renault Posted October 21, 2019 Share Posted October 21, 2019 Just the one Thanks very much for viewing Cheers Renault Link to comment Share on other sites More sharing options...
Iain Emms Posted October 21, 2019 Share Posted October 21, 2019 Splendid shot indeed. cheers Iain Link to comment Share on other sites More sharing options...
musterpilot Posted October 21, 2019 Share Posted October 21, 2019 Very nice indeed John Link to comment Share on other sites More sharing options...
Adam Banks Posted October 21, 2019 Share Posted October 21, 2019 Hmmm ... that's a helluva engine fire! Grand shot! Adam. Link to comment Share on other sites More sharing options...
TigerTigerM Posted October 21, 2019 Share Posted October 21, 2019 Very nice composition. Who does your clouds? TTM Link to comment Share on other sites More sharing options...
olderndirt Posted October 21, 2019 Share Posted October 21, 2019 Looks like a dissipating thunder bumper in back of you - very nice. Link to comment Share on other sites More sharing options...
renault Posted October 22, 2019 Author Share Posted October 22, 2019 17 hours ago, Iain Emms said: Splendid shot indeed. cheers Iain Thanks Iain. I appreciate it! Cheers Sir Pete 16 hours ago, musterpilot said: Very nice indeed John Thank you John Cheers! 16 hours ago, Adam Banks said: Hmmm ... that's a helluva engine fire! Grand shot! Adam. Thanks Adam Cheers Pete 12 hours ago, TigerTigerM said: Very nice composition. Who does your clouds? TTM Thanks TTM ? - "Renault" and his fumble fingers on the data refs switch Cheers Pete 9 hours ago, olderndirt said: Looks like a dissipating thunder bumper in back of you - very nice. Thanks OND Trying to outrun it I think See you Link to comment Share on other sites More sharing options...
Jack Sawyer Posted October 22, 2019 Share Posted October 22, 2019 Contest quality, but what is a data refs switch? Link to comment Share on other sites More sharing options...
TigerTigerM Posted October 22, 2019 Share Posted October 22, 2019 Clouds via xVision perhaps?? TTM Link to comment Share on other sites More sharing options...
BobT Posted October 22, 2019 Share Posted October 22, 2019 Geez that's a beauty Pete! Link to comment Share on other sites More sharing options...
adambar Posted October 22, 2019 Share Posted October 22, 2019 That's a dandy T! Link to comment Share on other sites More sharing options...
renault Posted October 22, 2019 Author Share Posted October 22, 2019 3 hours ago, BobT said: Geez that's a beauty Pete! Thanks Bob Appreciate the comment Cheers Pete 1 hour ago, adambar said: That's a dandy T! Thanks Adam See you Pete 4 hours ago, Jack Sawyer said: Contest quality, but what is a data refs switch? 4 hours ago, TigerTigerM said: Clouds via xVision perhaps?? TTM Uh, you might be sorry you asked, but this was a little project I had on the go this past little while. I hope it makes some sense. It was an interesting little challenge which I enjoyed doing to try to keep my mind in gear and keep up my interest in such things.... For several years before I retired I was involved in AI research, specifically using pattern recognition to improve history matching techniques of real data sets using finite difference simulation models. There is a particular class of technique called unsupervised learning that I liked. I still like to follow along with new developments and I use a computer language called R Script (which is designed for artificial intelligence problems) as a hobby to play around with new ideas - essential creating little "proof of concept" toy models. It helps (I hope) to keep my neurons from completely disappearing as I age. As much as I do physical excercise every day and enjoy flight simulation, I also try to read a lot and also to do things that really stretch my mental abilities. Unsupervised learning means simply that you use a computer algorithm to find patterns in data and then use those patterns to predict the correct outcome i.e. reproduce the characteristics of the pattern . Simply put , this method is called cluster analysis and involves letting a computer algorithm group input data into clusters based upon a similarity measurement i.e. similar clusters of data share measurements which are more similar while dis similar clusters are the opposite. The actual methods that I like to use are generally called K-means clustering and correlation clustering. About 8 months ago I wondered if I could generate weather pattern data for XPlane and use these techniques to look at cluster information for "patterns" which I would like and to reject those which I didn't like. XPlane uses parameter values called data references (data refs) for all kinds of things. A common example would be the fog_be_gone data reference that many of you might be familiar to control fog and haze. The data references are generally what is in XPlane plug ins and Fly With