Anyone here into deep learning and neural networks?

I used to but people like that actually enjoy their jobs

@phat-lasagna well, that’s good to know! Right now, it’s not so enjoyable.

It's my main line of work currently, but I don't think I'm much into it.

I was approached by some old boss I guess because he knew I was the most decent in math and statistical analysis.

I'm certainly not the best at it but the pay is good, the job is comfy and they are satisfied with the results so all good I guess.

I do, however, find it all a bit boring, not gonna lie. If I have to work in data sci, I'd rather work in data Viz tbh.

@CoreFusionX Data viz is one of my favorite things to do. I can go on and on and never sleep most of the time. But PyTorch is giving me so many issues that I cannot get to the fun data cuz part of the process. 😭

Would love to know more about your work day to day. It’s what I want to do. I’m rather obsessed with data science.

@EvanMarie how so?

@phat-lasagna Well, so I worked around the issue, but I’m not satisfied with things “working” unless I know why they didn’t work before. It was an intuitive thing that I came up with in my sleep that fixed it. But it was a tensors not matching up shape-The issue arose when I wrote a function to test 4 different loss functions at 10 different learning rates for 500 epochs, so 40 times going through 500 epochs, right? Well, EVERY time, it would stop somewhere in the middle to end of the second loss function and give a runtime error. But it was never the same place. So I ended up adapting the function to take one loss function and try all the different learning rates. And I ran the functions individually, and everything is fine. But I don’t understand why. And it makes me a little cray. Now my issue is with finding out my accuracy, not just my loss. PyTorch is only 1 week old for me, so I’m struggling. But I do like it. I just wanna be good at it.

@phat-lasagna OH! Wait! Did I answer the question you actually were asking? Or did you mean something else? WHOOPS! Have had my head buried for hours this morning.

Soon to be in neural networks. My first ever professional project about that is starting on Monday 😁

@EvanMarie lol sounds like ur hardware is just saying “nope”

@Earu That’s so exciting!!! Congrats!! I wanna hear all about it. Hit me up on discord if you ever wanna. I would
Love to nerd out! (✨ EvanMarie ✨#6114)

@phat-lasagna lol! Right! But I ended up winning!! And I got me an accuracy rate of over 95% currently. So I’m a happy snowman! (Sorry, Frozen reference. My previous career was opera singer / teaching little kids to sing Disney songs.)

@EvanMarie good shit. Oh no are u a Disney adult lmao

@EvanMarie

Right now, not really ready to go into DNN since dataset is still too small.

It's really a text classification problem along with a recsys and inference engine.

Thing is, input is very very noisy, therefore, at least in the beginning, I want to run other ML algos that offer more explanation as to their decision process, just so I can better evaluate performance.

Goal is also to be able to provide a semi ML-assisted labeling effort in order to then train DNN.

So yeah, my day goes between improving the custom labeling tool (sagemaker cost for labeling jobs is ludicrous), curating datasets, and mostly calculating stats and interpreting them in order to find better features to extract, etc.

Graphics programming (DX, ogl, vulkan, etc) is my real passion tho.

Hence why I usually have fun with data viz, but you know, whatever pays the bills, as I'm freelance now and can't really be picky about work (yet, I hope)

@CoreFusionX You’re so speaking my language. Love it!! I love when suddenly features jump out and starting screaming at you to look at them because their importance cannot be missed. I’m having that currently as FINALLY things are coming together with my data. Amazing accuracy given the situation, and really great plots resulting. It’s like a gift from the heavens when that happens. I look forward to working more with text. It’s so daunting but so insanely interesting. It’s honestly the direction I wanna go in. There’s so much there.

Have you seen the show “Invasion” on AppleTv? There’s a Japanese ML chick trying to decipher the “alien sounds”, and they honestly did an OK job not making it silly and annoying. I’m considering playing around with some SETI data now, after seeing that. Inspired to find language where there “is none”.

@phat-lasagna No, not at all actually. It was only because I was paid too. I’m a Miyazaki kid!

@CoreFusionX Ah, ok, that makes total sense that you would feel bored. Have you done much with Houdini? It’s getting big and quite lucrative. Or is that totally a different graphics tangent? I can see how data visualization could go A LONG WAY if more 3D animation were involved. And I don’t mean that TensorFlow type stuff. I mean, dynamic!

@EvanMarie

Not a big fan of Houdini itself, but have dabbled with blender, Maya, 3dsmax, unity, unreal, Godot, threesjs, you name it.

In truth it's shader development I love. Like, ultra low level development, which results in incredible eye candy 😂

@CoreFusionX Agreed! That’s awesome. We need more good eye candy.

@EvanMarie howl’s moving castle changed me as a child

@phat-lasagna They’ve all changed me. Howl’s is a major fave though. Mononoke is my boo. I gotta show you the painting of the Forest spirit I commissioned by a Japanese artist friend of mine! Will post.

@phat-lasagna

@phat-lasagna now I’m thinking I might need something from Howl’s! I cannot imagine picking ONE still frame though.

@EvanMarie that looks amazing I need to see that one. I keep waiting to get into a relationship with someone who I can watch all the ghibli films with.

@phat-lasagna Do not settle for less! If they don’t like Hayao, they best be out the door! And when you have kids, raise them on Miyazaki. They are better humans that way. I have proof. 💜💜💜

@EvanMarie fuck it, I’m watching Howl’s moving castle tonight - going to relive my days as an 9 year old

@EvanMarie thanks for the kind words lol

@phat-lasagna That’s awesome! I would too, except my kid has us watching Jojo’s Bizarre Adventure every night for now. Enjoy Howl!! Tell him I said hi! Hug Sophie for me! Give Heen a kiss! Oh, and Calcifer. The whole gang! Omg, I might need to watch too.

@EvanMarie your kid is based, JoJo is one of my favorites haha. I don’t see it as too much of a kids show off the top of my head but u guys do u lol.

@phat-lasagna NO JOKE! It gets RAUNCHY! But it cracks me up. I mean, my son is old enough. I don’t censor much. But my daughter gets REALLY uncomfortable at times when we watch. Honestly, few of the Miyazaki films are truly kids’ films.

I go around yelling “Oh my GOD!” With a Japanese accent like Joseph Joestar. He’s my fave!

@EvanMarie https://youtu.be/KWreHiN_E-Q

@phat-lasagna Dude, that stand FREAKED ME OUT!!!! No wonder Joseph was yelling so much that episode.

@EvanMarie true, it’s the most suspenseful. Have u seen stone ocean?

@phat-lasagna On Jojo? Is that the one where they’re in the submarine? I was like OMG! You are LITERALLY in Egypt! Swim!!!!

@EvanMarie nah it’s the new season, they released new episodes this year

@phat-lasagna ah, I see. I’m still on season 2. There are SO many episodes. I think I’m almost to the end of this season. Still trying to get Dio. I guess that’s always gonna be the case.

@EvanMarie yeah it’s pretty much a fever dream

@phat-lasagna Like the Oingo Boingo art style?!

@EvanMarie I feel like that’s more golden wind, not sure what stone ocean would be categorized under though 🤔

i use Stable diffusion on my system for getting inspiration of certain things. Inpainting is quite nice for that.