Accept index in addition to list for set_index

[michaelbilow]

Running this code generates an error, since columns are now considered an Index and not a list. I'm not sure if the error that gets raised (set_index ends up asking for an index that is the entire length of the data frame) is actually by design. In this case, it was a bit unintuitive to figure out how to catch, since Index objects and python lists often work so similarly.

import string
import pandas as pd

data1 = 'x'*5 + 'y'*5
data2 = string.lowercase[:5]*2
data3 = range(10)
data_dict = {'Cat': list(data1), 'SubCat': list(data2), 'Vals':data3}
df = pd.DataFrame(data_dict)
ordered_df = df[['Cat', 'SubCat', 'Vals']]

correct_df = ordered_df.reset_index([x for x in ordered_df.columns[:2]])
error_df = ordered_df.set_index(ordered_df.columns[:2])

[jreback]

So you want this?

In [23]: ordered_df.set_index(list(ordered_df.columns[:2]))
Out[23]: 
            Vals
Cat SubCat      
x   a          0
    b          1
    c          2
    d          3
    e          4
y   a          5
    b          6
    c          7
    d          8
    e          9

yeh, its a pretty trivial change. want to take a go at it. Appears innocuous.

[jreback]

change here to not is_list_like(...)

[michaelbilow]

Yep, exactly.

[StephenKappel]

It's valid to pass an Index or row labels to set_index with the expected behavior of that index becoming the index. So, to achieve the desired behavior, I am opening a PR that only treats an index like a list of column labels when the index is not the same length as the DataFrame. This should avoid breaking the existing behavior while making ordered_df.set_index(ordered_df.columns[:2]) possible in most cases.