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

Axl atom type loss review #91

Closed
wants to merge 75 commits into from
Closed

Axl atom type loss review #91

wants to merge 75 commits into from

Conversation

rousseab
Copy link
Collaborator

@rousseab rousseab commented Nov 4, 2024

Here I revamp the calculation of the atomic type loss, making sure that the predicted probabilities p(a_{t-1} | a_t) are normalized to sum to 1.

sblackburn-mila and others added 30 commits October 23, 2024 13:29
@sblackburn-mila
sample noisy version of the atom types
b4ce805
@sblackburn-mila
adding beta_t, q_t and others in the noise scheduler
1f18f10
@sblackburn-mila
broadcast function for q_t and q_bar_t
e30d7c0
@sblackburn-mila
loss as AXL and d3pm loss module
059e454
@sblackburn-mila
big chunk in the lightning base model to support atom type training
9df68c6
@sblackburn-mila
update instantiate diffusion model to AXL model
cabb9c6
@sblackburn-mila
score network base update for AXL
3f473af
@rousseab
Fix broken code.
8468bd8
@rousseab
Renamed various things related to drawing diffusion samples.
981a54b
@rousseab
Renamed various things related to drawing diffusion samples.
32e69c4
@sblackburn-mila
analytical score model and mace diffusion
65a42f0
@rousseab
Refactored away the weasel word "sampler".
b1c29e6
@rousseab
Fix executable path.
945eeb0
@rousseab
Use the name "noiser".
5c40e79
@rousseab
Fix data folder imports.
0b699d8
@rousseab
Fixing imports
8da066c
@rousseab
remove broken script
99cba10
@rousseab
Fixing imports
76c99a3
@rousseab
More consistent name.
ecab049
@rousseab
Put NoiseParameters in its own module.
3b9439a
@rousseab
Systematically use the ExplodingVariance object instead of recoding t…
1b38d87 
…he same stuff over and over.
@rousseab
No more dump.yaml after test.
e4d06d1
@rousseab
No more dump.yaml after test.
7162db2
@sblackburn-mila
egnn with AXL thing
fcdfe7c
@sblackburn-mila
various namedtupl snafu
50faeea
@sblackburn-mila
more fixes
b86ac2d
@rousseab
Merge most issues.
aff150d
@rousseab
Merged conflicts in the NoiseScheduler class.
b20bf38
@rousseab
Noisers.
f4c9420
@rousseab
Noisers
ffec518
sblackburn-mila and others added 28 commits October 28, 2024 10:14
@sblackburn-mila
bug fix in diffusion mace score network
a78e8db
@sblackburn-mila
analytical score
530dd14
@sblackburn-mila
dataloader renaming to namespace for atom types
754d32e
@sblackburn-mila
fixing data preprocessing tests
05598eb
@sblackburn-mila
Merge remote-tracking branch 'origin/main' into refactor_for_atom_typ…
6d03d69 
…e_diffusion
@sblackburn-mila
adding unit tests for d3pm loss
a8f01c6
@sblackburn-mila
variance samplers unit tests
00f61d9
@sblackburn-mila
atom type noiser tests
2c35f85
@sblackburn-mila
d3pm utils tests
69a6930
@sblackburn-mila
tensor utils unit tests
b1f0c6f
@sblackburn-mila
test egnn upgrades
56f6339
@sblackburn-mila
testscorenetwork upgrade
1e8b869
@sblackburn-mila
fixing score network unit tests
e5d2f7e
@rousseab
A different seed, and no more nans.
72ae772
@sblackburn-mila
fixing cartesion positions in mace - solving the unit test issue...
023508a
@sblackburn-mila
code review part 1
df29fee
@sblackburn-mila
code review part 2
6e72b76
@sblackburn-mila
code review part 3
cb2520c
@sblackburn-mila
first pass complete on code review
db026b7
@sblackburn-mila
saturday morning breakfast cereal comments
d1122a2
@sblackburn-mila
fixing variance_sampler and various unit tests
1255222
@sblackburn86
Merge pull request #90 from mila-iqia/refactor_for_atom_type_diffusion
f4ef300 
initial refactor & atom types diffusion methods
@rousseab
Sorting imports, passing linting tests.
cb544d8
@rousseab
More granularity in the loss modules. Also, more isort.
8a82129
@rousseab
Linting.
926eeca
@rousseab
More granular loss testing.
0535839
@rousseab
More granular loss testing.
9cfd069
@rousseab
Revamped atomic type loss function.
3a625b0
@rousseab rousseab requested a review from sblackburn86 November 4, 2024 20:06
@rousseab rousseab closed this Nov 4, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@sblackburn86 sblackburn86 Awaiting requested review from sblackburn86

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@rousseab @sblackburn-mila @sblackburn86