Reduce function if concatenated intermediate results doesn't fit in memory

I'm trying to decode a huge dataset of rougly 500k*10k size. What I'm doing is actually load the file into a datastore, replace the tall table variable names with a meaningful name, and decode the content of all columns with a meaningful sentences in my dictionary.

I have some wiered issues that didn't let my program go through using a tall table. So I'm now trying to use mapreduce.

I find an example here

https://www.mathworks.com/help/matlab/import_export/simple-data-subsetting-using-mapreduce.html

function subsettingReducer(~, intermValList, outKVStore)
  % get all intermediate results from the list
  outVal = {};

  while hasnext(intermValList)
    outVal = [outVal; getnext(intermValList)];
  end
  % Note that this approach assumes the concatenated intermediate values (the
  % subset of the whole data) fit in memory.
    
  add(outKVStore, 'Null', outVal);
end

My question is, in this example, if the contanenated intermediate results doesn't fit in memory, is there a better workaround for this?

Thanks!

[Peng Li]

Or actually I can write the data store to local in the reduce function?

[Peng Li]

anybody who can shed me some light please?