Binary operations between Numpy array and Pandas series convert to array dtype before executing

[8one6]

It looks like when you do a binary operation with a numpy array and a pandas series, the Pandas series is cast as the same type as the numpy array. In the case where the array is ints but the series is floats this can have odd results in Python 2.x with division. This was surprising to me because, element-wise, a float/int computation would upcast the denominator to a float. But not here in this vectorized case:

import numpy as np
import pandas as pd

j = np.array([0] * 5)
k = np.random.randn(5)

print j / k
print (j / pd.Series(k)).values
print (pd.Series(j) / k).values
print (pd.Series(j) / pd.Series(k)).values

####output####

[-0.  0. -0. -0.  0.]
[ inf  inf  inf  inf  inf]
[-0.  0. -0. -0.  0.]
[-0.  0. -0. -0.  0.]

So in the two "pure" cases (array and array, series and series) the ints are upcast to floats and we get 0's everywhere. In the 2nd mixed case (3rd case overall), everything gets converted to floats to match the array and everything goes fine. But in the 1st mixed case (2nd case overall) everything gets converted to ints to match the array and so you get what was, at least to me, an unexpected infinite result.

[jreback]

actually this is a bug
this line: https://github.com/pydata/pandas/blob/master/pandas/core/common.py#L1351 is wrong
it should be startswith __r

reverse ops are passed to pandas (so in this example it is passed as (Series(j),k)
with the name being __rtruediv__
this is done so pandas handles these kind of things

so zeros can be filled correctly (as number does integer division and things like that so u get weird results at times)

so if u would like to do a or with your test cases and make that small change should work