Re: report on first qiime tutorial with Debian packages
On 12.08.21 16:31, Nilesh Patra wrote:
On 8/12/21 4:06 PM, Steffen Möller wrote:
Hello,
TL;DR: Qiime tutorial is directly usable with the packages in salsa for 2020.11.1, no need for anything redundant from Debian. We need to work on JavaScript to get q2-diversity into main.
Done and pushed, please take look and let me know if it does its work well
It builds nicely and dpkg -c looks good, alpha (and also beta) diversity
analyses ran fine! Please remind me to ask you some tons of questions at
the next videoconference :)
I followed https://docs.qiime2.org/2021.4/tutorials/moving-pictures/ and must admit that I am impressed. This is some nice work. The command lines can be copy'n'pasted directly from that page.
What surprised me a bit was that I needed about all the modules we have currently functional - it is not that there is a core functionality in the qiime package and the additional modules are just making it somewhat nicer or so - no. You have phylogeny analyses or you don't. You have diversity anlalyses or you don't.
The first trouble I could spot with this very first attempt is with the q2-diversity package. This needs JavaScript, which I (not reading the README) had missed in my first package. This however is huge:
[..]
Yes. 640 packages.
Not at all.
It just needs webpack for bundling, which is in the archive, and it needs d3 as dev depends, which is again in the archive and coincidentally I maintain this one
I however had to patch out - to almost rewrite a couple of config files, builds now. Check package.json for these things.
Also, using npm is wrong that should not be used in the build - using an npm install on webpack will obviously fetch the entire world for you and hence the huge number of deps that you see
Feel free to send a respective patch upstream.
BTW, q2-diversity (Build-+)Depends on q2-emperor which again depends on python3-emperor which are not in the archive. And both do not look ready for upload yet (in interim state)
If you make them ready, let me know.
I'll have a look. My immediate questions are answered: Debian can
perform the compute and create output files. Since this was running
remotely, I admittedly have not yet checked if emperor (which provides
an interactive 3D visualisation in your web browser) is truly functional.
The other concern is that the tutorial provided a classifier in a file format of what was from scikit-learn IIRC that was from a newer version of what we have in the archive. They want to have their users classify that themselves in a future version, so that was not too much of my immediate concern.
Is it not possible to provide an old file format and get the results?
qiime checkes versions and just refuses to continue.
Or is the compatibility completely broken in the versions? Or is it possible to achieve that with some patching?
In any case, we have an entire release to get these in properly, so yes not an immediate concern
For whatever workflow we will be performing for our testing, this will
not be of concern. The pre-classification was only a shortcut of the
tutorial.
Best,
Steffen
Reply to: