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

Lattice generator #116

Open
wants to merge 91 commits into
base: dev_lattice_diffusion
Choose a base branch
from
Open

Lattice generator #116

wants to merge 91 commits into from

Conversation

sblackburn86
Copy link
Collaborator

In this PR, I introduce the pieces of code required to do a full lattice diffusion experiment. Namely:

  • the generation components for the lattice diffusion with the langevin predictor-corrector and adaptive corrector
  • modifications to the mlp network to take the lattice parameters as input
  • minimal fixes to the egnn & mace architecture to output a better dummy variable
  • changing the lattice parameters noising process to fit in an exploding variance scheme
  • fixes to the dataloader processor component to convert unit cell to lattice parameters
  • adding a convertor from lattice parameters to unit cell vectors for LAMMPS oracle

A few things are still required in a future PR:

  • handling of the lattice parameters in EGNN and MACE as inputs and outputs
  • a smarter way of managing "bad" (i.e. noisy) lattice parameters for calculating distances between cartesian coordinates in a graph (radial cutoff)
  • handling a different noise scheduler for X and L which implies modifications to the sigma input of the score network
  • fixing remaining broken unit tests

@rousseab
Compute log wrapped gaussians.
439d7d8
@rousseab
Permutation over atomic indices.
12d7871
@rousseab
Code a trivial factorial function to avoid importing something else.
83e99b8
@rousseab
Improved analytical score network.
58c3734
@rousseab
Refactor name of datamodule module.
ecfe3ce
@rousseab
Rename lammps preprocessor code.
b1f7255
@rousseab
Better name.
51a612c
@rousseab
Align test name.
7f10189
@rousseab
Better naming and clean data module instantiation.
d6fc645
@rousseab
A new data module for on-the-fly Gaussian datasets.
14825ca
@rousseab
The ability to instantiate a Gaussian data module.
8d54ef4
@rousseab
Make it easier to instantiate the analytical score network from a con…
bfb8506 
…fig file.
@rousseab
Add analytical score network to possible models to instantiate.
99bbaf4
@rousseab
Fix exp details.
778b4d9
@rousseab
Cleaner data module instantiation.
27c566a
@rousseab
More comments for debug server.
714b583
@rousseab
Make it possible to turn off the optimizer.
bb7e1f3 
This way, we can pass the analytical score network through the main code.
@rousseab
Class to plot the scores along a direction.
89fb2ad
@rousseab
Fancier score viewer.
9da231e
@rousseab
Bring scores back to cpu.
d0d0b23
@rousseab
New callback to show scores along a path.
f335322
@rousseab
Fix assert error comment.
4071d3a
@rousseab
Ensure that the EGNN score network works in 1D, 2D and 3D.
98afd87
@rousseab
an example using the on-the-fly data module
fd25a21
@rousseab
Utility class to compute sigma and its derivative.
0907f02
@rousseab
A utility module to only compute sigma(t).
da4040c
@rousseab
A dummy score network where we can compute the analytical derivatives…
e05e81f 
… by hand.
@rousseab
A score fokker planck regularizer.
674d4a6
@rousseab
A method to compute the loss contribution from FP.
b83d7a6
@rousseab
Pick up regularizer parameters from config.
d08b382
Simon Blackburn and others added 27 commits January 10, 2025 08:05
test sigma scaling function
8cf3363
axl model tests
6ca5fc0
forget to fix a typo
c52d069
@rousseab
Corrected a minor import issue.
f183212
code review for lattice noiser
34016a0
fixing bugs captured by code review
d19ee21
fixing bugs captured by code review part 2
1637566
fixes to unit tests for code review
943992d
code review
42e3e2d
update to training script
e563996
removing dependencies on alpha_bar and average density in the lattice…
76f3187 
… noiser
lattice noiser parameters update
ee91649
gaussian score as a vanilla variance exploding
3e6078d
updates to axl base model
ffeb77c
util function to compute number of lattice parameters
f54fee7
changes in mlp to take lattice parameters as input
5025b4c
circular import somehow..
5883a50
adding lattice diffusion in langevin PC and adaptative corrector gene…
4ebf002 
…rators
fixing generators unit tests
be6924a
missing ode solver fixes
f12b5ab
cleaning up an unnecessary unit_cell
6b5e5f3
other superfluous cell dimensions usage
0b7796c
fix in generic langevin generator
b846708
testing lattice updates in generic PC generator
7809437
adding checks on lattice updates in langevin PC & adaptive generators
af157b5
fixing some unit tests for consistency regularizers by removing unit_…
a6e5948 
…cell obsolete kwarg
fixing a bunch of unit tests and adding a few useful features such as…
7ec695a 
… parsing lattice parameters from dataset files
@sblackburn86 sblackburn86 changed the base branch from main to dev_lattice_diffusion January 12, 2025 18:24
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
@sblackburn86 @rousseab