Model validation - Githubissues

cct-datascience / setaria-predict

Using ED2 ensembles to train a spatially explicit model

0 stars 0 forks source link

Model validation #5

Closed Aariq closed 1 year ago

Aariq commented 1 year ago

[x] Learn how to validate RF models
[x] Fit another kind of regression as a comparison
[ ] Make Quarto report explaining choices

Aariq commented 1 year ago

Not 100% satisfied yet. RF performs better than glmnet, but still overpredicts in mixed forest and underpredicts in prairie. Not sure what the source of bias is.

Aariq commented 1 year ago

Fixed by using mean of log NPP instead of log of mean NPP as a response variable