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.

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

Uncertainty #59

Open
wants to merge 30 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
30 commits
Select commit Hold shift + click to select a range
0dea35c
fix cluster issue gpus-per-task
AlexDuvalinho Apr 18, 2024
904fea1
new yaml configs
AlexDuvalinho Apr 18, 2024
7c48139
update path trained gnn model
AlexDuvalinho Apr 18, 2024
72ae772
fa_frames => fa_method
AlexDuvalinho Apr 18, 2024
040b475
skip_co = concat is not possible
AlexDuvalinho Apr 18, 2024
83659c6
Merge only relevant changed from disconnected_gnn branch, to run depf…
AlexDuvalinho Apr 19, 2024
442ca59
remove edge_embed_type
AlexDuvalinho Apr 19, 2024
0d70e8e
create deup-depfaenet, add dropout_lin, modif class names
AlexDuvalinho Apr 19, 2024
fd9d1d1
add q
AlexDuvalinho Apr 19, 2024
2ab5c33
fix forward of output block depfaenet
AlexDuvalinho Apr 19, 2024
9f18bfd
new model checkpoints to create deup-dataset
AlexDuvalinho Apr 23, 2024
e0fb6f7
argparse deup_dataset + comments
AlexDuvalinho Apr 23, 2024
5b9c76f
fix chkpt_path + edge case error + new configs
AlexDuvalinho Apr 24, 2024
8599de7
adapt configs for v0 deup-faenet training on deup-dataset
AlexDuvalinho Apr 24, 2024
58b9927
fix module load
ginihumer Apr 25, 2024
5a5524c
return hidden state in wrapper
ginihumer Apr 25, 2024
6594960
`scatter` `q` in `energy_forward`
vict0rsch Apr 25, 2024
5ce2f3f
fix configs for depfaenet/faenet fine-tuning
AlexDuvalinho Apr 25, 2024
606fcd0
quickfixes
ginihumer Apr 26, 2024
fcf2650
update configs deup-depfaenet
AlexDuvalinho Apr 26, 2024
4d73707
test use deup-dataset in an active learning framework
AlexDuvalinho Apr 26, 2024
b426cce
Merge branch 'uncertainty-depfaenet' of https://github.com/RolnickLab…
AlexDuvalinho Apr 26, 2024
2bc9a1a
Merge pull request #61 from RolnickLab/ch/develop
AlexDuvalinho Apr 26, 2024
5061d09
Merge pull request #60 from RolnickLab/uncertainty-depfaenet
AlexDuvalinho Apr 26, 2024
175567e
deupdepfaenet configs
AlexDuvalinho May 17, 2024
03f3038
fix issues with q + enforce graph-level deup-dataset
AlexDuvalinho May 17, 2024
ae7b175
random instead of randon in yaml
AlexDuvalinho May 20, 2024
7c2714c
random, not stochastic
AlexDuvalinho May 20, 2024
b006540
signnet analysis (workshop submission)
AlexDuvalinho May 21, 2024
cc50335
denormalise predictions
AlexDuvalinho Oct 31, 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
141 changes: 141 additions & 0 deletions configs/exps/catalyst/gflownet.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,141 @@
job:
mem: 32GB
cpus: 4
gres: gpu:1
partition: long
time: 15:00:00

default:
wandb_project: ocp-deup # ocp-alvaro
wandb_tags: gflownet-model, depfaenet
test_ri: True
mode: train
# graph_rewiring: remove-tag-0
graph_rewiring: ""
frame_averaging: 2D
fa_method: se3-random
cp_data_to_tmpdir: True
is_disconnected: true
model:
mp_type: updownscale_base
phys_embeds: True
tag_hidden_channels: 0
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-all
# note: baseline faenet

# - config: depfaenet-is2re-all
# note: depfaenet baseline

# - config: depfaenet-is2re-all
# note: depfaenet per-adsorbate
# adsorbates: {'*O', '*OH', '*OH2', '*H'}

# - config: depfaenet-is2re-all
# note: depfaenet per-adsorbate long string
# adsorbates: '*O, *OH, *OH2, *H'

# - config: depfaenet-is2re-all
# note: depfaenet per-adsorbate string of a list
# adsorbates: "*O, *OH, *OH2, *H"

# - config: depfaenet-is2re-all
# note: Trained on selected adsorbate more epochs
# adsorbates: "*O, *OH, *OH2, *H"
# optim:
# max_epochs: 10

