I’m a data science student at lambda and I had a cool idea. Would you guys be interested in a website or app that can take a picture of your ball python and tell you the genes it has? For instance you upload a your picture and it tells you this is a Mojave ghi clown. I would train a program using convulated neural networks I think. But it could be fun haha. I want to see how accurate I could get it at classifying things like yellowbelly or orange dream. It could be a giant inaccurate dumpster fire but it would be a fun inaccurate dumpster fire
I’ll give you both sides of my thoughts.
YES! if this could be possible then it would completely change the whole game. If it worked I would happily pay really decent money for such a software.
Without sounding like a dream crusher, I think this is the most likely outcome. With the variation in different morphs, it would be extremely difficult to guarantee which genes are involved.
If you have the resources, then this is what it is all about. Enjoying what you are doing. I say give it a shot.
I would love you to try this and see how it works out! I had tried a little bit using Orange and some other image data mining software and it failed pretty abysmally (~60% success rate of differentiating between clown and Mojave), and would love to see someone more skilled in data science give it a try.
I’m sure it would definitely be possible and probably successful as our engineer built a program/robot that is super accurate at sorting fruit flies, and ball pythons morphs are way more easy to distinguish than that. Obviously the caveat is that on rare morphs/combos you probably wouldn’t have enough pictures to train with, but for more common morphs it should work.
There are too many variables to be accurate first the quality of the picture, let’s face it a lot of people ask for identification and provide pictures that are so bad they might as well take a picture in the dark and ask for ID.
Second there are too many combos that are starting to resemble each others (the more gene are stacked up the more it happens) not too mention the use of subtle mutations.
Even as a breeder sometime the only way to tell what you produce it is actually to breed it and isolate each genes.
My thought is this. Most of the time we look at morphs we see the colors, or drastic pattern changes. But maybe the program could pick up very slight changes in pattern, flaming etc. I think it will either be very accurate or like I said dumpster fire. But their are alot of variables to account for, BELs , Bellies photos, gets, etc
Not a new conversation Advanced genetic calculator
Bad pictures isn’t that bad I could just program in a raise error that returns a prompt saying " are you in a dark room ? Try again etc". The too many genes part is gonna be hard yes. But it depends on what I wanna make. Let’s say we keep it basic and only go for very well known morphs, I that should be easy. The hard parts are gonna be (like you said) stuff were it’s one gene but people break it into different lines. Butter and lesser, banana and coral glow. Etc etc. That I think I might be able to over come with lots of imagine training though. I think the real problem comes from training on inaccurate pictures. For example some says they have a BEL and it’s a mystic potion. I have seen that on morphmarket alot actually
I completely agree. At best a app for morphs could only say what a snake possibly is and not be 100% accurate. It is a cool concept though.
Also picture quality pixel wise isn’t a big deal with convoluted nueral networks
Actually no no, I just thought of the biggest problem actually. World’s firsts. It would be nearly impossible for a trained model to predict things it’s never seen. For example let’s go back in time to before BELs where a thing. Let’s say I trian the model on available morphs made the site and someone makes the first BEL and tries to classify it. The system caannot predict the whacky way some genes come together to make completely different snakes. Hmmmmmmmmmmmmmmnnn
Ah… But a MysticPotion is a BluEL. A “dirty” one, but a BluEL nonetheless.
For simple things, this might work but I think the hyper-variability would be your downfall. You also have to consider things that are not the same but look the same. Is it a BluEL or a BlkEL or an Ivory or an all-white Pied??? SuperCinny or Suma or Abyss??? Asphalt or Gravel or YB??
I wouldn’t think that the kind of folks who would have a world’s first would be using an app like this to identify the genes in their snake. I do think that it would be a cool database photo matching app that could give someone an idea of what they have with a probability percentage (that could be adjusted if say, they entered known parentage) like the Picture This app for plants. Users could continuously build and update the database. The hardest part would be verifying the validity of morphs so you would need some type of vetting process for that in order to maintain accuracy. I would think that an app could eventually learn pattern and variation to a point that could potentially rival the human eye. It’s a cool idea and think it would be a neat project.
Not meaning to go off topic but what combo is a Abyss? never heard that name before but it sounds cool.
It is a Suma GHI. Pure black with no genetic issues.
Lol the point of that statement was people lying haha classifying BELs would be simple
Oh really I got to see this
Hell to the YES THAT’LL BE A BLAST TO HAVE AND SO INFORMATIVE PLEASE DO SOMETHING LIKE THAT…YOU’LL SAVE A lOT OF HASSLE and save a lot of time
Even if it dosnt work with Ball morphs it might work with other species. Like rosy locals or corns.
How exactly are they lying?
If someone says they have a BluEL and their animal is a MysticPotion they are telling the truth. There is no lie there.
It is not actually pure black, they have some traces of pattern on them. But they are really nicely dark
The one I saw was listed as a Mojave lesser. I didn’t say that earlier because I wasn’t being specific enough sorry haha