When using BDA in the init worker the number of output rows per chunk depends on the highest frequency in the chunk. This means that the grid worker won't be able to concatenate by chan since output chunks will have a different number of rows. The frequency resolution (i.e. the number of imaging bands) is therefore frozen in by the --channels-per-image parameter and no further coarsening in frequency is currently possible. The solution would be to combine the bands in image space but this is not currently implemented. Concatenating by row should still be possible so --integrations-per-image can be used if more and/or smaller chunks are required during initialisation
When using BDA in the
init
worker the number of output rows per chunk depends on the highest frequency in the chunk. This means that thegrid
worker won't be able to concatenate by chan since output chunks will have a different number of rows. The frequency resolution (i.e. the number of imaging bands) is therefore frozen in by the--channels-per-image
parameter and no further coarsening in frequency is currently possible. The solution would be to combine the bands in image space but this is not currently implemented. Concatenating by row should still be possible so--integrations-per-image
can be used if more and/or smaller chunks are required during initialisation