[WIP] Refactor development regularization cocktails#161

Closed

ArlindKadra wants to merge 140 commits intomasterfrom

refactor_development_regularization_cocktails

Commits on May 4, 2021

First push for Mix/cut regularization

franchuterivera
authored and
ravinkohli
committed
Fix mypy

franchuterivera
authored and
ravinkohli
committed
Initial implementation of adversarial training

ArlindKadra
authored and
ravinkohli
committed
Modifying the code to have activation controlled batch normalization

ArlindKadra
authored and
ravinkohli
committed
Fix unused imports, fix coding style checks, fix unit test

ArlindKadra
authored and
ravinkohli
committed
Adding activation controlled weight decay, updating the style for code style check

ArlindKadra
authored and
ravinkohli
committed
Commit for passing style check

ArlindKadra
authored and
ravinkohli
committed
Style check try 2

ArlindKadra
authored and
ravinkohli
committed
Bug fix

ArlindKadra
authored and
ravinkohli
committed
Adding unit test for adversarial trainer

ArlindKadra
authored and
ravinkohli
committed
Adding code for activation controlled skip connections, with the additional choice of shake-shake and shake-drop being hyperparameters as the choice for the multi branch networks

ArlindKadra
authored and
ravinkohli
committed
Bug fix for the failing tests

ArlindKadra
authored and
ravinkohli
committed
Adding better conditions

ArlindKadra
authored and
ravinkohli
committed
Try at a fix

ArlindKadra
authored and
ravinkohli
committed
Temporary fix for the failing test

ArlindKadra
authored and
ravinkohli
committed
Failing code check

ArlindKadra
authored and
ravinkohli
committed
Failing code check v2

ArlindKadra
authored and
ravinkohli
committed
Add new update to fix break

ArlindKadra
authored and
ravinkohli
committed
Flake8 coding style fix

ArlindKadra
authored and
ravinkohli
committed
Unit test update

ArlindKadra
authored and
ravinkohli
committed
Removing duplicate unit test

ArlindKadra
authored and
ravinkohli
committed
In progress
ravinkohli
committed
Add cyclic property to lr scheduler and use_swa to trainer
ravinkohli
committed
swa working, se in progress
ravinkohli
committed
Fixed bug in update model with swa model, add predict with snapshot ensemble; todo: add tests for both
ravinkohli
committed
ADD tests for SWA, SE
ravinkohli
committed
Add pytest_mock to test dependencies
ravinkohli
committed
lookahead in progress
ravinkohli
committed
add lookahead hyperparameters
ravinkohli
committed
ADD tests for LookAhead
ravinkohli
committed
Fix flake mypy
ravinkohli
committed
Fix error in test
ravinkohli
committed
Added default for lookahead config
ravinkohli
committed
Addressed comments from francisco
ravinkohli
committed
fixed issues after merge
ravinkohli
committed
Add displayable path to model file assertion
ravinkohli
committed
add delay to display path tree
ravinkohli
committed
Add test for non cyclic scheduler
ravinkohli
committed
Fix errors in Adversarial training
ravinkohli
committed
Fix pickling error for swa model
ravinkohli
committed
Fix issues after rebase from refactor_development
ravinkohli
committed
Fix flake
ravinkohli
committed
Added n_restarts as hyperparameter for CosineAnnealing
ravinkohli
committed
fix bug with ealy stopping and swa
ravinkohli
committed
cont...
ravinkohli
committed
add min for 1
ravinkohli
committed
Addressed comments from shuhei, better documentation
ravinkohli
committed
Addressed comments from arlind, change in T_mul and T_0 calculations
ravinkohli
committed
Added debug information for API
ravinkohli
committed
Fix flake
ravinkohli
committed
Fix import
ravinkohli
committed
made test deterministic for feature preprocessing
ravinkohli
committed
Fix bug in parsing log
ravinkohli
committed
convert to int
ravinkohli
committed
Fix bug in testing
ravinkohli
committed
Fix flake after rebase

franchuterivera
authored and
ravinkohli
committed
Updating search space (#156)

ArlindKadra
authored and
ravinkohli
committed
bug fixes

ArlindKadra
authored and
ravinkohli
committed
fixing code style checks

ArlindKadra
authored and
ravinkohli
committed
bug fix for use_pynisher in the base pipeline

ArlindKadra
authored and
ravinkohli
committed
bug fix

ArlindKadra
authored and
ravinkohli
committed
Adding constant clause in the testing module for hyperparameter ranges, removing use_pynisher since it is not used anymore, hack for the moment to bypass illegal hyperparameter ranges for conditions

ArlindKadra
authored and
ravinkohli
committed
Fixing flake8 fail

ArlindKadra
authored and
ravinkohli
committed
flake8 fix

ArlindKadra
authored and
ravinkohli
committed
flake8 fix

ArlindKadra
authored and
ravinkohli
committed
Updating implementation for tabular regression

ArlindKadra
authored and
ravinkohli
committed
Addressing the comment from Ravin

ArlindKadra
authored and
ravinkohli
committed
Adding checks for the possible different value range of multibranch_choice

