Hi @mbhall88,
thanks for developing this tool, this approach is indeed very fast!
However when trying out on bacterial genome assembly data (Nanopore and Illumina) I was puzzled that nohuman was throwing the baby out with the bathwater. Therefore I investigated with the manual approach described in https://github.com/mbhall88/classification_benchmark with kraken2 and got:

277081 sequences (1116.92 Mbp) processed in 16.112s (1031.8 Kseq/m, 4159.33 Mbp/m).
  2814 sequences classified (1.02%).
  274267 sequences unclassified (98.98%)   

But the output file from nohuman contained 2814 sequences, which I trace the typo to be:

I submitted a PR to fix that unclassified reads are wanted, but I'm not familiar with Rust, so I could not recompile/test properly, let me know if I can try out something.

Best,
Charlie