convoluted connor @notwa@witches.town

@dd86k lol I woke up to two separate notifications this morning calling me lewd. maybe I am lewd...

try not to turn into a vampire werewolf skeleton this impending super duper moon

@dd86k I don't mind a DddDM (the first D being discord I guess)

didn't burn it this time but still kinda underwhelming, although this stuff is a little old at this point

wish I still had the drive to make stupid things for dumb fun as a kid. totally pointless webpages and aimless game maps/mods, incomprehensible video/audio shitposts... eventually you make something pretty neat by fluke, you know?

gonna make some gween tea tonight

i should probably change the default matplotlib font

DET. i want to make this prettier sometime but it's not a priority. https://witches.town/media/ZMDw73xiH_u8sIjQ-CQ

@Xkeeper i feel your pain though. i've been wanting to put out something that's more indicative of modern AI, but it's a lot of work writing it from scratch in plain Lua, and i keep getting burnt out working on it… writing backpropagation without a numpy-esque library is hell.

@Xkeeper unfortunately it's still an AI, just a very old and limited one. plus, its model is way too simple for the environment it's in, so it's basically doomed when it comes to learning correlations.

current mood: ba!
https://www.youtube.com/watch?v=L0DAQoqVzAw

all my good posts are just random excerpts of random chats i have

read-write-execute: the life of an author

implementing it was kinda fun. MNIST comes in some arbitrary order, so i first arg-sorted samples by class, then interleaved them (0 to 9 repeating) by swapping axes 0 and 1. after that i can flatten, roll, and split the samples into training and validation sets for each fold.

however this only works when all the classes have an equal number of samples. also, my batch training function shuffles by default, so i omitted it from the procedure.

another one off my list of stuff i should've implemented ages ago: stratified k-folding. i think? i'll double-check my code tomorrow.

valid loss: 0.50343 ± 0.02206
valid accuracy: 89.794% ± 0.631%