Interfacing the Novag Citrine

Karmazen & Oliver · Post by **Karmazen & Oliver** » Sat Oct 25, 2008 5:41 am

sje wrote:
bob wrote:however, that is a disappointing result. This is a trivial ending for a computer, if you have two key pieces of evaluation information:

(1) losing king has to be driven to the edge of the board;

(2) winning king has to come to help mate the king by getting close.

Amir Ban and I discussed this opening many years ago in r.g.c.c. He thought it was easy while I thought it would be nearly impossible to win without an endgame table. I tested it with crafty and at 35K node per second speeds, at 1 second per move, it had no problem forcing mate. It didn't play perfectly according to endgame table best moves, but it always won without reaching a repetition or 50 move draw. I played several hundred games with craftyA . 1 sec per move, no EGTB, craftyB using (at the time) your endgame tables. I was really surprised at how easy it was for the computer...
I have some ideas here concerning the Citrine, but they are only speculation; we really need Dave Kittinger here to comment authoritatively.

1) The Citrine node frequency in the endgame is about 8 KHz, and that's a fairly strong limit on effective analysis depth in a full width searcher. Its H8/300 16 bit CPU runs at about 10 MIPS, so there's about 1,250 instruction executions available per endgame node. Are these enough for extensive, class specific, endgame play?

2) I assume that the Citrine, like its Super Constellation grandpappy, uses significant resources (code and time) with root level prepreprocessing. But what works well in the middlegame may not always be as effective in the endgame.

3) The Citrine program ROM space is limited to 60 KB, and 24 KB of that is consumed by the opening book. That's not a very big book by desktop standards, and to make it smaller to include specific endgame class recognizers might have been too risky. On the other hand, I believe that the Citrine has at least a KBNK recognizer as the machine does very well with that class.

My experience so far with the Citrine is that it has been optimized for play against humans at the expense of strength vs other programs. This is not a bad decision given the target audience. I'm guessing that it would score about fifty percent in a long match with a 2200 elo human, but not as well against a typical 2200 elo program. The marketing claim of 2330 elo is an advertising fantasy typical of the commercial chess computer field.

It's still a fun machine, though. I'm glad I bought one and I recommend it to others.

OK, I agree. I play some games versus novag star diamond, and some fast games versus novag citrine,

My experience is very rare, A lot of people say that star diamond is a very strong program, have more ram, have more rom, have more hash-table, but I play more easy versus star diamond, that versus novag citrine... star diamond have the way to thinking that a computer do... and his big book can be a problem too... ok, I do some 0.5 points and I do some victorys on kings Indian or king gambit, and on exchange spain variatons do easy victorys...

but with Ctrine... there is other world, ok it have some problems in the ends games, similar all computers... but have some "idea" on middle game and in open game that are very surprising for that "small" program and with that CPU...

I think that novag do a "impressive compression of the program for that cpu, giving a group as a result more cheap.

but I also think that some evaluations of previous computers novag, are bad for style, in the levels semi fast and gast, of 5 plys to 11 plys, citrine is very human, is surprising that he/she finds very logical plays and although in some occasions it cannot support those long term good plays, the idea this present... it is very quick thinking good variations.

All that this well, but I don't have clear which is the program that novag installed in that it... I am surprised their valuations and their game level is "difficult" to classify for most of the fans. playing with sincerity, without looking for traps, the citrine game is a lot but human and realist that the style of star diamond.

I win star diamond more easy with a strategy anti-computers... Citrine go for other ways... and have some problems... because these ways are they are correct strategically... without having book of the Sicilian, it plays it perfectly or very good concepts ¡
and interesting test is to be developed without pressing the nuts to the program and without to give him basic objectives and to see like it goes making decisions, it doesn't try to occupy the center massively, it maneuvers in 3 lines and this dresser with the positions of 0.5-0.5... it is surprising. for a computer.

( the other problem for human is that CITRINE... another curious novelty, is that citrine doesn't care to make 0.5-0.5 points... it won't force the situation, and until the search in many moments. )

The style of CITRINE is very human, more human that others computer novag.

I believe that some algorithms have been omitted that harmed the natural selections of the program in versions like in star diamond...

