Author: Robert Hyatt
Date: 14:10:14 09/03/02
Go up one level in this thread
On September 03, 2002 at 17:02:48, George Sobala wrote: >I have no axe to grind with either side. In fact, through these discussions >Robert Hyatt's explanations seem considered and rational, whilst Vincent seems >to be going off the deep end. > >HOWEVER there is absolutely NO WAY that the following table is genuine raw data. >The times in seconds are all far too-perfect divisions of the time taken by one >processor. Real life does not spew out such perfection. Let me explain. E.g. in >pos 1, for 4 processors note that 2830/832 is 3.4014423. This is really close to >3.4 - so much so that 2830/3.4 is 832-plus-a-bit. Work your way down the columns >and you will see that the time for x processors in seconds is far far far too >often just what you would get if you divided the 1-processor time by a divisor >with one decimal place (e.g. 3.4, 3.6, 3.7 etc) - not a second or two more or a >second or two less. Why shouldn't we frequently see times that would be the >result of 2-decimal place divisors, such as 3.43 or 3.58? I can't answer. I can only tell you two things: 1. The original log was raw data. 2. It was "eaten" by a program written prior to my dissertation which started in 1986-1987. I don't know what sort of "tricks" I might have had to play in doing that. IE some of the logs we "ate" were from Harry Nelson and Lazlo Lindner running test positions for ridiculous amounts of times. So that they would produce node counts that would blow the int values on the machines I used to eat them. how I fixed this I simply do not recall. Several have suggested equally probable ideas. However, without the code, I don't have any idea and guessing is not particularly useful. The only real point is that (1) is absolutely true. Whether the data in the table was "mangled" a bit by reading, converting, adding/dividing, and converting again, I simply do not remember. So, as always, "Caveat Emptor"... > >Conclusion: this is inescapably NOT raw data. (And if you want "proof", Dr >Hyatt, take it to a statistician at your university.) However, that does not >automatically mean that it is invented or spurious. For example, the original >data in seconds could have been recorded, divided out and rounded down to >produce ratios accurate to one decimal place, and then (with this raw data lost >or mislaid) for the purpose of the publication a "back-calculation" may have >been done using the rounded divisors. It really was not "lost or mislaid" so that isn't it. But whether my "log eater" did a few "conversions" that might have hurt the least significant digits, I don't remember. I do know that the "log eater" was really written for my dissertation, which ran on a machine that could search 100 nodes per second per processor. The node counts did not get real big. I modified it for Harry later and it may well have been that the data was mangled a bit in the process. I don't remember. But your first suggestion certainly didn't happen that I recall... However, there is something that is "tickling" here, so give me a while to look back thru my notes... that might help. More if I figure this out... > >On September 03, 2002 at 16:12:13, Robert Hyatt wrote: > > >>Here it is: >> >> >>First, times in seconds: >> >>pos 1 2 4 8 16 >>1 2,830 1,415 832 435 311 >>2 2,849 1,424 791 438 274 >>3 3,274 1,637 884 467 239 >>4 2,308 1,154 591 349 208 >>5 1,584 792 440 243 178 >>6 4,294 2,147 1,160 670 452 >>7 1,888 993 524 273 187 >>8 7,275 3,637 1,966 1,039 680 >>9 3,940 1,970 1,094 635 398 >>10 2,431 1,215 639 333 187 >>11 3,062 1,531 827 425 247 >>12 2,518 1,325 662 364 219 >>13 2,131 1,121 560 313 192 >>14 1,871 935 534 296 191 >>15 2,648 1,324 715 378 243 >>16 2,347 1,235 601 321 182 >>17 4,884 2,872 1,878 1,085 814 >>18 646 358 222 124 84 >>19 2,983 1,491 785 426 226 >>20 7,473 3,736 1,916 1,083 530 >>21 3,626 1,813 906 489 237 >>22 2,560 1,347 691 412 264 >>23 2,039 1,019 536 323 206 >>24 2,563 1,281 657 337 178 >>
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.