user since
Sun Aug 18 2002 at 01:43:20 (21.6 years ago )
last seen
Sun Sep 26 2004 at 23:26:45 (19.5 years ago )
number of write-ups
17 - View apothanousa's writeups (feed)
level / experience
2 (Acolyte) / 368
mission drive within everything
node what everyone should know and doesn't seem to. Reach level 4. C! something deserving. Retire.
specialties
Things my mother taught me
school/company
The Penguin Juice Co., Inc.
motto
motto quia mottandum
most recent writeup
logical punctuation
Send private message to apothanousa

Warning: everything on my homenode is obnoxious and should not be read

For if it weren't, it would be in a real node. QED. Thank you for your attention.

apothanousa's Most Frequent Coolers

Thanks to the following people for C!ing my writeups:

Cools Perpetrator
3C! ariels
2C! Jet-Poop
1C! aneurin, arieh, dem_bones, Erenata, Excalibre, Gritchka, jaubertmoniker, jessicapierce, lj, olmanrvr, Ouroboros, Pseudo_Intellectual, Quizro, ReiToei, sgs, timeshredder.



Nitpicks about E2:

  1. When you search from a display of a single writeup (rather than the whole node) it doesn't seem to create a softlink. I claim this is reproducible, but if there are several different sets of generating code (like the ones you would never have realised were different if they hadn't said “sumbit”) I may have to specify more precisely the circumstances under which this happens. Is this a bug?
  2. How about standardising accented characters in node titles by doing search-and-replace to change HTML accents to ISO accents in
    • all existing node titles
    • all node creations, and
    • all search terms (because it's easier to type the HTML ones, even though we don't want them in the real titles because they're search-unfriendly)?
  3. Very likely this issue is far more complicated than I'm assuming it is, but how about attacking the slow server problem by responding to "easy" (fast) requests preferentially over "hard" (slow) ones when thene is a long queue (as opposed to responding in the order received?)—the idea being that the people who make the slow requests are the ones who have to wait, rather than everyone else having to suffer for them. (This might even encourage people to use more links and fewer searches, because they know it's faster.)
  4. In Node Heaven, if you don't have any “interesting” deleted writeups, you get the message ‘xx deleted writeups total, shown’ instead of ‘... total, 0 shown’. (The equivalent thing may or may not happen when you have 0 total writeups there, I can't tell since I don't.)
  5. If you log in while you're at a "Findings:" page, then you get a bizarre page which says "Findings:" at the top and nothing else. (Actually, my programmer's instinct says that this one could well require too much special-case coding to be worth doing. So forget it. Maybe.)
  6. If you search either for an existing nodename with "Ignore Exact" enabled, or under the other circumstances (which I can't reproduce but definitely exist: bug in search algorithm?) where you would get instead of the "Create Node" box the message "(a node called "—" already exists, so you can't create one)", when you're not logged in you get instead the message "If you Log in you could create a "—" node."; this is of course untrue since such a node already exists. In other words, the "already exists" message should take priority over the "must log in first" message, which it currently doesn't; or better yet a different message could be formulated for this case.

Interpreting the Honor Roll

The Honor Roll computation works by approximating merit by a normal distribution. But the next step is, bizarrely, to compute your LF in terms of the value of the pdf at your merit, rather than its integral? Okay, sure, it's invariant in the sense that it corrects for mean and s.d. in the way you'd want it to; but the effect isn't that of the graph Professor Pi draws in the Honor Roll superdoc. If you draw the graph to show LF against distribution (i.e. the one which tells you approximately how many noders get what LF) you find that it's constant for a longer time and drops more sharply towards the end:

1.00 |.....*******                                                                              
     |    ...     *****                                                                         
     |      ...        ****                                                                     
     |        ..           ***                                                                  
0.95 |         ...           ****                                                               
     |           ..              ***                                                            
     |            ..               ***                                                          
     |             ..                ***                                                        
0.90 |              ..                  ***                                                     
     |               ..                    **                                                   
     |                ..                    ***                                                 
     |                 ..                     ***                                               
0.85 |                  ..                       **                                             
     |                   ..                       ***                                           
     |                     .                         **                                         
     |                     ..                         **                                        
0.80 |                      ..                          **                                      
     |                        .                           **                                    
     |                        ..                            **                                  
     |                          ..                            **                                
0.75 |                           .                             **                               
     |                            ..                             **                             
     |                             ..                              **                           
     |                              ..                              ***                         
0.70 |                               ...                              ***                       
     |                                 ..                               **                      
     |                                  ..                                **                    
     |                                    ..                               ***                  
0.65 |                                     ..                                **                 
     |                                       ..                                **               
     |                                        ...                               ***             
     |                                          ...                               **            
0.60 |                                            ...                               **          
     |                                              ...                              ***        
     |                                                 ...                             **       
     |                                                    ....                           **     
0.55 |                                                       .....                        ***   
     |                                                           ......                     **  
     |                                                                 .........             ***
     |                                                                         .................*


