Data management
In order to produce classifications using Sentinel-2 sensor, some data are shared between IOTA²'s run : *STACK.tif
and *10M.tif
by dates. In order to save processing time it has been decided to write them on disk.
However, in large scale classifications, disk space could be more precious than processing time.
That's why I propose to add a new parameter full_pipeline
to allow IOTA² generating features without any (or the less possible) writing on disk.
The workflow could follow the one below :
superimpose every available bands -> concatenate them (to produce a time series) -> ... -> features extraction