Thirty#1

Open

FibonacciDude wants to merge 70 commits intomainfrom

Commits on Feb 12, 2022

load data and normalize untested

adelmotheboi
authored and
FibonacciDude
committed
dataloader load data complete with numerically stable normalization

adelmotheboi
authored and
FibonacciDude
committed
dataloader with batches train and test

adelmotheboi
authored and
FibonacciDude
committed
updated download_data.sh to make files writeable from python

adelmotheboi
authored and
FibonacciDude
committed
some progress on train, still some bugs to fix

adelmotheboi
authored and
FibonacciDude
committed
training code done, there might be bugs

adelmotheboi
authored and
FibonacciDude
committed
works

adelmotheboi
authored and
FibonacciDude
committed
fixed the bigggest bug of my life (the dimensions were meshed together), may have added anothe one with the majority voting

adelmotheboi
authored and
FibonacciDude
committed
code?

adelmotheboi
authored and
FibonacciDude
committed
fixed adabn, no avg functionality yet

adelmotheboi
authored and
FibonacciDude
committed
code works, problem with testing set though

adelmotheboi
authored and
FibonacciDude
committed
changed some code to be more consistent with the literature and more commonsensical. Also, test mean/std applied to all bug

adelmotheboi
authored and
FibonacciDude
committed
adding new things, not tested yet but scared computer is going to crash and code is going to be lost

adelmotheboi
authored and
FibonacciDude
committed
removed extra files

adelmotheboi
authored and
FibonacciDude
committed
training on validation data bug + dataloader faster through cpu -> gpu later

adelmotheboi
authored and
FibonacciDude
committed
Make dataloader load entire dataset at once (it fits). Made everything a little cleaner and a little faster

adelmotheboi
authored and
FibonacciDude
committed
added things to make better model (tweak)

adelmotheboi
authored and
FibonacciDude
committed
There was one main bug that I hadn't noticed all along: striding. Apparently, the striding was incorrect so the model wasn't learning much

adelmotheboi
authored and
FibonacciDude
committed
randomization in tasks bug and others. Init loss is strange, not convinced there aren't more...

adelmotheboi
authored and
FibonacciDude
committed
not completed yet, rough draft to make it predict the labels first to debug

adelmotheboi
authored and
FibonacciDude
committed
work in progress: root mean square and downsample

adelmotheboi
authored and
FibonacciDude
committed
let's see

adelmotheboi
authored and
FibonacciDude
committed
a whole mess of things, clean up is coming

adelmotheboi
authored and
FibonacciDude
committed
Simple. And it works!

adelmotheboi
authored and
FibonacciDude
committed
fixed bug but still works

adelmotheboi
authored and
FibonacciDude
committed
Added glove functionality indexing, I think it is correct but the code is too slow...

adelmotheboi
authored and
FibonacciDude
committed
For now, the loading of the glove and emg data are done. We have to test this correctness and add contrastive learning

adelmotheboi
authored and
FibonacciDude
committed
added contrastive, issue in learning may have been regularization (the emg model was not learning). there is still an issue with the test people... how do we solve that?

adelmotheboi
authored and
FibonacciDude
committed
Add stim==0 to norm, take out *l2 after model.l2(), make hyperparams easier to deal with, change a few constants. emg not working still

adelmotheboi
authored and
FibonacciDude
committed
new things (not much)

adelmotheboi
authored and
FibonacciDude
committed
little changes again, should be nothing serious

adelmotheboi
authored and
FibonacciDude
committed
it works somewhat better (I made it use all test reps for validation and all reps for test

adelmotheboi
authored and
FibonacciDude
committed
voting window

adelmotheboi
authored and
FibonacciDude
committed
new changes to pretrain (not done was adding ability to use all windows for the voting)

adelmotheboi
authored and
FibonacciDude
committed
new code, a mess

adelmotheboi
authored and
FibonacciDude
committed
new changes, contrastive (wrong and wrong)

adelmotheboi
authored and
FibonacciDude
committed
more incorrectness

adelmotheboi
authored and
FibonacciDude
committed
contrastive works, but only for one hot (see why other is hard)

adelmotheboi
authored and
FibonacciDude
committed
Okay, this actually works.

adelmotheboi
authored and
FibonacciDude
committed
Inter-subject myoelectric control. This code works

adelmotheboi
authored and
FibonacciDude
committed
It was using the data from the second database, not third

adelmotheboi
authored and
FibonacciDude
committed
update const

adelmotheboi
authored and
FibonacciDude
committed
something changed

adelmotheboi
authored and
FibonacciDude
committed
branch

adelmotheboi
authored and
FibonacciDude
committed
branch, crossval

adelmotheboi
authored and
FibonacciDude
committed
m

adelmotheboi
authored and
FibonacciDude
committed
Data used for analysis

adelmotheboi
authored and
FibonacciDude
committed
Removal of irrelevant files

adelmotheboi
authored and
FibonacciDude
committed

Commits on Mar 22, 2022

Commits on Mar 30, 2022

Commits on Apr 5, 2022

Commits on Jun 2, 2022

Commits on Jul 16, 2022

Update README.md
FibonacciDude
authored

Commits on Jul 17, 2022

Commits on Jan 20, 2024

Update README.md
FibonacciDude
authored

Commits on Jun 24, 2024

Update README.md
FibonacciDude
authored