How can I use the same column as used in 'values' for 'column' or 'index'?
For example:
pd.pivot_table(data, values='Survived', index=['Survived', 'Sex', 'Pclass'],
aggfunc=len, margins=True)
values and index use the same column Survived. When I try to run the above I get
ValueError: Grouper for 'Survived' not 1-dimensional
However, if instead of values='Survived' I use another column, the pivot_table works fine.
One issue I'm seeing is that you haven't set the columns argument when calling pivot_table (which tells pandas what values to use as the column headers for the pivot_table output).
A pivot table operation is actually a succession of groupby -> aggregate -> unstack. Say you have this DataFrame:
survived sex pclass other
0 False f a 29
1 True f b 6
2 True f b 22
3 False m b 55
4 False f a 59
.. ... .. ... ...
95 False f a 66
96 False f c 42
97 True m c 93
98 True m c 59
99 False f b 93
You can pivot this table using pivot_table:
pd.pivot_table(df, index='sex', columns='pclass', values='other', aggfunc=sum)
pclass a b c
sex
f 1000 840 306
m 728 851 1247
Or you can get the same result using groupby and unstack:
df.groupby(['sex', 'pclass'])['other'].sum().unstack()
pclass a b c
sex
f 1000 840 306
m 728 851 1247
The point of this short story is that pivot tables are actually groupby operations. In your case, you're trying to group by ['Survived', 'Sex', 'Pclass'] and aggregate 'Survived' again using len. That doesn't make much sense since 'Survived' is already part of output table index (which is why pivot_table gives you an error).
You can, if you really want to make this work, use groupby instead:
df.groupby(['survived', 'sex', 'pclass', 'other']['survived'].apply(len).unstack()
However, I think you actually want to achieve something else, not sure what though.