Skip to content

[Bioc-devel] Use and Usability metrics / shields

15 messages · COMMO Frederic, Dan Tenenbaum, Wolfgang Huber +4 more

#
Bioc developers!

It's important that our users be able to identify packages that are suitable for 
their research question. Obviously a first step is to identify packages in the 
appropriate research domain, for instance through biocViews.

   http://bioconductor.org/packages/release/

We'd like to help users further prioritize their efforts by summarizing use and 
usability. Metrics include:

- Cross-platform availability -- biocLite()-able from all or only some platforms
- Support forum activity -- questions and comments / responses, 6 month window
- Download percentile -- top 5, 20, 50%, or 'available'
- Current build status -- errors or warnings on some or all platforms
- Developer activity -- commits in the last 6 months
- Historical presence -- years in Bioconductor

Obviously the metrics are imperfect, so constructive feedback welcome -- we 
think the above capture in a more-or-less objective and computable way the major 
axes influencing use and usability.

We initially intend to prominently display 'shields' (small graphical icons) on 
package landing pages.

Thanks in advance for your comments,

Martin Morgan
Bioconductor
#
Dear Martin

great idea.
"Current build status? could perhaps be wrapped with "Cross-platform availability? into some sort of ?Availability / Accessibility??

I wonder how informative it would be to make metrics such as
(i) citations of the associated paper
(ii) full-text mentions e.g. in PubmedCentral
actually useful. (i) could be flawed if package and paper are diverged; (ii) would require good disambiguation, e.g. like bioNerDS http://www.biomedcentral.com/1471-2105/14/194 (or other tools? not my expertise). Do we have someone with capabilities in this area on this list?

PS  Martin you?ll like Fig. 2 of their paper.

Wolfgang
1 day later
#
Dear Martin,

All of these suggestions sound good.

Wolfgang's suggestion regarding possible associated papers might be also great.

Another useful information would be to point to other publications where a given package was used, and cited.
I don't know if it's technically possible, but it would be greatly informative to know how frequently a package is used, and how it performs, in real contexts.

Frederic Commo
Bioinformatics, U981
Gustave Roussy

________________________________________
De : Bioc-devel [bioc-devel-bounces at r-project.org] de la part de Wolfgang Huber [whuber at embl.de]
Date d'envoi : samedi 9 mai 2015 19:57
? : Martin Morgan
Cc: bioc-devel at r-project.org
Objet : Re: [Bioc-devel] Use and Usability metrics / shields

Dear Martin

great idea.
"Current build status? could perhaps be wrapped with "Cross-platform availability? into some sort of ?Availability / Accessibility??

I wonder how informative it would be to make metrics such as
(i) citations of the associated paper
(ii) full-text mentions e.g. in PubmedCentral
actually useful. (i) could be flawed if package and paper are diverged; (ii) would require good disambiguation, e.g. like bioNerDS http://www.biomedcentral.com/1471-2105/14/194 (or other tools? not my expertise). Do we have someone with capabilities in this area on this list?

PS  Martin you?ll like Fig. 2 of their paper.

Wolfgang
_______________________________________________
Bioc-devel at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/bioc-devel
2 days later
#
----- Original Message -----
This has now been implemented. You can see it at e.g.

http://bioconductor.org/packages/release/bioc/html/BiocGenerics.html

The shields are also available for use externally, for example if you want to display build status on a README in Github. Just pull in the image from:

http://bioconductor.org/shields/build/release/bioc/BiocGenerics.svg

Note that devel has a different build shield:

http://bioconductor.org/shields/build/devel/bioc/BiocGenerics.svg

Thanks,
Dan
#
Sweet; you went live with the badges/shields, e.g.

  http://bioconductor.org/packages/release/bioc/html/affxparser.html

A positive side effect is that now there's a link from the package
page to to the package's check results, which I always wanted :)

Thanks for adding this

/Henrik


On Sun, May 10, 2015 at 11:39 AM, COMMO Frederic
<Frederic.COMMO at gustaveroussy.fr> wrote:
#
----- Original Message -----
That was there before (and still is, see the bottom of the Details section). But yes, it was not very visible.