CITRINE style is better ? why ?¡

Steve B · Post by **Steve B** » Sat Oct 25, 2008 11:39 am

Excellent points Oliver
to be perfectly honest i think all of Kittingers programs for Novag have been basically the same for many years
perhaps from the Introduction of the Constellation Series in the 1980's
but there is something different about the Citrine
it can play silly moves in fast time controls that lose the game outright and then play very solid deep moves
i dont recall seeing this sort of wild divergence in move quality in other Novags

you might very well be right that Kittinger modified the Citrine program and removed certain algorithms
it might be interesting to put the Citrine through one of the BT test suites to see how it stacks about against other tested Novags

Regards
Steve

Steve B · Post by **Steve B** » Sat Oct 25, 2008 1:42 pm

Steve B wrote: it can play silly moves in fast time controls that lose the game outright and then play very solid deep moves
i dont recall seeing this sort of wild divergence in move quality in other Novags

an example of erratic play...
Le Citrine is black

1. d4 d5 2. c4 dxc4 3. Nf3 Nf6 4. e3 Bg4 5. Bxc4 e6 6. h3 Bh5
7. O-O Nbd7 8. Nc3 Bd6 (so far all book in the Queens Gambit Accepted..)
leaving Le Citrines book we have..
9. Re1 O-O 10. e4

[d]r2q1rk1/pppn1ppp/3bpn2/7b/2BPP3/2N2N1P/PP3PP1/R1BQR1K1 w - - 0 11

here on the Blitz Game/5 minute level ..Le Citrine essays forth the honker 10..a6??
allowing of course 11.e5 and the win of a piece

dont see this move on the Sapphire I (as an example) which plays 10..e5 at the same level

Bi-Polar Regards
Steve

Karmazen & Oliver · Post by **Karmazen & Oliver** » Sun Oct 26, 2008 7:15 am

Steve B wrote:Excellent points Oliver
to be perfectly honest i think all of Kittingers programs for Novag have been basically the same for many years
perhaps from the Introduction of the Constellation Series in the 1980's
but there is something different about the Citrine
it can play silly moves in fast time controls that lose the game outright and then play very solid deep moves
i dont recall seeing this sort of wild divergence in move quality in other Novags

you might very well be right that Kittinger modified the Citrine program and removed certain algorithms
it might be interesting to put the Citrine through one of the BT test suites to see how it stacks about against other tested Novags
Regards
Steve

Ok, some humans players are very confused with the style citrine, it can play some moves "bad" ( some problems with clock program for time-move or not mem reset previus serius game), and after this, find some ideas ...

that this confuse the human player, usually more pressure of time, and comment errors to finish off the game.

I always hear said for a lot of people: "citrine was lost, but I make a mistake the last..."

I always say that the games are not won until the game don't finish. it doesn't serve as anything that that it was better, but...

I think that this confirms that we are before a "onlooker" and strange program, CITRINE... anything conventional... let us see:

http://www.schachcomputer.info/html/body_bt2630.html

in BT-2630 compare computers novag:

Novag Citrine: 1974.
Nnovag Star Ruby-Obsidian: 1975.

I think that we only look this test, we can think that There are the same program... in fact... these 3 novag computar solved the sames problems ¡ and in similar time.. ( only one problem, number 22: this gives place to suspect something)... chances of the life, the solution is "a3", the same thing that plays citrine in yours example, XD, but this comment is irrelevant...

Ok. until here all correct one.

the surprise comes when analyzing the test BT-2450:

http://www.schachcomputer.info/html/bt2 ... ungen.html

Novag Citrine: 1997. ( ~ 2000 point )
Novag Star Ruby-Obsidian: 1930.

???¡¡¡

For example Saitek Cougar, Novag Shappire I, Novag Diablo, have similar results in Bt-2450 and Bt-2630...

I hope not to put the paw with the conclusions...

1- I think that Enginers of novag do a fantastic work "insert" the program citrine in that board ( a lot but cheap that the complex shappire design, I and II and star series ), in citrine all are in one chip... ¡

2- I think that they have used other test to evaluate the computers, and have they given a lot of importance at the real level of game, for example, does the hardware of other last computers perhaps allow some betters results in the test BT-2630, more spectacular ¡? but be let us realistic, who lets think to computers during more than more 5~7 minutes for play in one game ?

