np.fix doesn't work

[andylei]

In [28]: np.fix(Series([1,2,3], range(3)))
---------------------------------------------------------------------------
Exception                                 Traceback (most recent call last)

C:\alei1\basic_mktneutral\<ipython console> in <module>()

C:\Python26\lib\site-packages\numpy\lib\ufunclike.pyc in fix(x, y)
     43     x = nx.asanyarray(x)
     44     if y is None:
---> 45         y = nx.zeros_like(x)
     46     y1 = nx.floor(x)
     47     y2 = nx.ceil(x)

C:\Python26\lib\site-packages\numpy\core\numeric.pyc in zeros_like(a)
     92     if isinstance(a, ndarray):
     93         res = ndarray.__new__(type(a), a.shape, a.dtype, order=a.flags.fnc)
---> 94         res.fill(0)
     95         return res
     96     try:

C:\Python26\lib\site-packages\pandas\core\series.pyc in fill(self, value, method)
    822         """
    823         if value is not None:
--> 824             newSeries = self.copy()
    825             newSeries[isnull(newSeries)] = value
    826             return newSeries

C:\Python26\lib\site-packages\pandas\core\series.pyc in copy(self)
    342
    343     def copy(self):
--> 344         return Series(self.values.copy(), index=self.index)
    345
    346 #-------------------------------------------------------------------------------


C:\Python26\lib\site-packages\pandas\core\series.pyc in __new__(cls, data, index, dtype, copy)
    136
    137         if index is None:
--> 138             raise Exception('Index cannot be None!')
    139
    140         # This is to prevent mixed-type Series getting all casted to

i think the bigger problem is overriding the default behavior of the fill() method, which may contribute to other numpy / scipy functions not behaving correctly.

[wesm]

In principle I agree with you that fill should be renamed to something like fillna or fill_missing since a portion of the NumPy API relies on the fill function. Unfortunately changing it would definitely cause downstream breakage in the next release-- as far as API changes go it's not too terrible to go through and fix client code but it sucks.

[wesm]

I took a look at fixing this behavior-- I'm going to remove the overridden fill() function from Series, having created fillna functions in all of the classes. There actually is a bug in NumPy which has been fixed about a month ago-- so starting with NumPy 1.6 this can be fixed:

http://projects.scipy.org/numpy/ticket/929

[wesm]

{Series, DataFrame, WidePanel}.fill have been removed. So np.fix should work on NumPy >= 1.6 (does not work with 1.5.1, though)