ArlindKadra
authored and
ravinkohli
committed
Removing comment since it is not relevant anymore

ArlindKadra
authored and
ravinkohli
committed
Removing unecessary code that transforms to cpu, while it is on cpu

ArlindKadra
authored and
ravinkohli
committed
Fixing buggy implementation of the network head with constant updates

ArlindKadra
authored and
ravinkohli
committed
Flake8 previous implementation style fix

ArlindKadra
authored and
ravinkohli
committed
Flake8 previous implementation style fix

ArlindKadra
authored and
ravinkohli
committed
Flake8 previous implementation style fix

ArlindKadra
authored and
ravinkohli
committed
Update for mypy error

ArlindKadra
authored and
ravinkohli
committed
Update for mypy error

ArlindKadra
authored and
ravinkohli
committed
Updating implementation

ArlindKadra
authored and
ravinkohli
committed
Temporary bug fix

ArlindKadra
authored and
ravinkohli
committed
fixing mypy

ArlindKadra
authored and
ravinkohli
committed
Implementation fix for constant updates to skip connections. Multibranch choice was dependant on an illegal value

ArlindKadra
authored and
ravinkohli
committed
Fixing the implementation for weight decay in the case of fixed updates. The reason is the illegal value for weight decay when use weight decay is off

ArlindKadra
authored and
ravinkohli
committed
Fixing implementation of lookahead and snapshot ensembling for fixed search space updates

ArlindKadra
authored and
ravinkohli
committed
Update unit test for debugging purposes

ArlindKadra
authored and
ravinkohli
committed
Working fit_pipeline method, with test and example
ravinkohli
committed
Fixed bug in tabular regression
ravinkohli
committed
Fix bug in example single configuration
ravinkohli
committed
Addressed comments from Fransisco, making configuration required in fit_pipeline
ravinkohli
committed
Add configuration to example, fix in get_dataset
ravinkohli
committed
Fix mypy
ravinkohli
committed
Address Arlinds comment for task type documentation
ravinkohli
committed
Change way to get configuration fr using fit_pipeline
ravinkohli
committed
fix flake
ravinkohli
committed
update setup.py
ravinkohli
committed
Updating implementation of the reg cocktails so that it is compatible with fixed search space updates

ArlindKadra
authored and
ravinkohli
committed
Turning off small preprocessing and also removing the TAE assertion for when the budget is larger than 100

ArlindKadra
authored and
ravinkohli
committed
Decreasing the dataset size for the fit pipeline since it is giving time out errors with its current implementation

ArlindKadra
authored and
ravinkohli
committed
Updating implementation

ArlindKadra
authored and
ravinkohli
committed
Fixing bug where the number of layers is actually one less from the values it is given

ArlindKadra
authored and
ravinkohli
committed
Updating implementation

ArlindKadra
authored and
ravinkohli
committed
Create fit evaluator, no resampling strategy and fix bug for test statistics
ravinkohli
committed
Fix mypy and flake
ravinkohli
committed
Fix check for X_test while making test data loader
ravinkohli
committed
fix bug in lookahead hyperparameters where lookahead was repeated for the hyperparameter name
ravinkohli
committed
Make passing tests in api easier
ravinkohli
committed
Fix bug in trainer weighted loss code for regression
ravinkohli
committed
Changed to iris in example
ravinkohli
committed
Increased time in regression and epochs for pipeline_score
ravinkohli
committed
Additional metrics during train (#194)
ravinkohli
committed
Fix mypy and flake after rebase, added random state to mixup and cutout and changs no resampling for new code
ravinkohli
committed
fix bug in setup.py
ravinkohli
committed

Commits on May 7, 2021

Fixing issues with imbalanced datasets (#197)
ArlindKadra
authored

Commits on May 8, 2021

Add seed to dataset creation (#202)
ravinkohli
authored

Commits on May 10, 2021

fix bug in backend next_num_run (#203)
ravinkohli
authored

Commits on May 11, 2021

Commits on May 12, 2021

Fixing the scheduler to update for epochs and not batch iterations (#210)
ArlindKadra
authored

Commits on May 14, 2021

Commits on May 15, 2021

robustness in get_next_num_run (#215)
ravinkohli
authored

Commits on May 17, 2021

Commits on May 21, 2021

Fix bugs in cutout training (#233)
ravinkohli
authored

Commits on Jun 3, 2021

Cocktail hotfixes (#245)
ArlindKadra
authored

Commits on Sep 13, 2021

Commits on Sep 14, 2021

[refactor] Address Shuhei's comments
nabenabe0928
committed

Commits on Sep 15, 2021

[doc] Add referencing to each regularization techniques
nabenabe0928
committed

Commits on Sep 21, 2021

Commits on Sep 28, 2021

Addressing the constant training subset size in the test api
ArlindKadra
committed

Commits on Oct 21, 2021

Bug fixes (#249)

authored

Commits on Dec 7, 2021

[FIX] Passing checks (#298)

ravinkohli
and
nabenabe0928
authored