features to me. No I don't work for Chessbase either.

[A Distel]

Hi all,

Some have wondered if Fritz 15.0 would be a UCI engine that can work in any interface, or if it will just be an .eng file that works only in fritz gui. I asked that question to chessbase, and they replied simply that Fritz 15.0 "will be" a UCI engine.

I replied back to them asking if they'd just sell the UCI engine to the public for a lesser price (or at least have an option for that) and I hope they will respond back to me about that, although they haven't yet.

So at least we know it will be a UCI engine.

Sincerely,

Tim.

With current exchange rates - you can get Fritz 15 ( which allegedly supports up to 2048 cores ) via download from Chessbase without VAT in the US for ~ $62. That price seems pretty reasonable to me in light of all the "other " features to me. No I don't work for Chessbase either.

...and I suppose You don't work for ICGA either?

[John Conway]

Fritz 15 is obviously a new engine which Vas wrote from scratch after carefully studying Rybka 4.1 source code for 4 years.

[AdminX]

Fritz 15 is obviously a new engine which Vas wrote from scratch after carefully studying Rybka 4.1 source code for 4 years.

Looks like the only thing new is the name, and maybe some improvements added over Rybka 4.1. Note: I am speaking only about the engine and I doubt Vas started from scratch.

[MikeB]

Fritz 15 is obviously a new engine which Vas wrote from scratch after carefully studying Rybka 4.1 source code for 4 years.

That would be foolish to write from scratch. Rybka has a very good SMP implementation. Vas is obviously a very talented programmer and it is good to see him move on and put Rybka behind him. Wish him all the best.