lua scripts that many of us have such a love/hate relationship with. So what I did was to build a little bit of code that started out by randomly picking values for all the several dozen cloud data references that XPlane uses for generating weather. For each set of data I did a simulation run i.e. I put all these values into the corresponding data ref in XPlane and then went flying. If I liked the way the weather/cloud development looked , I kept the data - if it didn't I threw it away and started again. This took the better part of 6 months to do, but I ended up with about 250 sets of data that created weather patterns that I liked. I figure I made about 1500 data sets in total. In terms of AI this is an incredibly tiny data set, so the results at best may not be very reliable but …. I then used AI techniques to use cluster analysis to determine what were the combination of variables (the data refs) that combined together to give my criteria for "pleasing results". My criteria in the original selection was that I wanted good contrast, good volumetric rendering of the clouds and dis similar patterns converging. Basically as both Adam and OND said "try to get ahead of the thing before it blows up!" I then wrote another program which randomly picked starting values and using the learnings from the AI model I attempted to generate interesting cloud patterns. The model gave me all the values to use in the data refs and I then put them into XPlane by reading a data file generated by the model using a .lua script and kept flying/changing patterns. Being an AI technique I also used the new data I generated to analyze and use in further generation of new sequences - a bit of an attempt to put a bit of intelligence into the algorithm. This failed rather miserably * as you can read below. The particular screenshot was the result of this little experiment - I had the model generate 50 different weather sequences and I selected a series of them (sometimes you will get 3 -5 of them that are almost visually identical) . I chose about 20 of them that did show changing patterns and I thought you might be interested in see the 20 + images that lead up to this particular screenshot. I have built a composite below showing this. You read it from left to right and then down to the next row and repeat. * You can see how the patterns are similar, but also different for a while , but also how they start to converge - this is a result of the cluster means starting to become more similar as additional patterns are calculated. Left alone overnight it converges into a big grey blob with no structure as all the patterns merge into one. Not all ideas are good ones.... All of these images are similar , but also are different. The one I chose , marked by a star was the one I liked the best, but all the others also met my criteria, even though they are not necessarily all interesting to look at. What was fun for me was that even though the input data to generate these weather patterns was randomly chosen, the algorithm was capable of learning from the criteria rules it generated by only using cluster criteria from pattern information that formed part of the "this gives pleasing results" idea. It was a fun exercise to do once, but "I'm over it now" and am on to other stuff... And yes, you don't have to say it, using a good weather program would probably have accomplished the same thing in a much shorter time span . But I enjoyed myself and the challenge of seeing if I could do it.... Link to comment Share on other sites More sharing options...
craigeaglefire Posted October 22, 2019 Share Posted October 22, 2019 Crikey, Pete there actually is method in your madness! enjoyed the description of the process of your experiment & the screenshots are fascinating... the one you chose to post first is a winner... cheers, craig Link to comment Share on other sites More sharing options...
YAMBA1 Posted October 22, 2019 Share Posted October 22, 2019 this is just amazing I love those clouds Link to comment Share on other sites More sharing options...
Aussieflyer38 Posted October 22, 2019 Share Posted October 22, 2019 Great shot Pete. Don Link to comment Share on other sites More sharing options...
Jack Sawyer Posted October 22, 2019 Share Posted October 22, 2019 4 hours ago, renault said: I was involved in AI research Pete, this is astonishing to me, I never knew you did all this. I'm writing a novel about AI, it's sci-fi but it discusses the very subjects you mentioned. I would love to meet in person some day, I'd have a million questions for you. Many thanks for that explanation and in no way am I sorry I asked. Link to comment Share on other sites More sharing options...
lauriebe Posted October 26, 2019 Share Posted October 26, 2019 Wow!! What a shot R. Fantastic skies. Link to comment Share on other sites More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.