Creating a dataframe from a dict where keys are tuples

Question

I have the following dict, with keys as tuples:

d = {('first', 'row'): 3, ('second', 'row'): 1}

I'd like to create a dataframe with 3 columns: Col1, Col2 and Col3 which should look like this:

Col1   Col2  Col3
first  row   3
second row   4

I can't figure out how to split the tuples other than parsing the dict pair by pair.

Answer 1

Construct a Series first, then resetting the index will give you a DataFrame:

pd.Series(d).reset_index()
Out: 
  level_0 level_1  0
0   first     row  3
1  second     row  1

You can rename columns afterwards:

df = pd.Series(d).reset_index()   
df.columns = ['Col1', 'Col2', 'Col3']   
df
Out: 
     Col1 Col2  Col3
0   first  row     3
1  second  row     1

Or in one-line, first naming the MultiIndex:

pd.Series(d).rename_axis(['Col1', 'Col2']).reset_index(name='Col3')
Out[7]: 
     Col1 Col2  Col3
0   first  row     3
1  second  row     1

Answer 2

Not that elegant as @ayhan's solution:

In [21]: pd.DataFrame(list(d), columns=['Col1','Col2']).assign(Col3=d.values())
Out[21]:
     Col1 Col2  Col3
0   first  row     3
1  second  row     1

or a straightforward one:

In [27]: pd.DataFrame([[k[0],k[1],v] for k,v in d.items()]) \
           .rename(columns={0:'Col1',1:'Col2',2:'Col2'})
Out[27]:
     Col1  Col2  Col2
0   first   row     3
1  second   row     1

Answer 3

I was curious if it were possible to use MultiIndexes, so I made an attempt. This may have its benefits if you want to specify levels. But simply following the pandas documentation example ( MultiIdex) I came up with an alternative solution.

First I created a dictionary of random data

s = {(1,2):"a", (4,5):"b", (1,5):"w", (2, 3):"z", (4,1):"p"}

Then I used pd.MultiIndex to create a Hierarchical index from the dictionary's keys.

index = pd.MultiIndex.from_tuples(s.keys())


index
Out[3]: 
MultiIndex(levels=[[1, 2, 4], [1, 2, 3, 5]],
        labels=[[0, 2, 2, 1, 0], [1, 3, 0, 2, 3]])

Then, I pass the dictionary's values directly to a pandas Series, and explicitly set the index to be the MultiIndex object I created above.

pd.Series(s.values(), index=index)
Out[4]: 
1  2    a
4  5    b
   1    p
2  3    z
1  5    w
dtype: object

Lastly, I reset the index to get the solution requested by OP

pd.Series(s.values(), index=index).reset_index()
Out[5]: 
level_0  level_1  0
0        1        2  a
1        4        5  b
2        4        1  p
3        2        3  z
4        1        5  w

This was a bit more involved, so @ayhan's answer may still be preferable, but I think this gives you an idea of what pandas may be doing in the background. Or at least give anyone the opportunity to tinker with pandas' mechanics a bit more.

Answer 4

You can easily create a data frame form a dict:

import pandas as pd

d = {('first', 'row'): 3, ('second', 'row'): 1}

df = pd.DataFrame.from_dict({'col': d}, orient='columns')

df

        |     | col |
 ------ | --- | --- |
 first  | row |   3 |
 second | row |   1 |

Now for cosmetic purposes, you can get your output dataframe with:

df = df.reset_index()
df.columns = 'Col1 Col2 Col3'.split()