Often the input maps are in float32, in order to save memory is a good idea to keep them in memory also at float32.
However, as soon as we want to evaluate the model and then bandpass integrate, we want to make sure they are executed in double precision so we avoid accumulating rounding errors.
Often the input maps are in float32, in order to save memory is a good idea to keep them in memory also at float32.
However, as soon as we want to evaluate the model and then bandpass integrate, we want to make sure they are executed in double precision so we avoid accumulating rounding errors.