When we compare the computers-middle level with the test BT-2630 we don't see the differences for that. These test has too complex problems for the level of those computers dedicate, all seem to have the same level, because they solve few problems, the more easy, (the easiest), (11 problems in both cases, SR,Ob,Cit)

On the other hand, the test BT-2450 is more realistic, for that level computers, in function of the use that will give to those computers, they are tactical solutions with a more short horizon problem, and then we can see the "notables" differences among the programs... to pass from 2000 ELO to 1930 is a notable differs...( it is of supposing that in quick games Citrine has very good results against those computers, in fact, citrine versus CT15.2 on palm 130, in rapid games, Citrine have a good result when the games are < 30 min, example 15 min, if more time, CT15.2 good betters results... of course... have more ram, a big program, the surprise is that in fast game win citrine... some games.)

I think that that difference comes given ... because the citrine program runs in a single chip, CPU+RAM+ROM, all in one encapsulated, with quicker accesses to memory, certain it is that it is not a lot of memory, and for that reason their results in BT-2630 don't improve, but we return to the main objective, the destination of that computer is the fan-ches-human, and these persons won't allow to think so much time to it and they neither the one interests one computer very expensive of price.

therefore, my conclusion is that the program of CITRINE works very well in the levels that citrine has to work well. (rapids and semi-rapids.) and well in normal game... LEVEL~price.

Unfortunately not everything is advantages, due to the classic habit of novag dont ´do severe test to some models, Citrine has some it misleads, but these they are caused for that "the internal" clock ( dedicated for thinking) doesn't work too well, difference is also noticed when many games are played without having restarted the computer ac-dc`power.

Another of the remarkable characteristics is that it doesn't make lose the time to the human player, it´s good for rhythm...

for example: if this in a quick answer level, "C3" for example, and the human takes more time in responding, CITRINE will move quickly if the human's play is the move was thinking for citrine, ( this is important, because if citrine play similar human, ( is not better versus other computers, but is very good for this level versus human, beacuse have more time, or his evaluations have more plys), (that thought in the time of the contrary-player one makes it very dangerous for any human player, because in fact this playing with your time, putting more pressure to the human player, in the other hand, when others computers are thinking in ours time, using its time for move too ¡, they also use the time that "supposedly" this dedicated to its move-play, making the game more slow or boring for human, citrine don´t do that.)

I believe that the citrine program is not in any previous computer, the differences among the test BT-xxxx demonstrates it, possibly to convert the program to the available size, some algorithms have been eliminated, a book more small that SD, and this gives an answer as a result more lives and energetic in fast games, ( nodos/seconds move around 4500 n/s ~ 5.500 n/s and 11.000~12.000 in end games, );

I also believe that other evaluation scales have been added, (+algorithms), simply for that I cannot believe (MYSELF) that a computer of that intermediate-level finds some "hypermodern" plays-moves in the treatment of some openings, as the Sicilian. ( any computer novag previus, (example star diamond, saphire II) ) I was never surprised so much them ...

their errors make citrine human, and their peculiar programming makes it only, surprising.

long live for Novag. CITRINE

http://img511.imageshack.us/img511/8083 ... ramyd5.jpg

if somebody needs the perfection, they already knows where it can look for it, ARENA GUI + CITRINE board + Deep Fritz 11. or Rybka... but which that is had a good time here ?

bye. from spain. Oliver

Karmazen & Oliver · Post by **Karmazen & Oliver** » Sun Oct 26, 2008 7:53 am

Steve B wrote:
Steve B wrote: it can play silly moves in fast time controls that lose the game outright and then play very solid deep moves
i dont recall seeing this sort of wild divergence in move quality in other Novags
an example of erratic play... ( for ply max: 5 ... are you sure ??? )

Le Citrine is black
1. d4 d5 2. c4 dxc4 3. Nf3 Nf6 4. e3 Bg4 5. Bxc4 e6 6. h3 Bh5
7. O-O Nbd7 8. Nc3 Bd6 (so far all book in the Queens Gambit Accepted..)
leaving Le Citrines book we have... ( are you sure ? I don´t think that is a book move )
9. Re1 O-O 10. e4

[d]r2q1rk1/pppn1ppp/3bpn2/7b/2BPP3/2N2N1P/PP3PP1/R1BQR1K1 w - - 0 11

here on the Blitz Game/5 minute level ..Le Citrine essays forth the honker 10..a6??
allowing of course 11.e5 and the win of a piece

dont see this move on the Sapphire I (as an example) which plays 10..e5 at the same level

Bi-Polar Regards
Steve

OK. I don´t think that... I like your example a lot...

