miller VS DataProfiler

jq and miller[1] are essential parts of my toolbelt, right up there with awk and vim.

[1]: https://github.com/johnkerl/miller

> This works really well if your problem can be solved in one or two liners.

My personal comfort threshold is around the 100-line mark. It's even possible to write maintainable shell scripts up to 500 lines, but it mostly depends on the problem you're trying to solve, and the discipline of the programmer to follow best practices (use sane defaults, ShellCheck, etc.).

> It go bad very quickly when, say, you have two CSV files and want to join them the sql-way.

In that case we're talking about structured data, and, yeah, Perl or Python would be easier to work with. That said, depending on the complexity of the CSV, you can still go a long way with plain Bash with IFS/read(1) or tr(1) to split CSV columns. This wouldn't be very robust, but there are tools that handle CSV specifically[1], which can be composed in a shell script just fine.

So it's always a balancing act of being productive quickly with a shell script, or reaching out for a programming language once the tools aren't a good fit, or maintenance becomes an issue.

[1]: https://miller.readthedocs.io/

where mlr is from https://github.com/johnkerl/miller

if this class of problems (i.e., csv/tsv data) is your main target you may find miller (https://github.com/johnkerl/miller) much more useful in the long run

That said, you may be interested in Miller (https://github.com/johnkerl/miller) which provides similar capabilities for CSV, JSON, and XML files. It doesn't use a SQL grammar, but that's just the proverbial lipstick on the thing. I'm not the author, but I have used it and I see some parallels in use cases at the very least.

worth giving Miller a shot

CSV utilities (still haven't pick a favorite one...): https://github.com/harelba/q https://github.com/BurntSushi/xsv https://github.com/wireservice/csvkit https://github.com/johnkerl/miller

It's been possible to skip tokenization for a long time, my team and I did it here - https://github.com/capitalone/DataProfiler

For what it's worth, we actually were working with LSTMs with nearly a billion params back in 2016-2017 area. Transformers made it far more effective to train and execute, but ultimately LSTMs are able to achieve similar results, though slow & require more training data.

Great call out -- would you be willing to write up an issue for that on the repo? Thank you! https://github.com/capitalone/DataProfiler/issues/new/choose