First of all, thanks for this amazing tool! It's incredibly helpful and makes my old data processing scripts look like a shameful mess 😆

Problem

I have a table of model fits, similar to the following:

mlr -t --from models.tsv head -n 4

I'd like to do some processing with miller, like this:

mlr -t --from models.tsv sort -n score then filter '$residual < 4' then uniq -f hash

The idea is to select best models by likelihood (score), apply threshold on residual and then deduplicate models by hash, as models with the same hash have the same topology, so I can just keep the fit with the best likelihood score. The problem is that miller will output just the column used by uniq -c, i.e. hash.

Proposal

Add a new parameter to uniq (and count-distinct), e.g. -k, which will keep the rest of the columns after deduplication on some field. the param could even accept comma-separated list of columns to keep, keeping all by default.

Notes

Note that I can achieve the same by piping to awk:

mlr -t --from models.tsv sort -n score then filter '$residual < 4' | awk '!a[$3]++' | mlr --t2m head -n 4