v0.3.2

Changes

• Improved training and alignment validation @vdplasthijs (#114)
• Load unlabelled data for alignment training @vdplasthijs (#112)
• Prediction model training changes @gabrieletijunaityte (#111)
• Fix dynamic top-k index scores @vdplasthijs (#98)

v0.3.1

What's Changed

• Change dependabot to ignore patches, work weekly and onto develop by @vdplasthijs in #91
• Change dependabot by @vdplasthijs in #92
• Update yield Africa configs and add earthaccess dependency by @robknapen in #83
• Add dynamic gate fusion strategy to EncoderWrapper by @robknapen in #93
• YieldAfrica dataset: drop records with missing tessera tiles by @robknapen in #95
• Tessera preprocessing: robustness and multiprocessing by @robknapen in #94
• Add ordinal classification of augmentation features to merge pipeline by @robknapen in #96
• Add YieldAfricaCaptionBuilder with Tessera+caption datamodule config by @robknapen in #97
• Small fixes by @gabrieletijunaityte in #99
• Pin PL version because of compromise by @vdplasthijs in #101
• Display mode for R2 and MAPE metrics by @robknapen in #102
• Cleanup of hydra configurations for crop yield experiments by @robknapen in #104
• Updated crop yield concept captions. Examples only for now. by @robknapen in #103
• Dtype parameterization by @gabrieletijunaityte in #106
• Fix cnn and add SatCLIP by @gabrieletijunaityte in #107
• Add batch size finder by @gabrieletijunaityte in #108
• Sync main and develop, many minor updates by @vdplasthijs in #110

Full Changelog: v0.3.0...v0.3.1

v0.3.0

Changes

• Inference model, encoder wrapper and more @vdplasthijs (#90)
• Add Kenya (KEN) data and experiment configs for hydra @robknapen (#82)
• Gated fusion, feature normalisation, and MAPE metric - to help CY use case @robknapen (#81)
• Data layer: normalisation stats, custom CSV name, LOCO country features @robknapen (#80)
• Add data augmentation pipeline scripts for CY use case @robknapen (#79)
• Dependabot package updates @gabrieletijunaityte (#84)
• Inference model and adopted (pre-trained) geo-encoders @gabrieletijunaityte (#72)

v0.2.0

What's Changed

• Introduce evaluation on concept captions by @gabrieletijunaityte in #57
• Setup method of the models by @gabrieletijunaityte in #59
• Crop Yield Africa use case - initial dataset and example tabular regression by @robknapen in #60
• Feature/spatial splits by @robknapen in #61
• Added concept captions for biodiv UC by @vdplasthijs in #62
• Geo-encoder wrapper by @gabrieletijunaityte in #63
• Introduce multi-encoder, crop yield UC and concept captions by @gabrieletijunaityte in #67

New Contributors

• @robknapen made their first contribution in #60

Full Changelog: v0.1.2...v0.2.0

More use cases, tabular data, LLM text encoder, location caption generation.

In this version, we release a number of changes and improvements.

Introduce additional use case implementations for SatBird-USA-summer and urban heat for Guatemala.
Add tabular data as input geo features.
Decouple training and validation metrics from the model class.
Introduce more auxiliary variables in the butterfly UC dataset.
Automate location caption generation for Butterfly UC.
Add an LLM as a text encoder.

What's Changed

• Updated captions for butterfly data by @vdplasthijs in #41
• Generate caption templates by @vdplasthijs in #46
• Satbird implementation by @gabrieletijunaityte in #45
• Urban heat islands by @BachirNILU in #47
• Tabular data for predictive model by @BachirNILU in #47
• LLM as text encoder implementation from LLM2CLIP by @gabrieletijunaityte in #42
• Metrics configurable through yaml by @gabrieletijunaityte in #50
• Minor test updates by @vdplasthijs in #53
• Change auxiliary data detangling by @gabrieletijunaityte in #54

New Contributors

• @BachirNILU made their first contribution in #47

Full Changelog: v0.1.1...v0.1.2

Pooch and paths fixed for first time setting up repo

Changes

• Feature/fix pooch @gabrieletijunaityte (#38)
• Minor updates to fix test for first-time users @vdplasthijs (#37)
• typos @cn241 (#34)
• Average encoder (for FM ims) @vdplasthijs (#32)
• Update minor README @vdplasthijs (#31)

Dataset re-organisation, improved predictive models

Main changes:

• New dataset structure (documented in README.md) and dataset downloading is implemented through pooch (for ready-to-use open source datasets).
• Tests implemented.
• Supports for more EO modalities.
• Second dataset (Satbird) implemented.
• Prediction model improvements: more metrics, MLP head option implemented, L2 norm moved from loss function.

Alignment model and predictive model.

We created an abstracted torch-hydra framework, including both an EO-text alignment model and an EO predictive model. Also including instances of the model to perform the S2BMS (butterfly) task.