Skip to content

Disconnected GNN: Catalyst-Adsorbate separator #44

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.

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

Open
wants to merge 134 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
134 commits
Select commit Hold shift + click to select a range
42553e7
no idea what I changed
AlvaroC-ML Jun 12, 2023
fe76d29
Created a file with disconnected gnn and an edge deleter
AlvaroC-ML Jun 13, 2023
064699f
vectorized implementation of edge deleter
AlvaroC-ML Jun 13, 2023
e897615
Changed config file to run faenet vs disconnected.
AlvaroC-ML Jun 14, 2023
e29b571
Created new flag 'is_disconnected'
AlvaroC-ML Jun 14, 2023
83fa9a9
Corrected some typos.
AlvaroC-ML Jun 14, 2023
afa0ae0
Deleted disconnected model, as now it is a feature on data transforms…
AlvaroC-ML Jun 14, 2023
fca863c
Fixed problem that the function assumed there were no tag 0 vertices.
AlvaroC-ML Jun 14, 2023
78e829e
Fixed a mistake where not all edge attributes are filtered
AlvaroC-ML Jun 15, 2023
e91c007
Modified the is_disconnected flag to be a variable so it can be used …
AlvaroC-ML Jun 21, 2023
0719a7f
fixed some typos
AlvaroC-ML Jun 21, 2023
61887f3
Deleted disconnected model, and copied best configs to my experiments…
AlvaroC-ML Jun 26, 2023
cec7663
Added flag for only using a subset of the dataset with a unique datap…
AlvaroC-ML Jun 26, 2023
a2e6b33
Added if statement in base trainer to use only the desired subset of …
AlvaroC-ML Jun 26, 2023
82c7c59
fixed some typos
AlvaroC-ML Jun 26, 2023
431a7c7
Fixed a bug when closing the dataset and modified config files to mak…
AlvaroC-ML Jun 28, 2023
ab814b5
Added graph splitter. Still need to reindex edges to not cause errors…
AlvaroC-ML Jul 3, 2023
540a485
Implemented: getting separate predictions for ads, cats. Their output…
AlvaroC-ML Jul 3, 2023
59928cf
Middle point before implementing a needed modification to make skip_c…
AlvaroC-ML Jul 5, 2023
1e6346f
Skip_co correctly implemented
AlvaroC-ML Jul 5, 2023
49cb00c
Set it so using the disconnected model turns on the is_disconnected f…
AlvaroC-ML Jul 5, 2023
c9d3c84
Changed the name of disconnected model to dependent as to make psace …
AlvaroC-ML Jul 7, 2023
34e232c
Created a new dataset file where graphs will be split before transfor…
AlvaroC-ML Jul 7, 2023
a997c04
Implemented graph splitter. Currently debugging modification to data_…
AlvaroC-ML Jul 7, 2023
39fe297
Now that the new dataset is implemented, we correct so the dependent …
AlvaroC-ML Jul 10, 2023
f2f89ef
Renamed some models, started debugging a simplified version of model 3
AlvaroC-ML Jul 11, 2023
7854976
Created debug file to debug from an interactive node
AlvaroC-ML Jul 12, 2023
6435dcc
Finished implementing basic version of model indfaenet
AlvaroC-ML Jul 12, 2023
6a5f3d8
Fixed small typo
AlvaroC-ML Jul 12, 2023
0985b83
fixing bug
AlvaroC-ML Jul 12, 2023
50687f5
Bug of model 3 has been fixed.
AlvaroC-ML Jul 13, 2023
b5a2414
Rewrote a line to make it nicer.
AlvaroC-ML Jul 13, 2023
ef44257
Finished implementing indfaenet
AlvaroC-ML Jul 14, 2023
a37986c
Implemented disconnected_mlp in indfaenet
AlvaroC-ML Jul 17, 2023
fd26418
Cleaned up some code, and set it ready for some hyperparameter tweaking
AlvaroC-ML Jul 19, 2023
745351f
Deleted a wild point trace.
AlvaroC-ML Jul 19, 2023
3020a46
Implemented a first version of the transformer
AlvaroC-ML Jul 26, 2023
c617918
Changed implementation to a better one
AlvaroC-ML Jul 26, 2023
684327d
Added positional encoding to allow the model to distinguish between a…
