On 25.07.19 12:02, Dylan Aïssi wrote: > Hi Steffen, > > Le mar. 23 juil. 2019 à 18:53, Steffen Möller <[email protected]> a > écrit : >> This reminds me of an ongoing discussion with Michael in particular >> about a nice higher-ranking paper on Debian Med. Especially with the >> very adoption of Conda, also by the industry pushing their binary >> software, we need to position ourselves somewhere. There is a document >> on Google docs that emerged from our Sprint in Vilnius at >> https://docs.google.com/document/d/1VuS6prCXaUvviEhJtHFZ4BWjZtBE6QyZ4wRlWmARuX4/edit?usp=sharing >> that I now extended to invite for self descriptions of various length. >> @Thomas, we would need to know how much you need. >> >> Concerning the paper I do not see a point to appear anywhere with it >> before we don't have a series of workflows properly established and as >> part of our CI. Nothing to go out in 2019 I tend to think. > I guess if the paper targets a high rank journal, it should speak a > bit about machine learning and deep learning as it is a current trend > for selling dreams :-). And this is especially true regarding our > field where there is more and more paper with machine learning > methods. Yes. It is also a bit of a carot to spawn discussions like this one, so we can find out where we want to be going. We yet fail to describe the biological workflows that Debian supports. That together with the distribution's core infrastructure was my personal threshold to get a paper out. You may well be right that machine learning is now too much of a thing in computational biology to not stress in whatever paper that we have workflows for that, too. > I don't know how to include in the paper the work of Mo Zhou regarding > the Debian Deep Learning policy. But if the DP-Policy become as > important as the DFSG for a more general policy unrelated to Debian > (i.e. OSD), this is something we should discuss I guess. The discussion would be about FAIR principles and DFSG, where this would fit in just nicely. Non-interpretable machine-learning blobs are nothing where either computational biology nor Debian shines too much. Neither want them. But we all want "apt install pytorch" - which is a bit of an intended pun as a reminder for "conda install pytorch" to have achieved exactly that. > I saw a lot of criticisms against Debian because we are not producing > new things but we are only following the other works (except maybe the > Reproducible builds project). So, this new DP-Policy could a proof > that it is wrong and we are still in the leadership.
For the Bio part of the paper we just need to show that Debian reflects the state of the art for NGS / genomic sequencing / ... you name it. And I don't want to hijack Mo's work - which may become an important position-paper in its own right. The distribution's infrastructure like reproducible builds, CI, extra platforms etc then give the extra incentive to use [aka prepare docker images with] our packages instead of others. I know at least two biological external-to-Debian projects that have their very own Conda channel to distribute their free-as-in-beer-or-worse software. The Debian equivalent would be a PPA or something better. If we don't come up with new ideas then we will just continue patching upstream to update to the latest versions of libraries so Conda folks have better experiences. Ubuntu must see that, too. Their answer is snap and friends. Maybe it is time to ask (again?) if the launchpad could also build for Debian distributions? Best, Steffen
