EXECUTIVE SUMMARY:   
   - We can ask an AI "what is odd about the 1970s". Already we know the   
     1970s is a bit odd because it featured a rash of reports of cattle   
     with their faces cut off across the western US. The reports got so   
     thick even the FBI was called on to look at it.   
   - The AI chugs away and spits out a list of known datasets that can be   
     simply manipulated to finger the 1970s and no other years.  The top   
     fits include our old friends Saturn, Uranus and Flying Disks.   
   - A lot of other data series can point out the 1970s, but they also   
     tend to "ping" other years as well. The planets and UFO data are the   
     "best explanation" the s/w can find.   
   - The models found seem to agree very roughly -- given we asked a   
     rather vague question.   
   - The models also predict cattle mutilations -- if they are in fact   
     related to Saturn, Uranus and Disks -- still continues at a lower   
     level than the 1970s. We might also predict another big up-tick   
     around the time when Saturn and/or Uranus again approach perihelion.   
     The "Saturn" model says around perihelion -- 2002, then 2031.   
     The "Uranus" model says around 10y after -- 2059.   
     Did the 2002 peak (+-) happen?   
      
      
   We've had a look at this before. But something the s/w has thrown up   
   at me this week relates and is fairly interesting from an AI point of view.   
      
   We recall we've looked at the "big FBI Report" into cattle mutilations   
   in the 1970s. As an apologia for the "report", the FBI pointed out   
   looking at butchered cows was not generally in its area of   
   responsibility.  Which might explain the rather sketchy document it   
   produced in the 1970s that listed a couple dozen reports of animal   
   mutilations with some additions at the end (the quality of the   
   document when it got to the PDF-making stage makes it fairly unclear)   
   written in what seemed to be pencil.   
      
   We ran those numbers through my s/w and it did come up with an   
   interesting list of possibilities for what best explained the pattern   
   of dates seen in the report.   
      
   But the s/w can do a bit more. AI's can answer "vague" questions as   
   well as specific ones.   
      
   Given cattle mutilations hit the headlines in the US during the 70s --   
   some wikipedia pages say the rash of reports lasted from 1973-1983 --   
   we could ask the program -- what is so strange about the 1970s?   
      
   As we recall the s/w has a list of 10s of 1000s of datasets from   
   various govt and NGO agencies around the world related to mostly   
   weather and satellite observations, but also a lot of data on the   
   evolution of economic data in rich and poor countries, plus all manner   
   of other odds and sods the program has uploaded by using google/bing   
   searches.   
      
   So what is weird about the 1970s? We give the AI a dataset that shows   
   years 1950-2020 with a 1 or 0 next to them depending on whether the   
   year is "interesting" (i.e. within the 1970s) or not.   
      
   It chugs away for a few mins and spits out a lit of datasets that   
   "look similar" to the target data we gave it, along with some stats   
   showing "how similar" it is. All the stats has been run through a few   
   packages I trust and the s/w knows how to run and also how to   
   interpret the text output (e.g. looking for error messages or unusual   
   conditions in the output ASCII plots :) and have shown to be   
   statistically significant in at least 2 ways better than 90%   
   confidence (IOW 10% or less chance the relationship could be just due   
   to chance).   
      
      
   Suspect			Lag	R2   
   			(y)   
   sdsaturn-v		1	 0.78547796   
   sduranus-v		10	 0.77604360   
   ufo-Disk		0	 0.39706604   
   preband80		10	 0.28790204   
   ant170			1	 0.22532102   
   minmoon-r		0	 0.16713256   
   preseg160		5	 0.13669447   
   minkenya		10	 0.13381989   
   cosmic-OULU		1	 0.12696118   
   minaravgArcocean	0	 0.09475654   
   stormseg-160		1	 0.09380508   
   lat-80			0	 0.05671197   
      
   Each line gives the name of a dataset in the code the AI uses, a   
   number of years it uses to "lag" the suspect data to make it look   
   closer to the target data (in this case the list of years 1970-1979)   
   and the "explanation power" of the suspect data to predict whether the   
   years listed are "interesting" or "not interesting".   
      
   It claims the "sdsaturn-v" lagged by 1 year predicts around 78% of the   
   interesting years.   
      
   The 2nd line is a similar thing, but for Uranus, and lagged by 10   
   years not 1. Kinda befits a planet that is much further out and has a   
   longer orbit.   
      
   And the 3rd item suggests sightings of "Disk" objects also predicts   
   the 1970s around 40% of the time.   
      
   It seems the program is suggesting the best explanation for the 1970s   
   appearing in our list is "something to do with Saturn, maybe Uranus   
   and Disk UFO's and maybe other things that look like predictors of   
   western US winter weather"..   
      
   Let's have a look at a couple of these in more detail.   
      
   Here's the model for the "sdsaturn-v" dataset.   
      
   (Serial corr detected; estimated rho = 0.671802)   
   y = 0.137581*x + -0.220647   
   beta in 0.137581 +- 0.0174059  90% CI   
   alpha in -0.220647 +- 0.166409   
   T-test: P(beta>0.000000) = 1.000000   
   r2 = 0.78547796   
      
   Dataset:   
   (binned into decades for brevity)   
   Decade	av x          av y      av model-predicted y   
   1950	3.27411		0	0.229809   
   1960	3.42266		0	0.250246   
   1970	14.7949		4.08	1.81486   
   1980	3.25553		0	0.227251   
   1990	3.55803		0	0.268871   
   2000	3.89511		0	0.315246   
      
   It shows the variable/model "predicts" years other than the 1970s   
   should be 0, but the 1970s years should be modestly interesting.   
      
   The interest part comes because Saturn reached its closest approach to   
   the sun in 1973. This makes the variation of the "v" angle -- the   
   angle between the planet's perihelion and current position in degrees   
   -- suddenly switch from 360 to 0 during the year making the "stddev"   
   quite large which is why the AI has spotted it.  The stats don't think   
   this can be a total coincidence.  And, beside, nothing else it has   
   does any better and most of it does a lot worse.   
      
   Here's the model for the "Disk UFO" dataset.   
      
      
   (Serial corr detected; estimated rho = 0.504612)   
   y = 0.113894*x + -1.28638   
   beta in 0.113894 +- 0.0313676  90% CI   
   alpha in -1.28638 +- 0.324333   
   T-test: P(beta>0.000000) = 1.000000   
   r2 = 0.39706604   
      
   Dataset:   
   (binned into decades for brevity)   
   Decade	av x          av y      av model-predicted y   
   1930	1		0		-1.17248   
   1940	2.33		0		-1.02062   
   1950	7.78		0		-0.400535   
   1960	15		0		0.422033   
   1970	35.8		5.22		2.79103   
   1980	15.8		0		0.513149   
   1990	18.4		0		0.809273   
   2000	14		0		0.308139   
      
   Again the model seems to "ping" into it largest value (2.8) for the   
   1970s and registered very low for other years.  The "Disk" data also   
   (not surprisingly) peaks in this dataset (the NUFORC database) for the   
   1970s at about 36 sightings a year, and falls off pretty quickly   
   before and after.   
      
   Certainly not enough to convict in a court of law, but the stats is   
   suggesting "something" related to Saturn (and maybe Uranus) and maybe   
   also related to "Disk" UFO's seemingly has an affinity for the decade   
   where cattle mutilations hit the headlines and supposedly "peak" in   
   the US.   
      
   Of course if these links are anything they also predict that the   
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)