# - config: depfaenet-is2re-all
# note: depfaenet full data

# - config: depfaenet-is2re-all
# note: To be used for continue from dir

# - config: depfaenet-is2re-all
# note: Fine-tune on per-ads-dataset 4 epoch
# continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4023244
# adsorbates: "*O, *OH, *OH2, *H"
# optim:
# max_epochs: 4
# lr_initial: 0.00015

# - config: depfaenet-is2re-all
# note: Fine-tune on per-ads-dataset 10 epoch
# continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4023244
# adsorbates: "*O, *OH, *OH2, *H"
# optim:
# max_epochs: 10
# lr_initial: 0.00015

# - config: depfaenet-is2re-all
# note: Fine-tune on per-ads-dataset 10 epoch
# continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4023244
# adsorbates: "*O, *OH, *OH2, *H"
# optim:
# max_epochs: 20
# lr_initial: 0.0001

# - config: depfaenet-is2re-all
# note: Fine-tune on per-ads-dataset 20 epoch
# continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4023244
# adsorbates: "*O, *OH, *OH2, *H"
# optim:
# max_epochs: 20
# lr_initial: 0.00015

- config: depfaenet-is2re-all
note: Depfaenet per-ads-dataset
continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4023244
adsorbates: "*O, *OH, *OH2, *H"
optim:
max_epochs: 12
lr_initial: 0.0002

- config: depfaenet-is2re-all
note: Depfaenet per-ads-dataset
continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4023244
adsorbates: "*O, *OH, *OH2, *H"
optim:
max_epochs: 10
lr_initial: 0.0001

# - config: depfaenet-is2re-all
# note: Fine-tune on per-ads-dataset starting from fine-tuned model
# continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4071859
# adsorbates: "*O, *OH, *OH2, *H"
# optim:
# max_epochs: 10
# lr_initial: 0.0001

# - config: depfaenet-is2re-all
# note: Trained on selected adsorbate
# adsorbates: "*O, *OH, *OH2, *H"
# optim:
# max_epochs: 25
# lr_initial: 0.0001

# - config: depfaenet-is2re-all
# note: Trained on selected adsorbate
# adsorbates: "*O, *OH, *OH2, *H"
# optim:
# max_epochs: 25
75 changes: 75 additions & 0 deletions configs/exps/catalyst/reproduce-configs.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,75 @@
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: "reproduce-best-config"
test_ri: True
mode: train
graph_rewiring: remove-tag-0
note: "repoduce-top-run"
frame_averaging: 2D
fa_method: se3-random
cp_data_to_tmpdir: True
is_disconnected: 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-all
note: baseline faenet

- config: indfaenet-is2re-all
note: baseline with top configs

- config: indfaenet-is2re-all
note: baseline with runs' configs
model:
tag_hidden_channels: 32
pg_hidden_channels: 96
energy_head: weighted-av-final-embeds
complex_mp: True
graph_norm: True
hidden_channels: 528
num_filters: 672
num_gaussians: 148
num_interactions: 5
second_layer_MLP: False
skip_co: concat

- config: depfaenet-is2re-all
note: baseline with top configs

- config: indfaenet-is2re-all
note: so that ads get old dimensions
model:
hidden_channels: 704
num_gaussians: 200
num_filters: 896
28 changes: 28 additions & 0 deletions configs/exps/deup/datasets/data-with-depfaenet.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
job:
mem: 32GB
cpus: 4
gres: gpu:1
partition: long

