This is a question for more of the tech savvy. With how far technology has come how likely do you all think it would for machine learning to analyze images of morphs to give us a potential image of what the offspring could look like. It would be like the MM calculator but would assign characteristics to different genes to give us a potential idea of what could be produced, mostly for new “worlds first” morphs or combos that don’t have many examples of. It could also be usefull for morph identification as it looks at it and gives you a list of things or could be with it showing you the most relevant results first. Obviously not a replacement for keeping track of what parents produced what clutch and what the genetics involved were.
No likely IMO some results are sometime completely unexpected.
As for reverse putting a snake picture in and getting an ID, not likely either, first because you would need consistency and quality picture each time, many cannot even provide that when asking here for ID, second there are many combos that start to look alike, third there can be a lot of variation in some animals, fourth some are too subtle.
On top of what Deb has said, I’m not a computer expert but I believe the processing power needed to instantly cross reference a picture with half a million+ combos is beyond our grasp.
Though if it was possible it would definitely change the whole game
What @sluggardpick318 is suggesting would most certainly require machine learning, and while it may be possible to some extent, it’s going to require computing power, server space and coding knowledge well beyond what the average person has access too. We’re actually running a program in my lab now where they are using machine learning to assess the difference between healthy neurons and diseased neurons before differences are visible by human observation or currently available programs. It’s pretty cool, but is taking a high powered computer, a huge amount of server space, an engineer and three bioinformaticians to run it. You would anywhere from 100-1000+ pictures of each morph to “train” the software correctly which would be wildly time consuming and require a lot of different snakes whose morph is known 100%
The biggest problem however is that many morphs are unpredictable (would you have guessed that pairing two Mojave’s would produce a pure white snake before anyone had ever produced one?). While we can make educated guesses based on previous pairings, some morphs visual appearance comes completely out of left field.
I do however think it would be useful to use a tool like this to identify offspring from complex clutches, an open source data mining software such as Orange might be pretty helpful here.
Having a tad of knowledge in each of the subjects. I would say it would be possible to a point. Morph ID is entirely possible, there’s no reason a machine shouldn’t be better at iding morphs than we are. It could take variables into account far better than us. However I am unaware how long it would take to build and how much data would need to be fed into it, to get it to that point. Predicting future offspring, you can predict all you want and even get close on some of them, other would be wildly different. I could see it being a fun toy rather than useful.
Biggest issue I see is making sure all your data is legit, a couple mislabeled morphs could make the whole program useless. If you’ve been around, you know there is a lot of mislabeled animals out there.
Unfortunately we still need a good reference sequence first and to map out all of the known mutations. We were talking about the possibilities of this a while back in another thread
I just tried to use WoBP’s images for Mojave and Mystic to see if I could cluster them using hierarchical clustering and it failed abysmally. We’ll see if building a neural network/machine learning program will work because now I’m curious.
If anyone has any good pictures of their Mystics or Mojave’s laying around, I’d love some more pictures to train this on
