You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently we import a set of TIFs into HDF5 format then extract ALL those covariates (with an optional halfwidth parameter) to create a tf records containing the train/test and query X data.
In order to exclude a particular TIF, or to try different combinations of TIFs, we need to import them to separate HDF5 files, resulting in a lot of data duplication.
The proposal is to add an option to the extract commands to list the covariates by name which we want to extract to a new tf record. The default would be to extract all covarates. It would likely also require a command to query the names of all covariates within an existing HDF5 file.
It could potentially work as --include/--exclude flags to allow for filtering out covariates.
The text was updated successfully, but these errors were encountered:
Ok, so you can ignore any covariates within the model config. So it's just a trade-off between extracting a new tfrecord vs the overhead of reading in unused covariates during training. While I think this would still be useful, I guess its not that important as there is a way to achieve it already.
Currently we import a set of TIFs into HDF5 format then extract ALL those covariates (with an optional halfwidth parameter) to create a tf records containing the train/test and query
X
data.In order to exclude a particular TIF, or to try different combinations of TIFs, we need to import them to separate HDF5 files, resulting in a lot of data duplication.
The proposal is to add an option to the
extract
commands to list the covariates by name which we want to extract to a new tf record. The default would be to extract all covarates. It would likely also require a command to query the names of all covariates within an existing HDF5 file.It could potentially work as
--include
/--exclude
flags to allow for filtering out covariates.The text was updated successfully, but these errors were encountered: