These are chat archives for thunder-project/thunder

13th
Feb 2015
Jason Wittenbach
@jwittenbach
Feb 13 2015 05:39
@freeman-lab (and anyone else with an opinion): I'm looking at implementing methods on Series that allow for using single- or multi-level indices (ala multi-indices in Pandas) for aggregating/sub-selecting time series data. The base functions that I have written to work with such index structures seem to be working well, but when I go to expose the functionality, I find that I'm recreating a lot of the functionality is already implemented in code like Series.seriesStat (for aggregation...plus related helper functions) and Series.select (for selecting). How do we feel about simply adding keyword arguments onto these functions that indicate that they should consider the index structure? Or is this too complicated (keyword arguments would also be needed to indicate which levels/values to group/select over) and a completely different set of functions should be implemented?
Jeremy Freeman
@freeman-lab
Feb 13 2015 06:23
Hm, how much is really being reimplemented? For example, Series.seriesStat is basically two lines, and a lookup table. You could move that lookup table out so both functions can use it, but otherwise doesn't seem like too much overlap?
Ben Poole
@poolio
Feb 13 2015 12:03
@freeman-lab or other devs, could you checkout #100 when you get a chance? I'd like to have RASL integrated by COSYNE, and that PR is an important first step.
Jason Wittenbach
@jwittenbach
Feb 13 2015 13:49
True, the functions are really simply, but the question is: do we want another set of them, just with "byIndex" (or something like that) at the end of the name -- e.g. seriesStatByIndex?
Jeremy Freeman
@freeman-lab
Feb 13 2015 14:02
@poolio absolutely, very excited about it, will have comments today!
Jeremy Freeman
@freeman-lab
Feb 13 2015 14:36
@jwittenbach in that case, yes, I think we want another set of functions
See e.g. map and mapValues
Jason Wittenbach
@jwittenbach
Feb 13 2015 16:38
ok, sounds good
Davis Bennett
@d-v-b
Feb 13 2015 21:54

will there be a performance difference between these?

imDat = imDat.toSeries().toTimeSeries().detrend(method='nonlin',order=2).normalize()

and

imDat = imDat.toSeries()
imDat.saveAsBinarySeries(path)
imDat = tsc.loadseries(path).toTimeSeries().detrend(method='nonlin',order=2).normalize()
well, besides the obvious time spent writing files to disk in the second case