AlvaroC-ML Jul 27, 2023
9c83bec
Forgot to renable inference time calculation
AlvaroC-ML Jul 27, 2023
31ec588
Modified training files.
AlvaroC-ML Jul 27, 2023
9a71546
fixed typo in experiments folder
AlvaroC-ML Jul 28, 2023
acf2d8a
Started implementing transformer interaction faenet, init mostly done
AlvaroC-ML Jul 28, 2023
5aead8a
Did good progress on the implementation of energy forward, modified b…
AlvaroC-ML Jul 28, 2023
189f7ce
Implemented everything up to the actual transformer block.
AlvaroC-ML Jul 28, 2023
e8af33f
Finished implementation of transformer interactions, and will redo tr…
AlvaroC-ML Jul 30, 2023
dc7d717
Changed initial learning rate for all dataset training
AlvaroC-ML Jul 30, 2023
2d45a38
Made transformer larger
AlvaroC-ML Jul 30, 2023
f5883a7
Readied an experiment
AlvaroC-ML Jul 30, 2023
7a223ad
Started implementing a second type of transformer-like block.
AlvaroC-ML Jul 31, 2023
e7cee7a
Finished correcting a HUGE bug on attention based models.
AlvaroC-ML Aug 1, 2023
181ee53
Changed training file.
AlvaroC-ML Aug 1, 2023
c0cec2a
Improved the implementation of tifaenet to use sparse matrices. It st…
AlvaroC-ML Aug 4, 2023
1120d03
Implemented attention tifaenet through sparse matrix. Now changing ba…
AlvaroC-ML Aug 7, 2023
95a415c
Finished implementing heterogeneous graphs dataset. Most of the chang…
AlvaroC-ML Aug 10, 2023
98a049f
Code worked fine. Implemented the requirement of specifying the versi…
AlvaroC-ML Aug 10, 2023
6dc745b
fixed small typo
AlvaroC-ML Aug 10, 2023
558d701
fixed a mistake in coding
AlvaroC-ML Aug 10, 2023
8ccb813
Forgot a set_trace
AlvaroC-ML Aug 10, 2023
0ca9573
Fixed mistake where version 2 wasn't being used. Added attention head…
AlvaroC-ML Aug 11, 2023
5e1485c
Small typo
AlvaroC-ML Aug 11, 2023
b7cfdf2
Fixed another typo
AlvaroC-ML Aug 11, 2023
a249d35
fixing some last mistakes
AlvaroC-ML Aug 11, 2023
eb3f71d
Forgot an ipdb set trace.
AlvaroC-ML Aug 11, 2023
e2ab7ac
Testing the addition of edge weights to the graph model
AlvaroC-ML Aug 14, 2023
886e1c2
forgot to remove a set trace
AlvaroC-ML Aug 15, 2023
2ae470b
fixed typo
AlvaroC-ML Aug 15, 2023
50f794d
Better disconnected edge weights implemented
AlvaroC-ML Aug 15, 2023
9d19590
Last update before cleaning up code base.
AlvaroC-ML Aug 17, 2023
b5bdd3c
Deleted sparse matrix implementation, which never ran fast enough to …
AlvaroC-ML Aug 17, 2023
b614b71
Changed named of TIFaenet to AFaenet. ALSO I changed the largest z-in…
AlvaroC-ML Aug 17, 2023
c606298
Cleaned up the config files
AlvaroC-ML Aug 17, 2023
86d7079
Combined personalized datasets into one
AlvaroC-ML Aug 17, 2023
00b4f9f
Changed the name of tifaenet to afaenet
AlvaroC-ML Aug 17, 2023
6298539
Data base cleaned, making sure results still make sense
AlvaroC-ML Aug 18, 2023
d1df105
Fixed and improved the implementation of indfaenet
AlvaroC-ML Aug 21, 2023
4b9d648
Added other experiments to make sure they still work
AlvaroC-ML Aug 21, 2023
89c9f6d
Forgot to delete a set trace
AlvaroC-ML Aug 21, 2023
c8444f6
Fixed some errors caused when using indfaenet
AlvaroC-ML Aug 21, 2023
2c93b91
Fixed a small typo that affected the normal data mode
AlvaroC-ML Aug 22, 2023
4b161e1
Forgot to uncomment the task part. I also didn't use the is_disconnec…
AlvaroC-ML Aug 23, 2023
8b461ed
Comments added, removed a point trace.
AlvaroC-ML Sep 3, 2023
ce00cc9
modified a training file
AlvaroC-ML Sep 3, 2023
f92b8f5
Implemented all disconnected schnet models except for aschnet
AlvaroC-ML Sep 4, 2023
e7c9d58
