It would be useful (and probably reasonably easy) to add a button
to archive an experiment on the JIST process manager. The idea is
that once you have run a given experiment (successfully or not),
you can then automatically generate a .tar.gz or a .zip file with
the layout and the experiment results.
I see two versions: one that includes all the generated data of the
experiment tree and a copy of inputs, so that the archive file is
all you need to ever check and / or reproduce the experiment (there
is the issue of software stability, but we'll assume it's not a
problem here). This type of archive is important to keep and store
for papers, possibly even to distribute if you want to create open
data sets. The second version would only save the layout and the
experiment tree status information (all the .input, .output,
debug.out, debug.err) but not the data itself. That lighter archive
could be very useful for storage of less important data (or when
only the final results should be kept), and also to help users
debug their pipelines when something go wrong (rather than having
to ask for every piece of information).
Additionally, it would be good to add the possibility to write up
some text summary (including also creation date, software versions,
anything useful...) when creating the archive and store it inside
as HTML or XML or .txt
|