Chess2u
Would you like to react to this message? Create an account in a few clicks or log in to continue.

Chess2uLog in

pollthe question is on top of the post and not here

(from the highest elo) A B F (to the lowest elo)
3
43%
(from the highest elo) A F B (to the lowest elo)
2
29%
(from the highest elo) B A F (to the lowest elo)
1
14%
(from the highest elo) F A B (to the lowest elo)
0
0%
(from the highest elo) B F A (to the lowest elo)
1
14%
Total Votes:
7

descriptionimportant facts and useful knowledge on engines and tourneys Emptyimportant facts and useful knowledge on engines and tourneys

more_horiz
poll question:


what do you think will accrue if engine A & F have hundreds of matches when engine A (elo 4000), engine B (elo 3900) and engine F (elo 2000) are on the list; do you think engine A's elo will go up or remain same or go down when A & F have hundreds of matches and the result of A Vs. F is 200 wins, 1 loss and 10 draws for A? what's gonna be the ranking after hundreds of matches between A & F (if you know how elostats work and how the elo is calculated based on which formula, you know the answer); what's gonna be the order of engines (the ranking) from the strongest to the weakest? (I'll post the answer later, but we can find the answer before)


***


stockfish / 1 ply search

I found out a bug or a weird symptom for SF engines

when SF is fighting with other engines

I can pause the tourney in winboard

winboard 4.70 is convenient cuz I can pause the tourney anytime
to do something else

then when I continue (un-pause) the tourney

only SF's search depth becomes 1 ply
(only when I watch a movie or I do some other things
which are CPU or RAM intensive)
then SF loses
since it can only search 1 ply ahead


***


elostat can't run .doc file

I just saved a plethora of matches in .doc file
the elostat 1.3 in Arena couldn't calculate for the rating list

perhaps cuz of the format and added functions of .doc file

so I saved the pgn file with wordpad or notepad
which don't support the fancy functions

then now the elostat can run the pgn well


***


elostat
having matches with weak engines is not good

I found out something extroardinary about elo calculating

I think there are 2 prominent elo calculating methods

and for instance,


engines ------------------------- elo ratings

A ------------------------- 3500
B ------------------------- 3450
F ------------------------- 2600

if there are 3 engines with the repective elo ratings

currently A is the best
and B is only 50 elo behind A
and F is nowhere around A or B

what do you think will accrue if A & F have hundreds of matches?

some may think
since A will have 100 wins and 0 losses
along with 0 draws
(the actual result should be something similar
like 100 wins, 1 loss and 6 draws)

A can gain more elo
which will result in A having higher elo

some may think
no, that's not the case
since the elo diff between A & F is too large
so even if A wins hundreds of rounds over F
A can't gain elo
so A's elo will remain as it is
no elo change will occur

but the answer is actually A's elo went down a lot
so if the strong engine A fights many matches
with weak engines like F
A's elo goes down
that's what I found out


***


I heard bouquet is not based on ivanhoe
it is based on ippolit
that was recently on the nTCET chat box
and bouquet is fantastic!
it is on top of the tourney now!!!


***


opening book match(tourney)

I can't understand how opening book Vs. opening book tourney can be organized

book is something that is used by 2 engines

how one engine can use book A
while another engine can use book B?

moreover, how can we get one book is better than another book?

I don't understand how opening book tourney can be done

stronger opening book means the book only has the strongest openings hence little variety in openings?

weak opening book means the book has multifarious openings hence the book is good?

if engine A choses a move from book A
then engine B choses a move from book B
and then some
how can this go on?

how can 2 opening books fight with each other?

is it even possible?

I don't have any knowledge on this matter!

descriptionimportant facts and useful knowledge on engines and tourneys EmptyRe: important facts and useful knowledge on engines and tourneys

more_horiz
I also found out some more

2 plies means 1 move
and some ppl are confused of this

1 move ahead = 2 plies ahead!