Dan
#
Can it be that the ?in Bioc? shield is incorrect?
For instance, it says ?9.98 years? for vsn but the first commit was in Oct 2002
Curiously ?9.98 years? is stated for many old packages - surely we can use R for more precise date arithmetic?

	Cheers
	Wolfgang
#
----- Original Message -----
It is not counting from the first commit (as it says in the tooltip text) but from the day of the first release in which the package appeared. 

So it does not count the time that the package is in devel only.

Dan
#
----- Original Message -----
Actually, it turns out that I don't have data for packages committed prior to BioC 1.6. I was using the release manifests (and associated release dates) to calculate "years in bioc"; I'll switch to using svn logs.

Thanks,
Dan
#
On 05/10/2015 11:39 AM, COMMO Frederic wrote:
Thanks for these suggestions.

I like the idea of linking into the scientific literature, initially as part of 
the 'Citation' section on each landing page rather than as a shield (maybe a 
shield in the long term). As Wolfgang mentions it is a little more challenging 
than a one-liner to match the information available from a CITATION file (or 
automatically generated) to an appropriate search in PubMed, and because 
citations are an important formal metric it seems important to get this 
more-or-less right.

For what it's worth the more-or-less continuous stream of papers citing 
'Biocondcutor' are listed at

   http://bioconductor.org/help/publications/

The links in the 'Literature Search' box query various resources for use of the 
term 'Bioconductor'.

We have so far kept the distinction between 'available' and 'build', partly 
because builds sometimes fail for transient (e.g., connectivity) reasons or, in 
devel, because of an incomplete check-in that does not compromise the end-user 
availability and functionality of the version available via biocLite().

It's kind of amusing that (a) most of the information was already available, 
often on the landing page (like the links to build reports that Henrik mentions, 
or years in bioc), so the shields are serving just to emphasize these; and (b) 
the 'green' implies some-how 'good', but many of the shields (e.g., years in 
Bioc, posts, commits, downloads) are actually never not green. Maybe these 
shields should be white?

Thanks again for the feedback; initial response seems to be positive.

Martin

  
    
#
The common shield convention is to use blue or orange when the information
is not qualitatively good or bad, but the color choice is just subjective
in the end.

On Thu, May 14, 2015 at 10:47 AM, Martin Morgan <mtmorgan at fredhutch.org>
wrote:

  
  
#
----- Original Message -----
It does seem though that we should indicate the non-changing nature of these shields with some kind of color change. Perhaps we can come up with one that works with the other design elements on the page.

BTW, the 'posts' tag does change color; if there are 0 posts tagged with a package name, the shield is yellow; otherwise it's green.

Dan
5 days later
#
Regarding the 'posts' tag, I can see that it includes a "closed
questions" component. For example,
http://www.bioconductor.org/packages/release/bioc/html/derfinder.html
is 3/1/9/0 right now meaning that 0 questions are closed. From
https://support.bioconductor.org/info/faq/, only moderators can close
questions. That seems like quite a bit of work for the moderators. So
maybe it would be best to drop the "closed questions" component. Or
alternatively, can the author of a package moderate the posts that
have a tag corresponding to their package?

As for 'build: warnings', it seems like it will show for some devel
packages all the time. For example,
http://www.bioconductor.org/packages/devel/bioc/html/regionReport.html
always has a warning in the Windows build machine due to a mismatch in
the version of Rtools installed.

I do like these changes and the addition of shields =)
On Thu, May 14, 2015 at 10:56 AM, Dan Tenenbaum <dtenenba at fredhutch.org> wrote:
#
----- Original Message -----
Perhaps the wording is wrong; what 'closed' is supposed to mean is that the original poster has accepted an answer. I'll change 'closed' to 'accepted'.
This is a bug in devtools and may have already been fixed (but not yet propagated to CRAN).
IMO this should be reflected in the build shield.

Dan
#
So, lots of things are happening in a few months: Jim Hester starts
working at Bioconductor, we get Bioc shields/badges, Jim's covr
package is released on CRAN, snare drum, ...  am I to eager if I
already now start wishing for a hi-hat as well?

/Henrik
On Tue, May 19, 2015 at 12:47 PM, Dan Tenenbaum <dtenenba at fredhutch.org> wrote: