earthpulse / eotdl

Earth Observation Training Datasets
https://eotdl.com
MIT License
17 stars 6 forks source link

Q1+ models STAC extension #162

Closed fmariv closed 2 weeks ago

fmariv commented 5 months ago

Create a new STAC extension for Q1+ models ingested in EOTDL.

ML Model spec Name Description STAC Support Actions
Id Unique id of the model Core -
dataset Link to datasets used for training (preferably on EOTDL). No ML Model extension
Version Version of the model No ML Model extension
Hparams List of hyperparmeters used for training the model (including feature extraction pipeline if necessary) No ML Model extension
Type Neural network, SVM, RF, … No ML Model extension
Features Link to the feature store used to train the model (if required) No ML Feature extension
DOI   Scientific extension -
Citation   Scientific extension -
Inputs Links to sample inputs for trying the model, pre-processing steps (pipeline) and other aspects such as normalization statistics, inputs names, etc. No ML Model extension
Outputs List of output names, type, tasks, post-processing steps (pipeline) and other relevant aspects. No ML Model extension
juansensio commented 2 weeks ago

First example here https://github.com/earthpulse/eotdl/blob/main/tutorials/notebooks/05_q2_model.ipynb

Full support for ingesting/staging in CLI & API