import pandas as pd
index = pd.MultiIndex.from_tuples([(1,2, None), (1,2,3)])
print(index.get_level_values(2))

gives

Int64Index([nan, 3.0], dtype='float64')

However, I would like to get

Int64Index([nan, 3], dtype='object')

That is, Pandas should not automatically change the data type. How can I do this?

I think you can convert level values to string by astype, str.split it by . and get first value. Then create new MultiIndex:

#change multiindex
new = zip(index.get_level_values(0),index.get_level_values(1),index.get_level_values(2).astype(str).str.split('.').str.get(0))new_index = pd.MultiIndex.from_tuples(new)
print (new_index)
MultiIndex(levels=[[1], [2], [u'3', u'nan']],labels=[[0, 0], [0, 0], [1, 0]])print(new_index.get_level_values(2))
Index([u'nan', u'3'], dtype='object')