Adding low-rank covariance - #4

Open

vivianwhite wants to merge 77 commits into

devfrom

new_experiments

Commits on Jun 28, 2024

adding files from muawi_resnets
vivianwhite
committed
fb83f37
removing comments and unused lines
vivianwhite
committed
0f10d2c
remove comments and unused lines
vivianwhite
committed
e3ab5ba
fixed and tested conv_modules
vivianwhite
committed
8411f74
add whitespace
vivianwhite
committed
fe6cbc7
add whitespace
vivianwhite
committed
a472f7d
trying to setup define_models
vivianwhite
committed
2ec009e
fixed models dir
vivianwhite
committed
122c758
Delete models directory
vivianwhite
committed
28841fe
add utils file
vivianwhite
committed
7ddc12c
fix white spaces
vivianwhite
committed
55670ba
fixing V1 cifar10 model
vivianwhite
committed
a779085
add replace_linears function
vivianwhite
committed
b873254
adding recursive functions script
vivianwhite
committed
204bf3a
some updates, more to come tomorrow
vivianwhite
committed
2aeaf98
Delete refactor/resnet.py
vivianwhite
committed
6a236c5
update define_model options
vivianwhite
committed
a9317a7
normalization back to normal
vivianwhite
committed
11d9477
updates anda adding turn_off_grad function
vivianwhite
committed
6905850
adding V1-init functionality
vivianwhite
committed
aa28aab
updating v1-init
vivianwhite
committed
e537fda
cleanup
vivianwhite
committed
15b8cd5
initial commit of refactored rainbow (have not run anything yet)

MuawizChaudhary
authored and
vivianwhite
committed
29978b4
minimal changes to ensure this script runs

MuawizChaudhary
authored and
vivianwhite
committed
82180c4
cleanup, modified print statements, added seed argument, removed width 8, removed evaluation on affine models, etc etc

MuawizChaudhary
authored and
vivianwhite
committed
09aade2
removed affine argument

MuawizChaudhary
authored and
vivianwhite
committed
6c6dcc9
moved recursive modules around

MuawizChaudhary
authored and
vivianwhite
committed
15422ab
added recurse_preorder. made each replace function a wrapper that returns the modifed network (tho the network is modified, so this is reduandant)

MuawizChaudhary
authored and
vivianwhite
committed
1397314
removed fact_2_conv

MuawizChaudhary
authored and
vivianwhite
committed
44da047
changes

MuawizChaudhary
authored and
vivianwhite
committed
a4bf746
did TODOs
vivianwhite
committed
fff0899
noting origin of code
vivianwhite
committed
f36a459
added class conditional decerators. added looping over multiple callbacks, could be better coded

MuawizChaudhary
authored and
vivianwhite
committed
38de471
Revert to 09f4833

MuawizChaudhary
authored and
vivianwhite
committed
0cca6a1
seperated rainbow functionality from runner script

MuawizChaudhary
authored and
vivianwhite
committed
42f7542
changes to rainbow sampling so we can sample multiple times and calculate variance

MuawizChaudhary
authored and
vivianwhite
committed
fff82a5
refactored a decent amount of rainbow sampling

MuawizChaudhary
authored and
vivianwhite
committed
1dd2048
fact-conv changes

MuawizChaudhary
authored and
vivianwhite
committed
5f44973
reduce amount of code in rainbow

MuawizChaudhary
authored and
vivianwhite
committed
657db4f
initalize with empty_like instead

MuawizChaudhary
authored and
vivianwhite
committed
c680aea
added verbose option

MuawizChaudhary
authored and
vivianwhite
committed
edda0b4
removed my beloved og

MuawizChaudhary
authored and
vivianwhite
committed
68e84e4
deleted learnable_cov.py which had been renamed to V1_covariance.py

MuawizChaudhary
authored and
vivianwhite
committed
07803e0
width scaling should be first

MuawizChaudhary
authored and
vivianwhite
committed
6594292
making rainbow sampling more like a properly called class. similar to a kymatio front end.

MuawizChaudhary
authored and
vivianwhite
committed
c078cfc
removed specification of device from init, rely instead on layer to provide that.

MuawizChaudhary
authored and
vivianwhite
committed
642c480
formatting changes

MuawizChaudhary
authored and
vivianwhite
committed
d3116ec
formatting

MuawizChaudhary
authored and
vivianwhite
committed
992e98b
added README.md

MuawizChaudhary
authored and
vivianwhite
committed
0d0d60f
added citation to README.md

MuawizChaudhary
authored and
vivianwhite
committed
2b48a03
Update README.md

MuawizChaudhary
authored and
vivianwhite
committed
9dc5666
renaming folder
vivianwhite
committed
01b39c8
Delete layers directory
vivianwhite
committed
c92f6ac
Delete refactor directory
vivianwhite
committed
e05bbcd
moving scripts
vivianwhite
committed
0a1c34b
Delete V1-models/scripts/s_f_sweep directory
vivianwhite
committed
428ae58
updating LC RSN scripts
vivianwhite
committed
c6fa1a4
removing hardcoded stuff
vivianwhite
committed
345a598
fixed save dir
vivianwhite
committed
1c5295f
fixed save dir
vivianwhite
committed
1b080e9
new experiments

MuawizChaudhary
authored and
vivianwhite
committed
2ae3414
implementing diag factconv2d
vivianwhite
committed
378cbd9
minor changes

MuawizChaudhary
authored and
vivianwhite
committed
2ffc071
conv_modules.py
vivianwhite
committed
74aa5a3
add diagchan, call by setting args.net='resnet18-diagchan'
vivianwhite
committed
2366ef8
added low-rank setup
vivianwhite
committed
c720d95
low-rank experiments
vivianwhite
committed
4a04e1d
adding covariance module
vivianwhite
committed
0342b2e
adding learnable low-rank diagonal covariance
vivianwhite
committed
a8c7b5d
adding low rank experiments
vivianwhite
committed
ec927a4

Commits on Jun 29, 2024

updating covariance modules
vivianwhite
committed
5cceccb

Commits on Jul 1, 2024

logging param count
vivianwhite
committed
5513509
update turn_off_covar_grad function
vivianwhite
committed
f5f70f9
adding LearnableCov.py back
vivianwhite
committed
0a2e945

Commits on Jul 15, 2024

adding nonlinearities study
vivianwhite
committed
1f9f43c

Commits on Jul 20, 2024

adding resnext model
vivianwhite
committed
db9ba1b

Commits on Aug 22, 2024

added diagonally dominant experiment and updated scripts
vivianwhite
committed
de9820b