MV[index] array indexing problem

[dnadeau4]

>>> import numpy
>>> b=numpy.array([1,2,3])
>>> c=numpy.arange(10)
>>> c
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> c[b]
array([1, 2, 3])
>>> b
array([1, 2, 3])
>>> import cdms2
>>> d=cdms2.MV2.arange(10)
>>> d
variable_2
masked_array(data = [0 1 2 3 4 5 6 7 8 9],
             mask = False,
       fill_value = 999999)


>>> d[b]
> /software/anaconda2/envs/devel/lib/python2.7/site-packages/cdms2/avariable.py(1440)__getitem__()
-> if type(key) is tuple:
(Pdb) key
array([1, 2, 3])
(Pdb) n
> /software/anaconda2/envs/devel/lib/python2.7/site-packages/cdms2/avariable.py(1443)__getitem__()
-> if isinstance(key, int) and key>=len(self):
(Pdb) 
> /software/anaconda2/envs/devel/lib/python2.7/site-packages/cdms2/avariable.py(1445)__getitem__()
-> speclist = self._process_specs([key], {})
(Pdb) 
/software/anaconda2/envs/devel/lib/python2.7/site-packages/cdms2/avariable.py:1198: FutureWarning: comparison to `None` will result in an elementwise object comparison in the future.
  if (numpy.newaxis in specs):
ValueError: 'The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()'
> /software/anaconda2/envs/devel/lib/python2.7/site-packages/cdms2/avariable.py(1445)__getitem__()
-> speclist = self._process_specs([key], {})

[jypeter]

I regularly get this problem (or something similar) when passing cdms2 objects to matplotlib or other non-PCMDI packages. This is merely annoying for me, because I know the workaround. But it's much more troublesome for the average user

My workaround is to cast the array back to numpy, or numpy.ma, depending on what I'm doing:

>>> d.filled()[b]
array([1, 2, 3])
>>> numpy.ma.asarray(d)[b]
masked_array(data = [1 2 3],
             mask = False,
       fill_value = 999999)

Can you do some magic so that the beginning users are not frightened away by this horrible error? I don't even understand the error! It's a good thing I found this workaround years ago (must have been on a lucky day)

[doutriaux1]

@jypeter you could also use d.asma()

[jypeter]

I could, if I knew about it!

There are lots of available methods if I type dir(d)! But even if I had looked at help(d.asma), I'm not sure I would have found it was what I needed

Help on method asma in module cdms2.tvariable:

asma(self) method of cdms2.tvariable.TransientVariable instance

[dnadeau4]

I am not sure how to fix this one, If we have axes lon/lat/level/time I need to slice them as well. This example is good to extract data from indexes. [1,2,3]. I don't think that can easily be done in CDMS and it would be a new slicer for CDMS.

[doutriaux1]

@dnadeau4 you should do either way, in the subslice function it barfs if it's not a slice (or something like that) just test for an array first and then send the subsetting to numpy.ma. This sort of operation drops all dimensions/attributes anyway. so just return somthing like:

 return MV2.array(self.filled()[input_slicer_array])

[dnadeau4]

ok I can do that....

[dnadeau4]

I thought I could trigger a discussion on a better slicer, similar to numpy.take...

[durack1]

@dnadeau4 both CDAT/cdat#1288 and #55 cover this territory

[dnadeau4]

@durack1 nice

[dnadeau4]

8833705