Re: r27012 - /branches/nmrglue/test_suite/system_tests/nmrglue.py -- December 08, 2014 - 17:54

Hi Troels,

It's all done.  It took a while longer than I though.  That was my
fault, as I made the nmrglue objects of the relax data store a lot
more flexible.  Hence the commit is far more complicated than actually
required (http://article.gmane.org/gmane.science.nmr.relax.scm/24773).
You now have:

cdp.nmrglue -> data_store.nmrglue.Nmrglue_dict dictionary instance (it
inherits from data_store.data_classes.RelaxDictType).
cdp.nmrglue[key] -> data_store.nmrglue.Nmrglue instance (it inherits
from data_store.data_classes.Element).
cdp.nmrglue[key].dic -> the nmrglue dic object.
cdp.nmrglue[key].udic -> the nmrglue udic object.
cdp.nmrglue[key].data -> the nmrglue numpy.float32 data object.

The Nmrglue.test_save_state system test now passes.  Have a look at a
relax save file for what is now being created.

Regards,

Edward


On 8 December 2014 at 15:39, Edward d'Auvergne <edward@xxxxxxxxxxxxx> wrote:
> Ok, so I'll start moving things around in your branch.
>
> Regards,
>
> Edward
>
> On 8 December 2014 at 15:37, Troels Emtekær Linnet
> <tlinnet@xxxxxxxxxxxxx> wrote:
> > That sound good !
> >
> > 2014-12-08 15:34 GMT+01:00 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:
> > > Hi,
> > >
> > > I wouldn't bother with the matplotlib data storage.  If you like, I
> > > could give this 10 minutes of work on the nmrglue branch and have the
> > > Base64 string in XML storage fixed.  The system test is in place
> > > already to implement this, and I know exactly what needs to be done,
> > > so it won't take long.  Then we can keep the numpy format as is.  I'd
> > > probably restore some of your original ideas and have all data in
> > > cdp.nmrglue_data which a dictionary with
> > > data_store.data_classes.Element instances as its elements, and then
> > > write to_xml() and from_xml() methods to handle the numpy data as
> > > Base64 and all other structures are passed to the Element base class
> > > methods.
> > >
> > > Regards,
> > >
> > > Edward
> > >
> > >
> > >
> > >
> > > On 8 December 2014 at 15:19, Troels Emtekær Linnet
> > > <tlinnet@xxxxxxxxxxxxx> wrote:
> > > > Hi Edward.
> > > >
> > > > I am trying to look at my data now with the plots I have in my mind.
> > > >
> > > > The branch is actually in a state where I can use it.
> > > >
> > > > The only trouble I have, is the storing of the numpy array.
> > > >
> > > > I am actually not planning on saving the state, but will continuously
> > > > just
> > > > read from the original file.
> > > > But since the numpy array is saved in cdp, to get access, then storing
> > > > the
> > > > state should be possible.
> > > >
> > > > All plots will just be made with matplotlib after the shuffling around
> > > > with
> > > > the numpy array.
> > > >
> > > > Best
> > > > Troels
> > > >
> > > >
> > > >
> > > > 2014-12-08 15:14 GMT+01:00 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:
> > > > > Hi Troels,
> > > > >
> > > > > > I think I will have to leave the implementation in relax.
> > > > >
> > > > > Would it take much to get the branch in a reasonable state for merging
> > > > > back?  Even if not everything is implemented?  Well, apart from
> > > > > bringing in nmrglue 0.5 when that is released.  I'm sure that in the
> > > > > future some PhD student or postdoc will appreciate and take up what
> > > > > you have already done and expand on it.  You've already put a lot of
> > > > > polish on this, so it'd be a pity to not merge it.
> > > > >
> > > > >
> > > > > > Time goes away from pursuing the analysis, and goes into rather
> > > > > > annoying
> > > > > > stuff.
> > > > >
> > > > > With coding, it's a bit hard to know what will be quick and what will
> > > > > take a lot of time.  It's a problem I have too.
> > > > >
> > > > >
> > > > > > The idea is simple.
> > > > > > Load one or two spectrum.
> > > > > > Plot the contour, or histogram.
> > > > > > Or make a correlation plot of the two spectra.
> > > > > >
> > > > > > Store the data.
> > > > >
> > > > > Storing the data in any other format always takes more effort when
> > > > > combining data from different sources/software.  But this is pretty
> > > > > much sorted out for the nmrglue data.  You could even store the
> > > > > nmrglue data in the relax data store as a Base64 string and a new
> > > > > variable for the ndarray shape, as a shortcut.
> > > > >
> > > > >
> > > > > > Storing the axis plots is a different question.
> > > > > >
> > > > > >
> > > > > > http://stackoverflow.com/questions/7290370/store-and-reload-matplotlib-pyplot-object
> > > > > >
> > > > > > I cannot actually do it at the moment.
> > > > > > So essentially, it just storing a numpy array
> > > > >
> > > > > For matplotlib data, I don't think that storing these data structures
> > > > > will ever be possible to implement in a compatible way.  The reason is
> > > > > because matplotlib is too much of a moving target.  They do not have a
> > > > > stable API!  So I would never recommend that you tackle such a problem
> > > > > in this way.  That's just asking for pain ;)
> > > > >
> > > > > There is a much better alternative which would be more compatible with
> > > > > future matplotlib versions.  And that is that all the information
> > > > > required to create the matplotlib plot is stored in a pure
> > > > > data_store.data_classes.Element class container as simple variables
> > > > > (str, bool, int, float, or a list, tuple or dict of these).  This is
> > > > > similar to the GUI setting storage.  Then there could be a user
> > > > > function to recreate the plot using the currently stored settings.
> > > > > This would work much better.  And it would be orders of magnitude
> > > > > quicker to implement than working with matplotlib data structures.
> > > > >
> > > > > Regards,
> > > > >
> > > > > Edward