Skip to content

Inference model and adopted (pre-trained) geo-encoders#72

Merged
gabrieletijunaityte merged 81 commits into
developfrom
aligment_experiments
Apr 15, 2026
Merged

Inference model and adopted (pre-trained) geo-encoders#72
gabrieletijunaityte merged 81 commits into
developfrom
aligment_experiments

Conversation

@gabrieletijunaityte
Copy link
Copy Markdown
Contributor

@gabrieletijunaityte gabrieletijunaityte commented Apr 7, 2026 •
edited
Loading

What does this PR do?

Changes:

  • Checkpoints now save model hyperparameters, to allow model intialisation without the need for a configuration file.
  • Tessera tile requests fixed. Downloads and merges tiles in npy and tif, fills missing values with 0s.
  • Add missing forward pass through extra-geo-projector

Introduces:

  • Inference model which:
    • can be initialised given alignment and predictive model checkpoints (with some sanity check, e.g., if geo_encoders share architecture).
    • can be initialised from a merged checkpoint.
  • Adopt encoder function which allows to initialise pre-trained geo-encoder based on a checkpoint.
  • Identity encoder, which just passes through the input
  • Elbow-k calculation for validation concept retrieval tasks.

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?

gabrieletijunaityte and others added 30 commits March 25, 2026 21:43
@gabrieletijunaityte
Log dynamic k per concept
d993e87
@gabrieletijunaityte
valuation configs
3122086
@gabrieletijunaityte
alignment poster inspection code
a9b443e
@gabrieletijunaityte
fix self.self
9d32277
@gabrieletijunaityte
Remove redundant val_val in logging the metric names
d2b4398
@vdplasthijs
Index dynamic top k -- not tested yet
beccf79
@vdplasthijs
Elbow method for theta k -- not tested yet
23e8b28
@vdplasthijs
update theta_k with calculated
ce5232c
@gabrieletijunaityte
Add missing pass through the extra projector
150f4f6
@gabrieletijunaityte
Add new type of error for missing file specification
83ef95d
@gabrieletijunaityte
Add missing pass through the extra projector
de7da47
@gabrieletijunaityte
fix if self.extra_projector missing
8c4b2ab
@gabrieletijunaityte
Formating hook changes
047097c
@gabrieletijunaityte
Fix hugging face dir
fc832d4
@gabrieletijunaityte
Prediction head initialisation print statements
10030a7
@gabrieletijunaityte
Add full freezer
d4bbc9b
@gabrieletijunaityte
Add configuration saving into ckpg
6c2e38d 
This is needed for the inference model re-configuration
@gabrieletijunaityte
Introduce inference model
561411c
@gabrieletijunaityte
Add configuration saving into ckpg
4ce2c21
@gabrieletijunaityte
tessera fixes
1470953
@gabrieletijunaityte
Merge branch 'aligment_experiments' of github.com:WUR-AI/aether into …
a5e3191 
…aligment_experiments

# Conflicts:
#	src/models/text_alignment_model.py
@vdplasthijs @gabrieletijunaityte
minor gee update
a4a94c5 
in case outside of radius_max, return radius_max
@vdplasthijs @gabrieletijunaityte
concept captions
51aed7d 
manually picked relevant captions for biodiv UC, inspecting histograms to look for sensible thresholds.
@robknapen @gabrieletijunaityte
Makes DBScan clustering more efficient and much faster.
4f0ed1d
@robknapen @gabrieletijunaityte
Crop Yield use case: spatial splitting
6911022
@robknapen @gabrieletijunaityte
Crop Yield use case: configs for various experiments
22c9cbb
@robknapen @gabrieletijunaityte
Adds RRMSE loss function for crop yield error comparison
f318689
@gabrieletijunaityte
Crop Yield use case: Adds Fourier harmonics as engineered location fe…
da121c4 
…atures.

# Conflicts:
#	tests/test_yield_africa.py
@gabrieletijunaityte
Fix encoder tests
9a62ceb
@gabrieletijunaityte
fix tests
ec9338b
vdplasthijs and others added 18 commits April 7, 2026 13:35
@vdplasthijs @gabrieletijunaityte
update theta_k with calculated
bf51184
@gabrieletijunaityte
Merge branch 'aligment_experiments' of github.com:WUR-AI/aether into …
09c7cf4 
…aligment_experiments
@gabrieletijunaityte
Cleaner print
67b0221
@gabrieletijunaityte
Move logging
d4fce8b
@gabrieletijunaityte
Fix trainable module typo, add to freeze all model upon testing
0e9c285
@gabrieletijunaityte
Introduce adopted encoder
331aba2 
Geo encoder from a checkpoint
@gabrieletijunaityte
optimise iterating through dataset one single time
048cef0
@gabrieletijunaityte
self.self fix
30bdb82
@gabrieletijunaityte
introduce identity encoder issue #65
3cef11c
@gabrieletijunaityte
Add setup for text encoder
c7fc5f9
@gabrieletijunaityte
Extract _setup method, add setup_flag, add update_configs method
3e78fae
@gabrieletijunaityte
Update configs from the checkpoint
bf7394f
@gabrieletijunaityte
Extract _setup method, add setup_flag
a504c6d
@gabrieletijunaityte
Clean up models and add setup flags, configurations into the model ch…
16f495d 
…eckpointing
@gabrieletijunaityte
Add another option for log dir
3542d14
@gabrieletijunaityte
Merge branch 'develop' into aligment_experiments
15ccdf8
@gabrieletijunaityte
Fix tests about missmatch in dim
f94a3bb
@gabrieletijunaityte
Updates to tessera downloading
15229a6 
(for larger areas, saves tiles with 0s for missing data)
Includes data checks
@gabrieletijunaityte gabrieletijunaityte marked this pull request as ready for review April 10, 2026 12:38
vdplasthijs
vdplasthijs approved these changes Apr 14, 2026
View reviewed changes
Copy link
Copy Markdown
Collaborator

@vdplasthijs vdplasthijs left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Great stuff, looks very neat! Just a few minor things, I'll approve so have a look and merge when you're happy.

Comment thread src/models/components/geo_encoders/encoder_wrapper.py Outdated
Comment thread src/models/components/metrics/metrics_wrapper.py
Comment thread src/models/base_model.py Outdated
Comment thread src/models/inference_model.py Outdated
This was referenced Apr 14, 2026
Closed
Open
Open
gabrieletijunaityte
gabrieletijunaityte commented Apr 15, 2026
View reviewed changes
Comment thread src/data/heat_guatemala_dataset.py
@gabrieletijunaityte gabrieletijunaityte merged commit a5d7ace into develop Apr 15, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@BachirNILU BachirNILU BachirNILU left review comments

@vdplasthijs vdplasthijs vdplasthijs 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.

4 participants

@gabrieletijunaityte @vdplasthijs @BachirNILU @robknapen