03 Jun 09
Originally posted by Macpo1) you do a blundercheck with any threshold you want (0 perhaps?), and check out the amount of deviation.
Hi everyone
I just got Fritz 10. How do you do Fritz match up analysis? I couldn't find...
Thanks!
2) you use "infinite analysis" with a definite time interval, say 30 seconds, and check how what percentage of the main line moves match up with the 3 best moves suggested by the engine.
03 Jun 09
1 edit
Originally posted by MacpoTo quote myself:
Hi everyone
I just got Fritz 10. How do you do Fritz match up analysis? I couldn't find...
Thanks!
You do this by using engine move matchup %'s in many objectively chosen games against strong opposition.
You look at the suspect's moves once the game goes out of book; ie hasn't been played before on a big database such as the 4.3m www.chesslive.de.
Next, you need to set up your engine so that you can look at it's top 3 choices for each move in a strictly consistent time interval. I use 30 seconds per move because with the hash table this allows quite high level analysis on a decent pc, but also allows for practical analysis.
With an average game with 20-30 non-book moves the analysis takes about 1.5 to 2 hours including write-up. If you allow 60 seconds per move the analysis takes ages & I tried this with a few games & there's very little difference in the end results between the longer & shorter time periods.
There should be at least 20 moves in the games once they go out of book so that the end results aren't skewed by for instance a blunder then a forced win in a few moves.
The expected matchup %'s for both top pre-computer era CC human players & OTB super-GM's are known to the game mods, so anyone who consistently tops these stats is either an unknown genius who chooses to grace RHP with his play or an engine user.
The figures for top human play are about
Top 1 match = 60%
Top 2 match = 75%
Top 3 match = 85%
and they take into account obvious & forcing moves & so on.
There was an argument in this forum that various moves should be discounted from analysis, but the figures given above would then also need re-calculating.
If you need more help, just ask in this thread.
Originally posted by Macpo
Sorry, can you be more precise? Cause I got many lines that appear, but that's all! I don't know how to see if it's fritz's move, fritz's second move, fritz third move...
Thanks!
I don't know how Fritz X works, but with Fritz XI you just play the moves or copy & paste the .PGN & then (in the default setup) right click in the bottom right pane.
A panel should then pop up & in it you have options to increase or decrease then number of lines Fritz is valuing.
03 Jun 09
3 edits
Originally posted by philidor positionThis is my preferred option, simply because the expected parameters for humanly possible play are now well known because of analysis of top GM games which I (amongst others) have done.
...
2) you use "infinite analysis" with a definite time interval, say 30 seconds, and check how what percentage of the main line moves match up with the 3 best moves suggested by the engine.
If you use the top 3 / 30 second per move analysis method you must be strict in the timings & even if your move is 3rd & scores exactly the same as Fritz's top choice, you must class it as 3rd choice!
If you are sending as evidence for games mods then game ids, & the parameters under which the analysis was done are very important.
I always used to put this as a header to the analysis:
Fritz 11 @ 30 seconds per move
Pentium 4 2.93GHz 1GB RAM
Hash Table 192MB
Database used www.chesslive.de
I write the analysis on a jotter with the following format:
e.g
********
goes out on 18...Nd4
(White player name)
(19) 3 (20) 1 (21) 1 (22) N/A (23) 2
(24) 1 (25) N/A (26) 2 (27) 1 (28) 3
(29) 1 (30) 2 (31) 1 (32) 3 (33) 1
(34) N/A (35) 1 (36) 2 (37) 2 (38) 1
(39) 1 (40) 2
(Black player name)
(18) N/A (19) 1 (20) 2 (21) 2 (22) 1
(23) 1 (24) 1 (25) 2 (26) 1 (27) 1
(28) 2 (29) 1 (30) N/A (31) 2 (32) 3
(33) 1 (34) 2 (35) 1 (36) 3 (37) 3
(38) 1 (39) 1 (40) 1
Note:
'N/A' in the analysis above simply means that this move was not in Fritz's top 3 at the 30 second point.
I actually ring the move numbers, not bracket them. I find I can fit 5 moves into each line of an average sized jotter this way.
I have found this to be the easiest & most accurate way of adding up the totals.
It's important to have the paperwork ready in advance, so you can just note down next to the move number what position Fritz rates the move after precisely 30 seconds in this example, whilst keeping the computer moving along at 30 second intervals.
Once this is done & the analysis is complete, you can add the Fritz analysis into the .PGN & finish writing up the results:
[gid ]000000000[/ gid]
Result:
White:
Top 1 Match: 10/22 (45,5% )
Top 2 Match: 16/22 (72,7% )
Top 3 Match: 19/22 (86,4% )
Black:
Top 1 Match: 12/23 (52,2% )
Top 2 Match: 18/23 (78,3% )
Top 3 Match: 21/23 (91,3% )
03 Jun 09
1 edit
Originally posted by Macpoin infinite analysis, you should see a plus and minus button in the main engine window. whenever you hit plus (either the button or the plus key in your keyboard), fritz will add another line to it's analysis.
Sorry, can you be more precise? Cause I got many lines that appear, but that's all! I don't know how to see if it's fritz's move, fritz's second move, fritz third move...
Thanks!
the one at the top, with the best score for the current side (white or black) is the continuation of fritz's most preferred move, the 2nd line is the continuation of it's second preferred move, etc. so the first moves of the different lines are the alternative moves that fritz considers, which are listed in order.
I must tell you, blunderchecking method is a lot easier and much less time consuming. you just add up the deviations and divide them by the number of moves.
06 Jun 09
1 edit
thanks for answers. I had one more question: with the blunderchek method, how do you do to have three alternative lines? for I only have one, so I can only make thje match up with fritz best line.
thanks again!
And I did good with my game then : 52/72/84. 🙂 And still, I excluded equalities from it...
07 Jun 09
Originally posted by Macpoyou're welcome.
thanks for answers. I had one more question: with the blunderchek method, how do you do to have three alternative lines?
you don't. you only have the deviation from the best line. but it's all automated, so you don't have to spend ages going through the game manually and note everything down.
07 Jun 09
1 edit
what if someone is a pretty good player (2000-2100 RHP rating, say), and they know about engine police, so they just use Fritz for moves that require the most delicate of judgements? (maybe once or twice per game) can you detect that?
Also, wouldn't many endgame sequences just naturally match up?
07 Jun 09
1 edit
Originally posted by BlackampEndgames are difficult! It's a mistake to think that endgames played by "strong" players will have a high match-up with a strong engine.
what if someone is a pretty good player (2000-2100 RHP rating, say), and they know about engine police, so they just use Fritz for moves that require the most delicate of judgements? (maybe once or twice per game) can you detect that?
Also, wouldn't many endgame sequences just naturally match up?
For example, take a look at this game between Weyerstrass (#1) and Woodworm (#10): Game 4054880.
This is a very complicated position and there is no chance whatsoever that the moves played between these two would have a high match-up with Fritz.