That problems similar example you say in that example, maybe it can have relationship at the time that CITRINE without being disconnected (reset previus a good game ) or to the control of time, ( I think that exits some defect on how use his time citrine, ( power ac-dc out, reset memory, hash, clocks...) there is something imprecise in that, when we play some games... for example the level "A2" is much worse (bad) that "C3", C3 is a good level fast..., it is necessary to find the ideal point of thought of Citrine,

( in your example, if we select a time that only permit ply 5, citrine do a6, this is a problem, ok, but it is not less certain than we cannot demand a lot of precession in depth 5, , when CITRINE arrive to ply 6, don´t play that, Cb6 is the first option... on levels "b4" and "c3", ( in C3, all is correct, and in C2 ot A2, only ply 5, play that error ( error for who ? only ply 5 ..., but this is necesary for a player that are play versus citrine and only have "C2" level... human handicap ?...

in occasions toward half of the game, it is good to restart the clocks, ( to convert a "C3" level a good level for all game, ( go setup, select other time, and go out setup and go in and select the first level again), because Citrine begins to make strange things in some moments, I believe that that is some errors in the programming of the levels and the time of game, of the clock, but I think that this doesn't have anything to do with the algorithms... ( I test that positions on diferents levels )

let us see this interesant example,

I test YOUR POSITION with Fritz 9 GUI and Fritz 9 ENGINE and ply similar citrine with that time:

when I select "ply max 5" Fritz 9 move with white "e4" and I press space, and F9 thinking with black and move f8-e8 ( Re8 )??? ¡¡¡ (I can assure you that I gave a sound laughter, XD)

it is not the first time that I find similarities in the game, when I force to the engines to think of the same depth

Fritz 9 engine (castrated to ply 5) move f8-e8 and it is thinking that white move in the next c1-g5 ( Bg5 ). F9 value position -0.02, when I move e5¡ the value change +1.96 for white ¡ ( the others options are c6 ??? and Cb6, but after two moves more, permit e5 too ¡ )

see you:

other engines lies in the analyzed depth or do other moves, Fritz 7, DeepFritz 7, and Fritz 8 don´t permit e5, in ply:5, Fritz 8 permit e5 on level ply:3... suprise too ¡

(rybka, CT and all engines not permit thinking only real ply:5, or a ply value similar novag citrine)

Good, if we keep in mind that Fritz 9 falls in the same trap that Novag CITRINE, to the same analysis depth, we can conclude that the problem is not a "BUG" in the algorithms, but rather CITRINE has not had time of discarding that play-move.

BIG
http://img373.imageshack.us/img373/6091/ply5sm9.jpg

in fact, it doesn't play that move in other levels. ( it´s not a citrine book move, excuse me, but it´s not...

)

a very interesting example... XD.

re-Time-problem Regards
Oliver.

Steve B · Post by **Steve B** » Sun Oct 26, 2008 11:50 am

hi Oliver
two interesting posts
i think you have asked me a question about Le Citrines book?
what i said was that these moves are in the computers book..

1. d4 d5 2. c4 dxc4 3. Nf3 Nf6 4. e3 Bg4 5. Bxc4 e6 6. h3 Bh5
7. O-O Nbd7 8. Nc3 Bd6

and the moves after 8..Bd6 were now out of book

the moves leading up to and including 8..Bd6 are also in the Sap I's book

i think you also asked if the playing level i used was Sudden Death ..ENTIRE game in 5 minutes?
if so yes
at this time control Le Citrine manages to search up to 6 ply but does not complete a full 6 ply search on average

sorry if i was confused and if you did not ask those questions

the point i was trying to make with the Sap I example was that the Sap I does not even consider a6 as a move choice .. even from a start up position ..it ponders Nb6 and then chooses e5
i could test this with any number of Novag computers because i have almost all of them
this is a position i play quit often and to the best of my memory i do not recall any Novag playing a6 on Blitz 5 or on any level
i use this as an example to show that there is indeed something different in the search extensions with Le Citrine which i think agrees with your earlier post?
Although this would be an example of a difference that makes the computer play weaker i imagine that there are different positions where the program changes make it play stronger
of course if i raise the time control Le Citrine does not choose a6 and instead chooses a different move such as c5

Best Regards
Steve

Karmazen & Oliver · Post by **Karmazen & Oliver** » Sun Oct 26, 2008 7:04 pm

Steve B wrote:hi Oliver
two interesting posts
i think you have asked me a question about Le Citrines book?
what i said was that these moves are in the computers book..

