[IPython-user] Parallel w/ scikits.timeseries dates

David Warde-Farley dwf@cs.toronto....
Thu Dec 10 20:09:55 CST 2009


On Thu, Dec 10, 2009 at 02:58:36PM -0800, Fernando Perez wrote:
> Hi Robert,
> 
> On Thu, Dec 10, 2009 at 6:43 AM, Robert Ferrell <ferrell@diablotech.com> wrote:
> >
> > For closure- on SciPy-User, Pierre pointed out to me that DateArray
> > and TimeSeries objects are picklable.  So this isn't a big an issue as
> > I had thought.  Date instances can be contained in a single-element
> > DateArray, for instance.
> >
> 
> Thanks for the info, but if I understand your report correctly, this
> is still going to be a bit annoying to use.  It seems that single
> scalars (Date instances) pulled out of DateArrays can't be pickled,
> right?  If that's the case, most parallelism systems in python are
> going to trip on timeseries objects, because it' s quite likely that
> at somep point in a computation you may break down an array and try to
> communicate individual scalars (say as the result of a longer
> computation, for example)...  Or did I misunderstand you?

If I understand the thread on Scipy-user correctly, there's information lost
when pickling DateArrays, too, so the thing to do would be to pickle an
entire TimeSeries object... so passing such a result around (a single date) 
will require wrapping in a DateArray and wrapping that in a TS. Which will be
annoying. Pierre said he was planning on fixing this soon, however.

David


More information about the IPython-user mailing list