Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

FAENet++ #66

Open
wants to merge 126 commits into
base: main
Choose a base branch
from
Open

FAENet++ #66

wants to merge 126 commits into from

Conversation

Ramlaoui
Copy link
Collaborator

@Ramlaoui Ramlaoui commented Jul 7, 2024 •
edited
Loading

Modifications for FAENet++ project with the following main changes:

  • New Canonicalization methods and config specification to support all of the added modules
  • Support for relaxations with on the fly graph rebuilding supporting transforms
  • Added new modules and models: Ewald on FAENet, SchNet, DPP. Support for EquiformerV2
  • Transfer learning (eg from S2EF models to IS2RE with backbone freezing)
  • Clean notebooks
  • Clean config files, reduce to min number with max information to reproduce results
  • Solve conflicts with Basile branch
  • Add documentation to readme -- explaining briefly how to use these extensions (e.g. canonicalisation)
  • Commit test_model_symmetries in single_trainer to account for qm7x
  • Clean unnecessary files
  • Add and update function comments, check everything
  • Verif that data_transform is readable
  • Add config for one (or a few) cano

Ramlaoui and others added 30 commits December 5, 2023 04:38
@Ramlaoui
ewald message passing raw
6f81484
@Ramlaoui
notebook to compare activations between models
eaf80a4
@Ramlaoui
quick addition of ewald to schnet
3f444e7
@Ramlaoui
update activations notebook
07ba182
@theosaulus
added indications for future implementations
4b88715
@theosaulus
added dirty implementation of learned canonicalisation function
23f2c6a
@theosaulus
removed comments
3750d1c
@theosaulus
bug fix
38ae651
@Ramlaoui
modify configs for ewald
d8b80e3
@Ramlaoui
ewald model for schnet and faenet
e520dea
@Ramlaoui
New tasks for S2EF relaxations
6ecd7d5
@Ramlaoui
update relaxations to current ocp updates
6879391
@Ramlaoui
adapt for relaxations
56796e7
@Ramlaoui
update logger name for ft
c758b0b
@Ramlaoui
update task
1306850
@Ramlaoui
add _unwrapped_model for relaxations
7382cb5
@Ramlaoui
fix results directory for the current config
5ce0486
@Ramlaoui
otf_graph in faenet class
bb4d388
@Ramlaoui
fix bug for relaxations metrics
eceafc5
@Ramlaoui
revoke debug options
ef619f8
@theosaulus
added othornormalization
dda4c54
@theosaulus
modif sh
4c38daa
@Ramlaoui
use directories with debug, revert after
9dcd071
@theosaulus
init separated UntrainedCanonicalisation class
0e2c35c
@theosaulus
currently not working - nodes with no edges
c6a09bf
@Ramlaoui
dive deeper in analysis
da5ba0a
@Ramlaoui
change notebook strucctures + studies on mispredictions
e3589ef
@theosaulus
added canonicalisation support - not running
d33e514
@Ramlaoui
uniform canocalisation calls
6e9dfc3
@Ramlaoui
fix bugs - now working
7a16a32
Ramlaoui and others added 27 commits May 19, 2024 21:09
@Ramlaoui
update plots order
eccdd7a
@Ramlaoui
add sum of interaction layers
d9cf179
@theosaulus
bug fix
cc7cc04
@theosaulus
bug fix
bb8c6f7
@Ramlaoui
fix fa when the positions are on gpu
64887f6
@Ramlaoui
transforms for relaxation in the case of fa
a3fd7dc
@Ramlaoui
update notebook
bc24537
@Ramlaoui
avoid splitting the batch when no trainable cano
88abff3
@theosaulus
fix + gram schmidt
384f910
@theosaulus
direct forces and support for qm7x sym
48aec96
@theosaulus
minor changes
5141584
@theosaulus
clean + added sign-inv E3-equiv net
1066e48
@theosaulus
cleaning
8b1a770
@theosaulus
added support for sign invariant E(3)-equivariant cano
2cece77
@Ramlaoui
add extra argument used on energy forward by the trainer
8eb0f73
@Ramlaoui
allow to freeze backbone for transfer learning
c0d9200
@Ramlaoui
linting
1f75cf1
@Ramlaoui
revert debug case to defau+t + lint
8647ef2
@Ramlaoui
Merge branch 'ewald-sum' into equiformer
2af5acd
@Ramlaoui
add ewald to dpp
6bc378e
@theosaulus
cleaning
640166c
@Ramlaoui
Merge branch 'equiformer' into learn-cano-func
52eabe6
@Ramlaoui
Merge pull request #65 from RolnickLab/learn-cano-func
b750be1 
Merge learned canonicalization into equiformer branch
@Ramlaoui
support for ewald with dpp and qm9
4ead6db
@Ramlaoui
fix memory leak for relaxations without transforms
c40b5aa
@Ramlaoui
add compatibility for old frame_averaging arguments in configs
e9c39a8
@Ramlaoui
Merge pull request #57 from RolnickLab/equiformer
442be83 
Fix relaxations for FAENet and add EquiformerV2 model
@Ramlaoui
Copy link
Collaborator Author

Ramlaoui commented Jul 8, 2024

  • [] Use Black
  • [] Regroup three notebooks into one
  • [] Handle S2EFtoIS2RE tasks
  • [] Remove hidden breakpoints
  • [] test_model_symmetries
  • [] unnecessary comments
  • [] Keep version of run_relaxations
  • [] Document and comment everything in the PR space
  • [] Remove unnecessary functionalities
  • [] Mention side effects of new functions
  • [] Revert base configs
  • [] List paper models and make sur the scores are the same (with wandb link)
  • [] Clean test file and add new experiments

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@Ramlaoui @theosaulus