1. d4 d5 2. c4 dxc4 3. Nf3 Nf6 4. e3 Bg4 5. Bxc4 e6 6. h3 Bh5
7. O-O Nbd7 8. Nc3 Bd6

and the moves after 8..Bd6 were now out of book

XD, excuse me, steve. My Citrine is a UCB ( don´t pacht, I didn't want that repairs me the citrine, it is an amusing faulty unit... XD)

I did this:

I test that position with setup positions, and in this case, citrine doesn't use the book although he/she has that position in the same one.

In that situation, when we use the level "C2", rapid game (5 min), use citrine uses a little bit less than time, because it has not saved seconds in the first plays-moves-book, for that reason it doesn't arrive to depth ply 6, only ply 5 when I test from setup positions.

The diferent is in book, when I play a game from initial position, the book of my citrine are "some bugs", when white move "Kc3", with black Citrine move ( book-rapid-move < 1 second ), BxKf3, and go out ours test positons, after this play others moves and after 2 o 3 lost again one piece, the pawn b7, lost, and lost too piece when move white "e5"... the book is very break in my Citrine, but...

I like to find those errors in the book, when I find some, I makes the following thing, I move-back-citrine-book-Bug, and I enter in the setup-mode, when I leaving the setup, the game will no longer follow any book, and CITRINE will think for it´s same... It´s a lot of funny, because the errors in the book ...

...these errors usually happen when they (novag) has been tried to personalize the selections of Citrine, surprisingly, even in quick-rhythms-games the citrine selections are very good, ( setup-not-book...), for that reason I think that in computers with a BIG book, as star diamond or shappire II, the book can be a problem if it is not very good, almost it is better than he/she thinks the computer for if same...

(in the pc engines, this also happens, but at another much higher level.)

Steve B wrote: the moves leading up to and including 8..Bd6 are also in the Sap I's book
i think you also asked if the playing level i used was Sudden Death ..ENTIRE game in 5 minutes?
if so yes

OK, in my case, in all levels Citrine move BxKf3 ... the book is bad after this, because lost pawn b7, and permit e5 too... ( I think that Novag cut-out this ... novag cuts that branch in the patch applied to the book )

Steve B wrote: at this time control Le Citrine manages to search up to 6 ply but does not complete a full 6 ply search on average

And I think that exits some diferents in time/move when we play a game from initial positions, and some moves are in the book and it´s diferent that insert position in setup... ( in level C2, citrine will use less time...)

OK, in my situation I can not test that, because my citrine have other selection book and in setup mode, citrine has not saved time in the first plays of the book, for that reason in the level C2 uses less time of analysis in that position after insert in setup-mode.

anyway the answer is very quick to know up to where citrine, the information in the LCD has goes changing and it is possible that it changes move-play and we don't see it... the exact ply ( see all information on LCD takes 10~15seconds)

Steve B wrote:sorry if i was confused and if you did not ask those questions

OK, don´t problem... some language complications exist.

Steve B wrote: the point i was trying to make with the Sap I example was that the Sap I does not even consider a6 as a move choice .. even from a start up position ..it ponders Nb6 and then chooses e5

ok course...

Steve B wrote: i could test this with any number of Novag computers because i have almost all of them
this is a position i play quit often and to the best of my memory i do not recall any Novag playing a6 on Blitz 5 or on any level
i use this as an example to show that there is indeed something different in the search extensions with Le Citrine which i think agrees with your earlier post?

OK, indeed their example is very developing, we can conclude that CITRINE has some evaluation algorithms very different to other novag computers ... the question es ¿are they are worse? I think that not, that in quick rhythms that programming is very effective, human, the problem is that position is the time... clock.

another question es: ¿ is an error in the algorithms? I think that not baceuse Fritz 9 engine, moves the same play exactly when we put like limit ply:5. and ply:6 ¡ too¡

Interfacing the Novag Citrine

Re: KQKR endgame

Re: KQKR endgame

Re: KQKR endgame

Re: KQKR endgame

Re: KQKR endgame

Re: KQKR endgame

Re: CITRINE similar Fritz 9 and ChessTiger 2009 on ply 4-5-6

Re: CITRINE similar Fritz 9 and ChessTiger 2009 on ply 4-5-6

Re: Interfacing the Novag Citrine

Re: Interfacing the Novag Citrine