The mstransform task is able to work with MSs that contain spectral windows with different polarization shapes. Split cannot handle such MSs, but this was overcome in mstransform.
A data column specifies which column will be used for the transformations. It is possible to make real a virtual MODEL column, by setting the parameter datacolumn to any of the values 'model', 'all' or 'data,model,corrected' and then setting the sub-parameter realmodelcol to True. This will copy the virtual column to a physical MODEL column in the MS main table. The virtual model column will be deleted afterwards.
The user can control whether to create WEIGHT_SPECTRUM/SIGMA_SPECTRUM columns in the output MS or not. If the input MS does not have a WEIGHT_SPECTRUM/SIGMA_SPECTRUM column, by setting the parameter usewtspectrum to True, the task will create a WEIGHT_SPECTRUM from the values in the WEIGHT column, and initialize SIGMA_SPECTRUM to 1/sqrt(WEIGHT_SPECTRUM). By default, usewtspectrum is set to False.
The resulting WEIGHT_SPECTRUM produced by mstransform is in the statistical sense correct for the simple cases of channel average and time average, but not for the general re-gridding case, in which the error propagation formulas applicable for WEIGHT_SPECTRUM are yet to be defined. Currently, as in cvel and in the imager, WEIGHT_SPECTRUM is transformed in the same way as the other data columns, notice that this is not formally correct from the statistical point of view, but is a good approximation at this stage.
The parameter keepflags from split is also available in mstransform. By default, mstransform keeps the flagged data in the output, depending on the data selection parameters. Regardless of this parameter, flagged data is never used in channel averaging. When keepflags is set to False, only partially flagged rows will be used in time averaging. This behavior is different from split.
mstransform('uid.ms', outputvis='myout.ms', datacolumn='data', spw='0:21~120,3,5')
will select 100 channels from spw 0 and all of spw 3 and 5