BUG: pandas 0.13 Series argmax/argmin returns index label instead of location (changed from 0.12)

[xdliao]

------------------------ pandas 0.12 --------
In [14]: pd.version
Out[14]: '0.12.0'

In [15]: a=pd.Series(range(3), index=('2','3','4'))

In [16]: a.argmax()
Out[16]: 2

In [17]: a.idxmax()
Out[17]: '4'

------------------------ pandas 0.13 --------
In [2]: pd.version
Out[2]: '0.13.0'

In [3]: a=pd.Series(range(3), index=('2','3','4'))

In [4]: a.argmax()
Out[4]: '4'

In [5]: a.idxmax()
Out[5]: '4'

[jreback]

these are equivalent are aliases to idxmin/idxmax

you could get the index position by doing

s.index.get_loc(s.idxmax()) for example or
np.argmax(s) will also work

prior to 0.13 iIIRC argmin/ max were just pass thrus to numpy

[xdliao]

Ok, the new behavior is more intuitive. It's probably good to mentioned the change in 0.13 release note.

[snth]

Just a quick note on the suggestions above:

np.argmax(s)

will not work to get the index position, however

np.argmax(s.values)

or

s.values.argmax()

does.

[shoyer]

Hmm. Looking back over this, I don't think it was a good idea to change this from the old behavior -- argmax is useful independently of idxmax. Probably too late to change back now, though.

[snth]

I agree. It seems superfluous to have them perform the same function when having argmax do something different would have been useful, as you point out.

[nmooman]

In addition to the above no.argmax() Snnth note
with pandas the following would work: np.argmax(data.values) or data.values.argmax()

You could load data file (mixed string and numerical) i.e., csv file using pandas
df=pandas.read_csv('filename1.csv') and do this to get the index:
np.argmax(df.values)
This will work.
If you use np.argmax(df,1) you will get this error like this (due to data has strings):
ValueError: could not convert string to float:

The above error message can be addressed by either uploading data using loadtxt('filename1.csv') in which the filename1.csv does not contain strings or even common delimiters.

On the other hands, df.idxmax would work on pandas loaded data that has strings and numerical. But it does not give you the index.