I’ve been trying to train this classifier using SILVA in the V3 V4 regions. I’m only using forward reads (whereas the original data included both forward and reverse paired end reads). I trimmed by data between 90-210 bps, therefore 120 length.
Wild guess: using the 80 % OTUs could be related to this issue. What happens if you run the same command but use the 90% OTUs? Note that we recommend using 99% OTUs for classification of real data, 80% likely would not give very sensitive classifications even if it could be used to train a classifier.
If that does not work, could you please post the full error output, found in this file: /var/folders/jk/4cm8m3mj58x1zh027dvn94jw0000gr/T/qiime2-q2cli-err-nurntpca.log
Or re-run the command with the --verbose flag appended to the end of your command, and post the full error traceback.
It looks like you got qiime feature-classifier classify-sklearn working with the 99% OTUs, possibly confirming my hunch from above.
Looks like you’ve moved onto a different command and a new problem. This one is even easier to fix.
You are not giving the barplot command a FeatureData[Taxonomy] artifact as input to the taxonomy parameter. It looks like you are using a sequence file. You need to use the output of classify-sklearn. E.g., the output named taxonomy120v2.qza in the first command example that you have.
Thanks very much for posting. I’m slightly concerned about the error that you saw originally (“this classifier does not support confidence values”). You shouldn’t have seen that error with those commands, so it means that there is probably a bug somewhere.
If it’s not too much trouble and you don’t mind sharing your data, would you mind please sharing the 80_otus.qza, ref_taxonomy.qza, and the rep-seqs-dada2.qza files that gave you that initial error with me? You could DM me if you don’t want to share them publicly.