API: Reshaping Series to its own shape raises TypeError

[goyodiaz]

This may be seen like a weird function call but it is actually triggered when plotting with development versions of matplotlib and the case is explicitly (and wrongly as it seems) by pandas:

>>> import pandas as pd
>>> x = pd.Series(range(5))
>>> x.reshape(x.shape)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/pymodules/python2.7/pandas/core/series.py", line 981, in reshape
    return ndarray.reshape(self, newshape, order)
TypeError: an integer is required

It is easily fixed just replacing the offending line by this one:

return ndarray.reshape(self, newshape, order=order)

[jreback]

actually....this should be changed entirely; something like this:

if not isinstance(newshape, tuple):
  newshape = tuple([ newshape ])
if newshape == self.shape:
      return self
return self.values.reshape(newshape,order=order)

IOW a reshape that does nothing is ok, other wise you get back an ndarray

a Series is by definition a 1-D object

[goyodiaz]

The current implementation (with the change I proposed) behaves well AFAICS:

>>> import pandas as pd
>>> s = pd.Series(range(4))
>>> s.reshape((4,))
0    0
1    1
2    2
3    3
dtype: int64
>>> s.reshape(4)
0    0
1    1
2    2
3    3
dtype: int64
>>> s.reshape((2, 2))
array([[0, 1],
       [2, 3]])
>>> s.reshape(3)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/pymodules/python2.7/pandas/core/series.py", line 981, in reshape
    return ndarray.reshape(self, newshape, order=order)
ValueError: total size of new array must be unchanged
>>> s.reshape((3,))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/pymodules/python2.7/pandas/core/series.py", line 981, in reshape
    return ndarray.reshape(self, newshape, order=order)
ValueError: total size of new array must be unchanged

and is pretty clear to me, except for the order check at the beginning which is irrelevant for my use cases.

But yours passes these same tests so I would be happy with it and I think matplotlib would be as well.

[jreback]

I think the order is just for compat with numpy
I agree not that useful
want to so a pull request for this change?

[jtratner]

Nitpicky: why not just add a comma at end instead of passing list to tuple
constructor?

[goyodiaz]

#4659