Author: Rolf Tueschen
Date: 02:41:41 02/21/03
Some time ago I read in CCC one of the many telling messages about SSDF and its technology. I don't like personal attacks and therefore I leave out concrete names. Since I have published my critic against the presentation of a "number one" people tried to defend the actual practice in SSDF. Here is one clear example for a terrible confusion: The argument goes like this: "You have number one in a lot of sports too. The person with most points at a certain time is presented as number one (the digit). It's only one that receives the gold-medal and it don't matter if (S)he wins with a thousand of a second or a whole minute. It may be sad but true." Analysis: Although the argument is not totally corrret because I know cases where 2 Gold Medals were awarded. But that is secondary here. I want to demonstrate why the actual practice of SSDF is outrageously false and why in other sports this is also respected! Demonstration: Perhaps some of you have seen swimming for instance. More than once I saw the following happen. With electronic time measurement you define beforehand how "exactly" the time should be taken for real. Say One thousands of a second. That is the last digit. Earlier with hand-made measurings the last digit was one tenth. Of course electronically it is more exact, but still this is not the end of the possible line (1 thousands). But you define this as the end. If you have equality then it makes NO sense to present a umber one. This should be clear. Take physics. You measure a stick of wood. You use a simple tool with millimeters as the last visible unity. Would it make sense to give results with a one thousands or even smaller digits? I don't think so. Now let's take statistics. You work with the famous standard deviation. And then you get a result with 1 point difference. The deviation is 30 points. Listen, such a result means that you ca be sure in 95% of all cases that your result is varying above the range of these 30 points. Now I ask you: If you have two "results" of a difference of 1 or say 18 points, could you say if the two results are the representation of a different value or is it all the same? [Perhaps my English is confusing this question, but it's the most important question!] In other words. A quantity of results in this space of 30 points around a value X could NOT be seperated [defined as] as better or worse, if the differences are 1, 2, 3, 4 ....30 points! Only with a difference of >30 [sic] you had the right to seperate two results! You remember the time measurement? >30 that is our time measurment up to one thousands, while above one thousands we have "darkness" like 1...30 in our stats results. You must always be aware of that the standard deviation defines the impossibility realm of exact data. Now it's foolish, so to speak, to give the exact standard deviation of results and then present a result that is impossible by definition! Here in SSDF when results differ by less than the SD, the given results are false. Conclusion: So, for the actual SSDF list, the three first programs could each be number one. We simply don't know and we simply can't say on the base of our data. In stats the presentation of your results on the base of your data is a whole chapter that you had to study to understand how that works; it's already of importance if you define what beforehand! Here in SSDF the only way to present the actual data is the possibility to create groups of progs. As I said earlier here in CCC, it could be like this 1.-3. are the three progs so and so... Rolf Tueschen
This page took 0.02 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.