Right now redundant calibration in the file_calibration notebook is effectively set to run for a fixed 50 iteration, for up to 4 reruns with antennas removed due to high chi-squared. While this heuristic seems to be producing ok solutions at the moment (for the H6C data we've looked at), this need not be the case in the future and is something we should watch out for. In general the redcal solution should be determined by a convergence criterion, not a fixed number of iterations. With a proper convergence criterion in place, hitting the maximum iteration limit set by the OC_MAX_ITER parameter should then be considered a failure to find a solution, and the integration/channel flagged.
Right now redundant calibration in the file_calibration notebook is effectively set to run for a fixed 50 iteration, for up to 4 reruns with antennas removed due to high chi-squared. While this heuristic seems to be producing ok solutions at the moment (for the H6C data we've looked at), this need not be the case in the future and is something we should watch out for. In general the redcal solution should be determined by a convergence criterion, not a fixed number of iterations. With a proper convergence criterion in place, hitting the maximum iteration limit set by the OC_MAX_ITER parameter should then be considered a failure to find a solution, and the integration/channel flagged.