Re: [Matplotlib-users] boxplot

[Eric F. <ef...@ha...>]

Pierre GM wrote:
>> And for the work I'm doing, I have a
>> different number of observations and data points on different days,
>> so it's a pain that the current boxplot infrastructure expects all of
>> the boxes to be in a single array. Hence my questions.
> 
> Ah OK, now I get it. Sorry for being a bit slow today.
> So yes, there's a problem here. The sequence is transformed into an array 
> with 'asarray', which obviously won't work if you have different sizes of 
> dataset.
> 
> But there's a trick, tested on numpy:
> 
> import numpy
> set1 = (rand(50)+1) * 100
> set2 = (rand(25)+2) * 100
> data = array([set1,set2], dtype=numpy.object_)
> boxplot(data,positions=[732659.,732660.])
> 
> The trick here is to force the data as objects, and not as floats. Your array 
> is in fact an array of arrays of different sizes. That's enough to fool 
> asarray.

It sounds like the real problem is that the initial use of asarray in 
boxplot is a bug--it should transparently support an object array, as 
you suggest (but numpy only), or an ordinary array, *or* a list or tuple 
of data vectors, and all this should be clear in the docstring and the 
example.  Correct?  I hope so, because I have made the change in svn. 
(Well, maybe not enough changes to the example yet.)

Eric