Finished implementing aschnet and also made some modifications to the…
AlvaroC-ML Sep 5, 2023
69f84d3
Modified config files and a couple other updates
AlvaroC-ML Sep 5, 2023
bb7713c
small mistake in config file
AlvaroC-ML Sep 5, 2023
f1b3335
fixed a config file
AlvaroC-ML Sep 5, 2023
fe398c6
prepared config files for jobs
AlvaroC-ML Sep 5, 2023
3e48f96
modified afaenet slightly
AlvaroC-ML Sep 5, 2023
10d882f
Started implementing depgemnet
AlvaroC-ML Sep 11, 2023
cbfd545
modified a config file
AlvaroC-ML Sep 11, 2023
5289e1c
Corrected mistake in config file
AlvaroC-ML Sep 11, 2023
ac60cdc
implementing more
AlvaroC-ML Sep 11, 2023
f64a3ff
Finished implementing dep and ind gemnet
AlvaroC-ML Sep 11, 2023
b03446d
readied config files
AlvaroC-ML Sep 11, 2023
a54ff2d
About to start incorporating agemnet_oc
AlvaroC-ML Sep 12, 2023
936e5fe
was implemented agemnet, need to stop for today and run tests
AlvaroC-ML Sep 12, 2023
06cb33d
changing to implementing gemnet_t
AlvaroC-ML Sep 13, 2023
f68c49b
fixed a mistake in aschnet
AlvaroC-ML Sep 15, 2023
fe10633
config file editing
AlvaroC-ML Sep 15, 2023
218df21
debugging gemnet
AlvaroC-ML Sep 15, 2023
73ee5a1
readied gemnet for training
AlvaroC-ML Sep 15, 2023
c56e117
config mistake
AlvaroC-ML Sep 15, 2023
2242eec
tried to implement gemnet, didn't work. Now rerunning faenet with the…
AlvaroC-ML Sep 15, 2023
1bf534d
config file for dpp
AlvaroC-ML Sep 16, 2023
f9d4b03
implemented almost all varients of dimenetpp
AlvaroC-ML Sep 16, 2023
6ea3ed2
IMPLEMENTED ALL DPP VARIANTS
AlvaroC-ML Sep 16, 2023
1520a0a
modifications to config files
AlvaroC-ML Sep 16, 2023
1f605df
removed an ipdb
AlvaroC-ML Sep 17, 2023
76e55cb
good config file included
AlvaroC-ML Sep 20, 2023
e6fe4d4
Fixed dependent models, and attention model
AlvaroC-ML Sep 24, 2023
a3588f0
fixed a mistake on two dep files
AlvaroC-ML Sep 24, 2023
cb69712
Changed independent models to have different dimentionalities
AlvaroC-ML Sep 25, 2023
a6e867d
Fixed config files
AlvaroC-ML Sep 26, 2023
d553ab0
reproduce results: new config
AlexDuvalinho Oct 30, 2023
7908f0b
add debug file
AlexDuvalinho Oct 30, 2023
122e0a8
🧱 (base model): Make future-proof with `**kwargs`
vict0rsch Nov 13, 2023
3ce8099
♻️ (utils): Import `build_config` from main
vict0rsch Nov 13, 2023
f7169dc
✨ (utils): `make_trainer_from_dir` for gfn
vict0rsch Nov 13, 2023
bca6df9
✨ (gfn): Import from #51
vict0rsch Nov 13, 2023
0795e1a
🐛 (transforms): Fix `GraphRewiring` call to work with gfn
vict0rsch Nov 13, 2023
62a6b14
♻️ (all): Blackify
vict0rsch Nov 13, 2023
de422e7
Merge pull request #52 from RolnickLab/prepare-disconnected
vict0rsch Nov 29, 2023
b4ba938
import changes from #48
vict0rsch Nov 29, 2023
7420ab5
Merge pull request #53 from RolnickLab/disconnected-per-ads-datasets
AlexDuvalinho Jan 6, 2024
6445a5c
fine-tune / retrain depfaenet on per-ads-dataset
AlexDuvalinho Jan 6, 2024
ca497e8
improve docu of continue_from_dir
AlexDuvalinho Jan 6, 2024
3f025ac
fix setup
vict0rsch Jan 10, 2024
4bb9aac
fix adsorbates filtering
vict0rsch Jan 16, 2024
1494d91
fix ads filtering for val_ood_cat&both
vict0rsch Jan 16, 2024
7c96385
config update
AlexDuvalinho Jan 27, 2024
5c526f7
Merge branch 'disconnected_gnn' of https://github.com/RolnickLab/ocp …
AlexDuvalinho Jan 27, 2024
f2591b7
add ocdata
AlexDuvalinho Apr 16, 2024
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
47 changes: 47 additions & 0 deletions configs/exps/alvaro/10k-training.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
# MODIFY THIS ONE FOR RUNS

