GEO-BON / bon-in-a-box-pipelines

BON in a Box 2.0 - Sampling optimisation and indicator pipelines
MIT License
11 stars 7 forks source link

Some SDM tasks from Taïga #137

Open frousseu opened 10 months ago

frousseu commented 10 months ago
glaroc commented 10 months ago

We should also make sure fit statistics like AUC are in the list of outputs

tpoisot commented 10 months ago

I was going to bring this up at the next meeting in 2024, but we should indeed make sure that there's a systematic validation of the SDMs with a series of reliable measures. There's been recent literature in ML showing that AUC (ROC-AUC anyways) can be high even for "bad" models.

jmlord commented 10 months ago

The enhancements listed above will obviously take time, and most probably happen at different moments. I think this issue should be exploded in a few other issues so we better track them.

jmlord commented 9 months ago

@frousseu "build decision tree to determine if a species can be modeled with any method (enough data, enough coverage, etc.)"

Should that be part of the tree? The tree could have exit points that are possible scripts, and exit points that are problems that explain why no model matched.