default:
config: depfaenet-is2re-all
wandb_project: ocp-deup
wandb_tags: depfaenet, MC-D,4621042
test_ri: True
mode: train
checkpoint: /network/scratch/a/alexandre.duval/ocp/runs/4621042/checkpoints/best_checkpoint.pt
restart_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4621042/
model:
dropout_lowest_layer: output
first_trainable_layer: dropout
dropout_lin: 0.3
cp_data_to_tmpdir: true
inference_time_loops: 1
deup_dataset:
create: after # "before" -> created before training (for deup) "after" -> created after training (for is2re) "" - not created
dataset_strs: ["train", "val_id", "val_ood_cat", "val_ood_ads"]
n_samples: 7

runs:
- optim:
max_epochs: 12
28 changes: 28 additions & 0 deletions configs/exps/deup/datasets/mc-faenet.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
job:
mem: 32GB
cpus: 4
gres: gpu:1
partition: long

default:
config: faenet-is2re-all
wandb_project: ocp-deup
wandb_tags: base-model, MC-D, 4616500
test_ri: True
mode: train
checkpoint: /network/scratch/a/alexandre.duval/ocp/runs/4616500/checkpoints/best_checkpoint.pt
restart_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4616500/
model:
dropout_lowest_layer: output
first_trainable_layer: dropout
dropout_lin: 0.7
cp_data_to_tmpdir: true
inference_time_loops: 1
deup_dataset:
create: after # "before" -> created before training (for deup) "after" -> created after training (for is2re) "" - not created
dataset_strs: ["train", "val_id", "val_ood_cat", "val_ood_ads"]
n_samples: 7

runs:
- optim:
max_epochs: 12
67 changes: 67 additions & 0 deletions configs/exps/deup/gnn/depfaenet-finetune.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,67 @@
job:
mem: 32GB
cpus: 4
gres: gpu:1
partition: long
time: 15:00:00

default:
wandb_project: ocp-deup
wandb_tags: depfaenet, no-concat, with-tag0, dropout
test_ri: True
mode: train
graph_rewiring: ""
frame_averaging: 2D
fa_method: se3-random
is_disconnected: true

runs:

- config: depfaenet-is2re-all
note: Depfaenet per-ads-dataset
continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4647488 #4647466 #4023244
adsorbates: "*O, *OH, *OH2, *H"
optim:
max_epochs: 10
lr_initial: 0.0002

- config: depfaenet-is2re-all
note: Depfaenet per-ads-dataset
continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4647488 #4647466 # 4023244
adsorbates: "*O, *OH, *OH2, *H"
optim:
max_epochs: 12
lr_initial: 0.0001

- config: depfaenet-is2re-all
note: Depfaenet per-ads-dataset
continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4647466 # 4023244
adsorbates: "*O, *OH, *OH2, *H"
graph_rewiring: ""
frame_averaging: 2D
fa_method: se3-random
cp_data_to_tmpdir: True
is_disconnected: true
model:
mp_type: updownscale_base
phys_embeds: True
tag_hidden_channels: 0
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: False
cutoff: 4.0
dropout_lin: 0.3
optim:
batch_size: 256
eval_batch_size: 256
lr_initial: 0.002
scheduler: LinearWarmupCosineAnnealingLR
eval_every: 0.4
max_epochs: 12
47 changes: 47 additions & 0 deletions configs/exps/deup/gnn/faenet-finetune.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
job:
mem: 32GB
cpus: 4
gres: gpu:1
partition: long
time: 18:00:00

default:
test_ri: True
mode: train
wandb_tags: faenet, no-concat, with-tag0, dropout, fine-tuned
wandb_project: ocp-deup
graph_rewiring: ""
optim:
batch_size: 232
eval_batch_size: 232
cp_data_to_tmpdir: True

runs:
- config: faenet-is2re-all
note: "fine-tuned faenet"
continue_from_dir: /network/scratch/a/alexandre.duval/ocp/runs/4647489
adsorbates: "*O, *OH, *OH2, *H"
frame_averaging: 2D
fa_method: se3-random
model:
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: 384
num_filters: 480
num_gaussians: 104
num_interactions: 5
second_layer_MLP: False
skip_co: False
cutoff: 6.0
dropout_lin: 0.3
dropout_lowest_layer: output
optim:
lr_initial: 0.002
scheduler: LinearWarmupCosineAnnealingLR
max_epochs: 14
eval_every: 0.25
Loading