job:
mem: 32GB
cpus: 4
gres: gpu:rtx8000:1
partition: long
time: 15:00:00

default:
wandb_name: alvaro-carbonero-math
wandb_project: ocp-alvaro
test_ri: True
mode: train
graph_rewiring: remove-tag-0
optim:
cp_data_to_tmpdir: true
wandb-tags: 'best-config-??' # Insert what model you're running if running one by one.
frame_averaging: 2D
fa_frames: se3-random
model:
mp_type: updownscale
phys_embeds: True
tag_hidden_channels: 32
pg_hidden_channels: 64
energy_head: weighted-av-final-embeds
complex_mp: False
graph_norm: True
hidden_channels: 352
num_filters: 448
num_gaussians: 99
num_interactions: 6
second_layer_MLP: True
skip_co: concat
edge_embed_type: all_rij
optim:
lr_initial: 0.001
scheduler: LinearWarmupCosineAnnealingLR
max_epochs: 20
eval_every: 0.4
batch_size: 256
eval_batch_size: 256

runs:
- config: schnet-is2re-10k

- config: gemnet_oc-is2re-10k
47 changes: 47 additions & 0 deletions configs/exps/alvaro/all-training.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
# MODIFY THIS ONE FOR RUNS

job:
mem: 32GB
cpus: 4
gres: gpu:rtx8000:1
partition: long
time: 15:00:00

default:
wandb_name: alvaro-carbonero-math
wandb_project: ocp-alvaro
test_ri: True
mode: train
graph_rewiring: remove-tag-0
optim:
cp_data_to_tmpdir: true
wandb-tags: 'best-config-??' # Insert what model you're running if running one by one.
frame_averaging: 2D
fa_frames: se3-random
model:
mp_type: updownscale
phys_embeds: True
tag_hidden_channels: 32
pg_hidden_channels: 64
energy_head: weighted-av-final-embeds
complex_mp: False
graph_norm: True
hidden_channels: 352
num_filters: 448
num_gaussians: 99
num_interactions: 6
second_layer_MLP: True
skip_co: concat
edge_embed_type: all_rij
optim:
lr_initial: 0.0005
scheduler: LinearWarmupCosineAnnealingLR
max_epochs: 20
eval_every: 0.4
batch_size: 256
eval_batch_size: 256

runs:
- config: afaenet-is2re-all
model:
afaenet_gat_mode: v1
45 changes: 45 additions & 0 deletions configs/exps/alvaro/best-configs-all.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,45 @@
# DON'T MODIFY THIS

job:
mem: 32GB
cpus: 4
gres: gpu:rtx8000:1
partition: long
time: 15:00:00

default:
wandb_name: alvaro-carbonero-math
wandb_project: ocp-alvaro
test_ri: True
mode: train
graph_rewiring: remove-tag-0
model:
edge_embed_type: all_rij
wandb_tags: 'best-config'
optim:
batch_size: 256
eval_batch_size: 256
cp_data_to_tmpdir: true
config: faenet-is2re-all
note: 'best-config-??' # Insert what model you're running if running one by one.
frame_averaging: 2D
fa_frames: se3-random
model:
mp_type: updownscale
phys_embeds: False
tag_hidden_channels: 32
pg_hidden_channels: 64
energy_head: weighted-av-final-embeds
complex_mp: False
graph_norm: True
hidden_channels: 352
num_filters: 448
num_gaussians: 99
num_interactions: 6
second_layer_MLP: True
skip_co: concat
optim:
lr_initial: 0.0019
scheduler: LinearWarmupCosineAnnealingLR
max_epochs: 20
eval_every: 0.4
62 changes: 62 additions & 0 deletions configs/exps/alvaro/dpp-config.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,62 @@
# MODIFY THIS ONE FOR RUNS

job:
mem: 32GB
cpus: 4
gres: gpu:rtx8000:1
partition: long
time: 15:00:00

default:
wandb_name: alvaro-carbonero-math
wandb_project: ocp-alvaro
test_ri: True
mode: train
graph_rewiring: remove-tag-0
optim:
batch_size: 16
eval_batch_size: 16

runs:
# - config: dpp-is2re-10k

# - config: dpp-is2re-10k
# is_disconnected: True

# - config: depdpp-is2re-all

- config: inddpp-is2re-all
note: so that cat get old dimensions
model:
hidden_channels: 256
num_spherical: 7
num_radial: 6
out_emb_channels: 192

- config: inddpp-is2re-all
note: dimensions both smaller
model:
hidden_channels: 128
num_spherical: 4
num_radial: 3
out_emb_channels: 96

