High memory usage when frequency averaging MWA correlator fits files

[ldress]

I have read in a ~13 GB set of MWA correlator fits files (from one 2 minute observation at 40kHz) whose frequency resolution I'd like to reduce. However, when I run frequency_average(n_chan_to_avg=2) on this data, the process consumes more than my available system memory (96 GB) and fails. Averaging over 4 channels succeeds but still uses almost all available memory. As a workaround, I can manually process a few files at a time and concatenate the results, but it might be nice to have a built-in process that doesn’t load as much data into memory at once.