. = Curve shown in in Honor Roll (horizontal axis = merit, from 0.0 to 2.0 standard deviations)
* = Curve as perceived by noders (horizontal axis = distribution, from median noder to best noder)
Both curves represent only the top 50% of noders; the bottom 50% correspond to a constant 1.00 to the left. Note that because of the different horizontal scales (which are the whole point) it doesn't strictly make sense to make comparisons like “one is higher than the other”, though I've tried to align them reasonably. The point is really the shape—facts like ‘over 75% of noders have LF above 0.80’ which you can read off from my graph, but about which the graph in Honor Roll is highly misleading.

Note that I am not advocating a change to the Honor Roll system, because I believe firstly that the general idea is an excellent improvement to E2, and secondly that the precise formula currently in use works well. But I find the graph above to be the only convincing evidence for the latter—I didn't believe it from reading the Honor Roll superdoc.

Professor Pi writes:

“[...] I think plotting it against distribution is actually more misleading. Because the norm. dst. is only an approximation of the actual distr, the graph (which is only an *example* on Honor Roll ) you will draw incorrect conclusions such as "x% has a LF above y". And I don't agree with your statement below the figure "curve as perceived...". People don't perceive this curve in terms of distribution, but in terms of what they can observe directly: their merit and LF score. I have some ideas on other ways to present this data, so that the average noder can understand it. That will eventually end up in a FAQ. [...]”
To some extent I think Professor Pi and I have different objectives. Professor Pi, being practical, sane and rational, having administrative obligations, and observing from experience that the system works well, infers that there is no need to change it, and is therefore expending effort in ensuring that noders understand the system that has been implemented. I, being impractical, pedantic and rational, having little practical experience of the system, and believing that no convincing quantitative justification for the formulæ has been given, have tried to determine theoretically whether the system works well, have in my opinion shown that it does1 (by drawing the graph), and have inferred that there is no need to change the system.

1As Professor Pi points out, the distribution of merit is not necessarily normal. But if the normal approximation is good enough for him, it's good enough for me (he says, having previously cast aspersions on one of PP's formulæ.)

For those who either doubt my computations or simply can't work out what the hell I think I'm doing, here's the source with which I drew the above graph:



#!perl -w
$PI=3.1415926535897932384626433832795028841971;
$xscale=30;$yscale=40;$xsize=90;$ysize=40;$xoff=0;$step=0.01;
@d=map {[(map {" "} 1..$xsize),"","\n"]} 0..$ysize;
$phi=0.5*sqrt(2*$PI)/$step;
for ($s=0;$s<3;$s+=$step) {
$phi+=exp(-$s*$s/2);
$d[-1+(1/(2-exp(-$s*$s/2))-1)*$yscale*2][$xoff+$xscale*$s]="."; # pdf
$d[-1+(1/(2-exp(-$s*$s/2))-1)*$yscale*2][($phi*$step/sqrt(2*$PI)-.5)*$xscale*3*2]="*"; # perceived LF
}
$t=-1;
map {if (++$t%4) {print "     |"} else {printf "%.2f |",1-$t/$yscale/2};print @$_} reverse @d;

My ekw settings, as if you cared


ekw_alinkcolor="#f038c0";
ekw_bgcolor="#f8f8f8";
ekw_headingfont="Tahoma,Verdana,Helvetica,Arial";
ekw_linkcolor="#1038c0";
ekw_logoaccenttext="#ffff90";
ekw_logobackground="#3058e0";
ekw_logoborder="#000000";
ekw_logofont="Tahoma,Verdana,Helvetica,Arial";
ekw_logofontsize="17pt";
ekw_logomaintext="#f0f0f0";
ekw_monofontsize="12pt";
ekw_oddrowcolor="#d0d0f0";
ekw_textcolor="#000000";
ekw_textfont="Arial, sans-serif";
ekw_textfontsize="12px";
ekw_vlinkcolor="#f038c0";
If the result doesn't look good to you, it's your own fault for not being red-green colour blind. (Or your monitor needs gamma correction, or both.)