EXECUTIVE SUMMARY:   
   - We look at the "explanatory factors" for UFO activity as determined   
     from a new tweak of a novel AI s/w.   
   - The new patterns are supposedly far more robust than those   
     previously determined by a robust correlation of individual   
     datasets against UFO sightings data.   
   - Some explanatory "Themes" in the new approach can be split between 1   
     or more effects that might influence observers and therefore   
     monthly UFO sightings and effects that certain natural phenomena   
     might have on UFO activity directly.   
   - But some Themes likely do not relate much to the behaviour of   
     observers and likely "mostly" effect UFO activity directly.   
   - Taking these factors into account it seems the biggest direct   
     influence on UFO activity seen here is cosmic rays.  Cosmic rays   
     detected in some locations predict a greater UFO activity (as   
     typically observed in NAm); some locations link more cosmic rays   
     with lower UFO activity. The division between +ve and -ve influences   
     might tag some regions as "home" locations to UFO's and some regions   
     as "areas of interest" for UFO's.   
   - Our old friends of the world's oceans, N Pole and S Pole figure   
     highly as influences of UFO activity. It's almost like these regions   
     feature as UFO bases of operations. But the Theme of "astronomical   
     data" features even more. While UFO's make work here the modeling   
     suggests they call some of the outer planets or moons thereof home.   
      
      
   We've looked before at which data -- selected from 1000s of "comma   
   separated variables" (CSV) files maintained by different agencies   
   around the world and generally down-loadable monthly via anon ftp or   
   similar -- seem to strongly correlate with UFO activity (I generally   
   use adjusted NUFORC monthly sighting counts 1950-present).   
      
   This work is part of a long term study I'm progressing into "unmanned   
   science" -- where computer s/w plays the part of a scientist or   
   scientists and tries to knit together everything it "knows" to   
   estimate how likely various theories about the world of its own   
   creation are to be true.   
      
   The AI-based programs have now upped the ante and moved from   
   individual data series as a basis for judging various "target" data   
   that it is trying to "explain", to a scenario where it groups   
   "explanation" datasets into "themes" that are meant to improve the   
   robustness of its decision processes as well as make its explanations   
   more intelligible to poor dumb humans.   
      
   It's a work in progress. This post will summaries a re-hash of trying   
   to understand UFOs via this procedure -- this time using the "themes" idea.   
      
   It does change the mental picture of who, what, where and why funny   
   objects are seen buzzing around the skies in almost every part of the   
   world. And while the Big Pentagon Report to be released at least to   
   Congress in the next few weeks may or may not tell the public anything   
   new or particularly interesting, we can try to preview what it might   
   reveal by looking at the latest numbers.   
      
   The AI has been hunting up many new data series and now has a   
   collection of many 10s of 1000s -- mostly satellite data averaged over   
   the world and produced monthly. But some are "gridded" that average   
   weather or other data over grid squares across the planet by latitude,   
   longitude, height/depth relative to MSL, and sometimes other   
   dimensions as well.   
      
   The latest run of the s/w to "explain" UFO activity produces the   
   following summary table:   
      
   Theme		"Explanation power" for adjusted   
   		monthly UFO sightings (R2)   
   npole	         0.90156   
   cosmic	         0.88612   
   clouds	         0.56150   
   cem	         0.55404   
   z	         0.53668   
   ocean	         0.53042   
   spole	         0.51988   
   ca	         0.49304   
   astro	         0.47792   
   q	         0.47556   
   seg	         0.44700   
   dep1.	         0.37424   
   rrg	         0.37424   
   ecl	         0.36502   
    r.	         0.36047   
   neptune	         0.35572   
   dep100.	         0.33925   
   RA(7)            0.33553   
   DecRA	         0.33111   
   uranus(7)        0.32529   
   pluto(6)         0.32222   
   dep10.           0.32048   
   gas(7)           0.31443   
   lon              0.31172   
   phyto(4)         0.29312   
   saturn(7)        0.24937   
   Dec(7)           0.24119   
   dep1000.         0.23808   
   lat              0.21242   
    rg(2)           0.20792   
   msl(5)           0.17274   
   tmp              0.16318   
   jma              0.16279   
   band             0.16062   
   arc              0.15234   
   sunmoon(6)       0.14372   
   wind             0.12643   
   sun(2)           0.12547   
   storm            0.12048   
   ant              0.11930   
   stormseg         0.10293   
   torn             0.08886   
   pre              0.08832   
   stormband(5)     0.08776   
   jupiter(5)       0.07538   
   hail             0.07357   
   radar            0.06714   
   chlor(2)         0.06454   
    elong(2)        0.06129   
   mercury(4)       0.05471   
   moon(3)          0.05425   
   venus(2)         0.05406   
    mag(3)          0.05223   
   land(5)          0.03455   
    FV(3)           0.02722   
   mars(2)          0.00346   
      
      
   The "theme" is a short code that indicates the commonality between a   
   bunch of datasets that are used to build models of the target dataset   
   (UFO sightings) from members of the bunch. The s/w tries to be super   
   robust in building all its models. If there is any doubt that a model   
   actually predicts something about UFO activity it is ignored.  The s/s   
   uses the selected explanatory data in several different ways to build   
   its models. The overall "score" for the Theme is the MINIMUM of all   
   the models that are built to explain the target. IOW the scoring is a   
   "minimax" procedure. Again, intended to produce statistically (very)   
   robust assessments of the various models being built inside the s/w.   
      
   So line 1 of the table scores the Theme "npole" as the "best   
   explanation" for UFO activity. Datasets such as cloud height, surface   
   temperature, water salinity, etc etc related to the region 60N-90N   
   when combined in various ways produces models that explain a minimum   
   of 90% of UFO sightings reports.   
      
   This might sound like an extraordinary high level. UFO sightings,   
   after all, are "mostly" mistakes or hoaxes -- they don't represent   
   anything more real than that. Estimates vary but we might assume about   
   95% of sighting reports are not anything of any real interest to us   
   here (although some of them may represent interesting new atmospheric   
   physics if nothing else ;).   
      
   But what is going on with this model building -- we are explaining   
   something about the observers as well as something about what they   
   might be seeing. The "north pole" has a strong affinity for UFO   
   sightings because it says something about how observers behave as well   
   as something about what Our Friends (tm; not necessarily friendly) are doing.   
      
   Given no other knowledge about the observers in question -- mostly   
   random citizens across the US and Canada -- standard rules of thumb   
   assign equal weight to each of the possible alternatives.  I.e. if   
   the "npole" explains 90% of UFO sightings then we might by rule of   
   thumb assign 45% to "what the observers are doing" and 45% to "what   
   the UFO's are doing".   
      
   So we might therefore say the affinity between the NPOLE and UFO   
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)