Skip to content

Improved training and alignment validation#114

Merged
gabrieletijunaityte merged 97 commits into
mainfrom
develop
Jun 1, 2026
Merged

Improved training and alignment validation#114
gabrieletijunaityte merged 97 commits into
mainfrom
develop

Conversation

@vdplasthijs

@vdplasthijs vdplasthijs commented Jun 1, 2026

Copy link
Copy Markdown
Collaborator

What does this PR do?

I think in short:

  • Fixed concept validation task metrics + support for unlabelled data.
  • Handling missing (tessera) data
  • Many minor improvements for training in model and loggers.

Before submitting

  • Did you make sure title is self-explanatory and the description concisely explains the PR?
  • Did you make sure your PR does only one thing, instead of bundling different changes together?
  • Did you list all the breaking changes introduced by this pull request?
  • Did you test your PR locally with pytest command?

vdplasthijs and others added 30 commits April 23, 2026 16:41
@vdplasthijs
Merge branch 'develop' into aligment_experiments
5954997
@vdplasthijs
Fix dynamic top k index scores
bb37d5d 
- Calculate theta_k for train only, and reuse for val and test.
- Catch edge cases max = min leading to zero div
- Printing optional, off by default
- Elbow method removes nans and skips leading zeros for tighter distribution.
- Update default concept caption v
@vdplasthijs
Merge branch 'develop' into aligment_experiments
8795be7
@gabrieletijunaityte
Remove missing tessera tiles from training split
49fb4ed
@gabrieletijunaityte
Remove missing tessera tiles from training split
d6be8fe
@gabrieletijunaityte
Add s2bms prediction configs
c09c2e3
@vdplasthijs
Fix caching
b9cf1ec
@gabrieletijunaityte
s2bms_experiment_1: aef 128
a3fa726
@gabrieletijunaityte
Remove missing tessera tiles from training split
b99e1d4
@gabrieletijunaityte
Remove missing tessera tiles from training split
4d01fc2
@gabrieletijunaityte
Add s2bms prediction configs
70134f2
@gabrieletijunaityte
s2bms_experiment_1: aef 128
da0f6a8
@gabrieletijunaityte
FIx data check for aef
0754317
@gabrieletijunaityte
For now deactivate tessera downloading
bbd57d2
@gabrieletijunaityte
Unified datasplit
4880073
@gabrieletijunaityte
Lower patience
99c2d56
@gabrieletijunaityte
LOg more frequently
4a4ed97
@gabrieletijunaityte
Allow "str" conversion in configurations
ae47cf9
@gabrieletijunaityte
Prediciton experiments: average encoder aef (1-2) and tessera (3-4) w…
945c5aa 
…ith lin or mlp pred-head
@gabrieletijunaityte
clean up
3c69908
@gabrieletijunaityte
Track wandb executed experiments
a15adcc
@gabrieletijunaityte
Add checkpoint dir configuration
1557577
@gabrieletijunaityte
Add checkpoint dir configuration
af4049a
@gabrieletijunaityte
Change where ckpt are saved and how they are logged to wandb and loca…
e913927 
…l csv log file
@gabrieletijunaityte
Longer training, larger lr
057fecd
@gabrieletijunaityte
garbage collection for oom
f715b40
@gabrieletijunaityte
Merge branch 'train/prediction' of github.com:WUR-AI/aether into trai…
71da776 
…n/prediction

# Conflicts:
#	configs/experiment/s2bms_experiment_1.yaml
#	src/data/base_dataset.py
@gabrieletijunaityte
restore object dict return
9ac6293
@gabrieletijunaityte
Remove missing tessera tiles from training split
00d6af9 
# Conflicts:
#	src/data/base_dataset.py
@gabrieletijunaityte
Remove missing tessera tiles from training split
3cefb8a
vdplasthijs and others added 27 commits May 21, 2026 17:02
@vdplasthijs
Checks if min/max corresponds to less than 50% of data
9f6b8e4
@vdplasthijs
set is_max if needed
51c33ae
@vdplasthijs
Allow for calcuting/saving is_max only
ada3d4c
@gabrieletijunaityte
Add splits for aef and tesera dataset union
bb9c3ca
@gabrieletijunaityte
Read aef from npy
454d42b
@gabrieletijunaityte
Log mse loss for the best model
8da67e4
@gabrieletijunaityte
Merge branch 'develop' of github.com:WUR-AI/aether into train/prediction
8a3390d
@vdplasthijs
store baseline accuracy
3cc5156
@gabrieletijunaityte
Add model config
b5fd5ca
@gabrieletijunaityte
Rename metrics config reference
d17db69
@vdplasthijs
Move concept caption threshold creation to datamodule
878c6f9
@vdplasthijs
Fix logging alignment model
b2035b5 
- Save mode
- Save average index
@vdplasthijs
Change concept caption files
b5a9f02
@gabrieletijunaityte
Merge pull request #98 from WUR-AI/aligment_experiments
4bd13de 
Fix dynamic top-k index scores
@gabrieletijunaityte
Merge branch 'develop' of github.com:WUR-AI/aether into train/prediction
5c43f58
@gabrieletijunaityte
Fix tessera downloading
7431913
@gabrieletijunaityte
Merge pull request #111 from WUR-AI/train/prediction
e494d14 
Prediction model training changes
@vdplasthijs
Download GEE data in parallel
42299c9
@vdplasthijs
Sample unlabelled aux data for UK s2bms
d54c435
@vdplasthijs
Add ability to load unlabelled data
ebb28cd
@vdplasthijs
adding unlabbelled s2bms data, bypassing hook
31c39a2
@vdplasthijs
Merge remote-tracking branch 'WUR-AI/develop' into aligment_experiments
c95faed
@vdplasthijs
Load unlabelled and original s2bms together
a58dcd7
@vdplasthijs
adding unlabbelled s2bms data splits, bypassing hook
1fa23a1
@vdplasthijs
adding unlabbelled splits, bypassing hook
5d6eda9
@vdplasthijs
Merge remote-tracking branch 'WUR-AI/aligment_experiments' into aligm…
ce4203d 
…ent_experiments
@gabrieletijunaityte
Merge pull request #112 from WUR-AI/aligment_experiments
e1e74d9 
Load unlabelled data for alignment training
@gabrieletijunaityte gabrieletijunaityte merged commit 45e6e90 into main Jun 1, 2026
11 of 12 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@gabrieletijunaityte gabrieletijunaityte gabrieletijunaityte approved these changes

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

@vdplasthijs @gabrieletijunaityte