- config: inddpp-is2re-all
note: so that ads get old dimensions
model:
hidden_channels: 512
num_spherical: 14
num_radial: 12
out_emb_channels: 384

- config: inddpp-is2re-all
note: so that their average is old dimensions
model:
hidden_channels: 340
num_spherical: 9
num_radial: 8
out_emb_channels: 256

# - config: adpp-is2re-10k
# model:
# gat_mode: v1
81 changes: 81 additions & 0 deletions configs/exps/alvaro/faenet-top-config.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,81 @@
job:
mem: 32GB
cpus: 4
gres: gpu:rtx8000:1
partition: long
time: 15:00:00

default:
wandb_name: alvaro-carbonero-math
wandb_project: ocp-alvaro
wandb_tags: "best-config"
test_ri: True
mode: train
graph_rewiring: remove-tag-0
note: "top-run"
frame_averaging: 2D
fa_method: se3-random
cp_data_to_tmpdir: True
model:
edge_embed_type: all_rij
mp_type: updownscale_base
phys_embeds: True
tag_hidden_channels: 32
pg_hidden_channels: 96
energy_head: weighted-av-final-embeds
complex_mp: True
graph_norm: True
hidden_channels: 352
num_filters: 288
num_gaussians: 68
num_interactions: 5
second_layer_MLP: False
skip_co: concat
cutoff: 4.0
optim:
batch_size: 256
eval_batch_size: 256
lr_initial: 0.002
scheduler: LinearWarmupCosineAnnealingLR
max_epochs: 9
eval_every: 0.4

runs:
# - config: faenet-is2re-10k

# - config: faenet-is2re-10k
# is_disconnected: True

- config: indfaenet-is2re-all
note: so that cat get old dimensions
model:
hidden_channels: 352
num_gaussians: 99
num_filters: 448

- config: indfaenet-is2re-all
note: dimensions of both smaller
model:
hidden_channels: 176
num_gaussians: 50
num_filters: 224

- config: indfaenet-is2re-all
note: so that ads get old dimensions
model:
hidden_channels: 704
num_gaussians: 200
num_filters: 896

- config: indfaenet-is2re-all
note: so that their average is old dimension
model:
hidden_channels: 468
num_gaussians: 132
num_filters: 596

# - config: indfaenet-is2re-10k

# - config: afaenet-is2re-all
# model:
# afaenet_gat_mode: v1
46 changes: 46 additions & 0 deletions configs/exps/alvaro/faenet-training.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,46 @@
# MODIFY THIS ONE FOR RUNS

job:
mem: 32GB
cpus: 4
gres: gpu:rtx8000:1
partition: long
time: 15:00:00

default:
wandb_name: alvaro-carbonero-math
wandb_project: ocp-alvaro
test_ri: True
mode: train
graph_rewiring: remove-tag-0
cp_data_to_tmpdir: true
wandb-tags: 'best-config-??' # Insert what model you're running if running one by one.
frame_averaging: 2D
fa_frames: se3-random
model:
mp_type: updownscale
phys_embeds: True
tag_hidden_channels: 32
pg_hidden_channels: 64
energy_head: weighted-av-final-embeds
complex_mp: False
graph_norm: True
hidden_channels: 352
num_filters: 448
num_gaussians: 99
num_interactions: 6
second_layer_MLP: True
skip_co: concat
edge_embed_type: all_rij
optim:
lr_initial: 0.0005
scheduler: LinearWarmupCosineAnnealingLR
max_epochs: 20
eval_every: 0.4
batch_size: 256
eval_batch_size: 256

runs:
- config: afaenet-is2re-all
model:
afaenet_gat_mode: v1
35 changes: 35 additions & 0 deletions configs/exps/alvaro/gemnet-config.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,35 @@
job:
mem: 32GB
cpus: 4
gres: gpu:rtx8000:1
partition: long
time: 18:00:00

default:
wandb_name: alvaro-carbonero-math
wandb_project: ocp-alvaro
graph_rewiring: remove-tag-0
model:
tag_hidden_channels: 32
pg_hidden_channels: 32
phys_embeds: True
otf_graph: False
max_num_neighbors: 40
hidden_channels: 142
graph_rewiring: remove-0-tag
optim:
batch_size: 32
eval_batch_size: 32
max_epochs: 30

runs:
- config: gemnet_t-is2re-all

- config: gemnet_t-is2re-all
is_disconnected: True

#- config: depgemnet_t-is2re-all

#- config: indgemnet_t-is2re-all

#- config: agemnet_t-is2re-all
Loading