***


engine's elo greatly depends on the TC
and number of cores used

when I run single core tourney
SF is always behind houdini
but when I run multicore tourney
SF's elo becomes higher by 20~40 points

perhaps it should have something to do with scailng issue
and search depth increase from SP to MP and efficiency


***


and during the winboard tourney
if we press "analysis" then
the match at the moment becomes null I think
and it will proceed to the next match

if SF's search depth becomes 1 ply ahead only
(mentioned in the above post)
and we wanna cancel that round

we can rescind that round which became incorrect

descriptionimportant facts and useful knowledge on engines and tourneys EmptyRe: important facts and useful knowledge on engines and tourneys

more_horiz
here is the answer to the question! thumb up 


what do you think will accrue if engine A & F have hundreds of matches when engine A (elo 4000), engine B (elo 3900) and engine F (elo 2000) are on the list; do you think engine A's elo will go up or remain same or go down when A & F have hundreds of matches and the result of A Vs. F is 200 wins, 1 loss and 10 draws for A? what's gonna be the ranking after hundreds of matches between A & F (if you know how elostats work and how the elo is calculated based on which formula, you know the answer); what's gonna be the order of engines (the ranking) from the strongest to the weakest? (I'll post the answer later, but we can find the answer before)


(from the highest elo) A B F (to the lowest elo)75% 75% [ 3 ]
(from the highest elo) A F B (to the lowest elo)0% 0% [ 0 ]

(from the highest elo) B A F (to the lowest elo)25% 25% [ 1 ] ------ the answer is this and the answer is surprising!


(from the highest elo) F A B (to the lowest elo)0% 0% [ 0 ]
(from the highest elo) B F A (to the lowest elo)0% 0% [ 0 ]

descriptionimportant facts and useful knowledge on engines and tourneys EmptyRe: important facts and useful knowledge on engines and tourneys

more_horiz

topic: why some engines are good at long TC but some engines are weaker at rapid?
or why engines have different ratings at short TC and long TC?
or why engines' ratings change a lot from short TC to long TC?

many ppl who watch nTCEC wonder
why some engines like hiarcs, rybka and junior are super strong
at long TC
but they become much weaker at shorter TC

the following is just my opinion and my obversation

I think those engines are slow searchers

so when they play short TC
rybka(R) searches for 15 moves ahead
when other engines like houdni(H) get 22 moves ahead

*

at short TC
if R gets only 15
while H gets 22
the diff is huge
even if the depth diff is only 7 (22-15=7)

the diff between search depth of 15 and 22 is huge

depth of 15 is much weaker than depth of 22 quality wise and eval wise

*

but at long TC
if R searches for 25 moves ahead
H gets 40 moves ahead

the search depth diff is 15 (40-25=15)
even if the depth diff is 15 that is huge
there is not that much actual diff between depth of 25 and depth of 40
quality wise and eval wise

once an engine's search goes over 20 moves ahead or so
the quality of the found move is superb already
the found move has the quality of the correspondent chess

that's why slow engines can do well in long TC

*

but when the TC is short

R searchs for only 15 (less than 20 or so)
when H gets 22
even if the searched depth diff is only 7

between the quality of 15 and the quality of 22
there is a huge diff in terms of eval and quality

the quality of 22 is that of correspondent chess
whereas the quality of 15 is only that of rapid chess

this is my observation and opinion
after watching chess engine matches for so many years


***
***


I'm using PCC March bin book and
this proved to be very good
for a long time
I thought of using a new book that is
PCC September bin book
you know they make and release the new book every month
when PCC March proved to be perfect
I will probably not use any newer books
since PCC March is just perfect
cuz if I use a new book
then I have to run thousands of matches
to prove that the new book is acceptable
that is
the book is not flawed
I have used so many different books
and some of them were only selecting the same or similar openings
again and again
so some engines are strong in those openings
and other engines are not strong in those openings
those engines which are good at those openings had higher elo ratings
and that's not accurate
also if I use those books which choose the same or similar openings
again and again
even if I run 1000 matches
it is like I only ran 600 matches
since many matches are overlapable
(the variety of the openings is low)
(some books choose similar or same openings a lot
I think cuz those openings are good and strong openings)
due to these reasons
from these grounds, if we use a new book
there are aforementioned side effects accompanied
so I'll stick with PCC March
that is just perfect
we know the old saying

if something is not broken, don't fix it

PCC March is very good
and I don't have to amend it  
or find a newer solution
since it is already perfect


***
***


this perhaps may be a useful tip
from watching many matches
I think that
when you open the engine parameter box in winboard
the left side is engine specific, engine native settings
whereas the right side is the polyglot setting
that is not engine specific but a general setting that is applied for all engines
so changing and experimenting with the left options
may be good and useful
since those options are originally designed for that specific engine
but changing the right side polyglot options
may be not useful
and perhaps it will only decrease the elo
since those right side options don't work with the given engine well in tendum in harmony
that's just my guess
after experimenting with many diff options
on both sides

descriptionimportant facts and useful knowledge on engines and tourneys EmptyRe: important facts and useful knowledge on engines and tourneys

more_horiz
Permissions in this forum:
You cannot reply to topics in this forum