2

I have a dataframe of lists, each value in the list represents the mean, std, and number of values of a larger dataset. I would like to create a subindex for three values in that list.

An example dataframe is:

np.random.seed(2)
d={i: {j:[np.random.randint(10) for i in range(0,3)] for j in ['x','y','z']} for i in ['a','b','c']}
pd.DataFrame.from_dict(d,orient='index')

Which gives:

    x   y   z
a   [1, 4, 5]   [7, 4, 4]   [0, 6, 3]
b   [7, 1, 9]   [1, 3, 8]   [3, 6, 2]
c   [1, 6, 6]   [6, 5, 0]   [6, 5, 9]

I would like:

    x              y              z
    mean std count mean std count mean std count
a   1    4   5     7    4   4     0    6   3
b   7    1   9     1    3   8     3    6   2
c   1    6   6     6    5   0     6    5   9
Wesley Kitlasten
  • 417
  • 5
  • 15
  • Seems like you could do [this](https://stackoverflow.com/questions/35491274/pandas-split-column-of-lists-into-multiple-columns) then [this](https://pandas.pydata.org/pandas-docs/stable/user_guide/advanced.html). – Nick ODell Dec 13 '20 at 22:52
  • 1
    helps if you put a np.random.seed(2) or some other number, so the randomised data is constant – sammywemmy Dec 13 '20 at 23:22

1 Answers1

3

You can concatenate the inner lists with numpy concatenate and numpy vstack, build the MultiIndex columns, then generate a new dataframe:

np.random.seed(2)
d = {
    i: {j: [np.random.randint(10) for i in range(0, 3)] for j in ["x", "y", "z"]}
    for i in ["a", "b", "c"]
}
df = pd.DataFrame.from_dict(d, orient="index")
df

        x          y            z
a   [8, 8, 6]   [2, 8, 7]   [2, 1, 5]
b   [4, 4, 5]   [7, 3, 6]   [4, 3, 7]
c   [6, 1, 3]   [5, 8, 4]   [6, 3, 9]

data = np.vstack([np.concatenate(entry) for entry in df.to_numpy()])
columns = pd.MultiIndex.from_product([df.columns, ["mean", "std", "count"]])
pd.DataFrame(data, columns=columns, index = df.index)


                   x                 y                    z
    mean    std count   mean    std count   mean    std count
a      8    8   6        2      8   7       2       1   5
b      4    4   5        7      3   6       4       3   7
c      6    1   3        5      8   4       6       3   9

UPDATE : October 5, 2021

Another option is to convert the initial dataframe to a dictionary and concatenate with pd.concat :

outcome = {k:pd.DataFrame([*v], 
                          columns = ['mean', 'std', 'count'], 
                          index = v.index) 
           for k,v in df.items()}

pd.concat(outcome, axis = 1)
 
     x              y              z          
  mean std count mean std count mean std count
a    8   8     6    2   8     7    2   1     5
b    4   4     5    7   3     6    4   3     7
c    6   1     3    5   8     4    6   3     9

sammywemmy
  • 27,093
  • 4
  • 17
  • 31