Pandas create a df from dict of dict of lists

Question

I've a data structure which is dynamically populated, so number of keys and sub-keys are unknown. I want to convert it into a Pandas df. The structure looks like this

    datastore = {
    "user1":{
        "time1":[1,2,3,4], 
        "time2":[5,6,7,8], 
        "time3":[1,2,3,4] },
    "user2":{ 
        "time1":[1,2,3,4], 
        "time2":[5,6,7,8] }
}

A dict of dicts with lists for value

I want to convert it into pandas df like this

index users times x y z k
0     user1 time1 1 2 3 4
1     user1 time2 5 6 7 8
2     user1 time3 1 2 3 4
3     user2 time1 1 2 3 4
4     user2 time2 5 6 7 8 
....

I've tried pd.DataFrame(dict), from_dict method but couldn't get it to work. Any help would be appreciated.

EDIT: Sorry about the syntax error, fixed

Your "dict of dicts" is syntactically incorrect. We would waste our time placing the quotes,commas, and colons around. Please clean it up. — DYZ, Jul 24 '17 at 01:24
This question seems similar, may be helpful https://stackoverflow.com/questions/13575090/construct-pandas-dataframe-from-items-in-nested-dictionary — Hatshepsut, Jul 24 '17 at 02:10
Thanks for the reply, the difference is that the number of users and time are dynamic and are named dynamically. Also for each time in user I want to create a new row. The link you posted combines both the keys to be the index. Edit: I can use the solution you linked and break the index in to two columns. However, I would like to know if there is a better solution — someRandomGuy, Jul 24 '17 at 02:15

piRSquared · Answer 1 · 2017-07-24T15:22:44.800

Option 1

pd.DataFrame.from_dict(datastore, 'index').stack() \
    .rename_axis(['users', 'times']) \
    .apply(pd.Series, index=list('xyzk')).reset_index()

   users  times  x  y  z  k
0  user1  time1  1  2  3  4
1  user1  time2  5  6  7  8
2  user1  time3  1  2  3  4
3  user2  time1  1  2  3  4
4  user2  time2  5  6  7  8

Option 2

pd.DataFrame(
    [[u, t] + l for u, td in datastore.items() for t, l in td.items()],
    columns='users times x y z k'.split()
)

   users  times  x  y  z  k
0  user1  time1  1  2  3  4
1  user1  time2  5  6  7  8
2  user1  time3  1  2  3  4
3  user2  time1  1  2  3  4
4  user2  time2  5  6  7  8

Timing

%timeit pd.DataFrame.from_dict(datastore, 'index').stack().rename_axis(['users', 'times']).apply(pd.Series, index=list('xyzk')).reset_index()
%timeit pd.DataFrame([[u, t] + l for u, td in datastore.items() for t, l in td.items()], columns='users timets x y z k'.split())

100 loops, best of 3: 2.72 ms per loop
1000 loops, best of 3: 556 µs per loop

DEBUG
If you copy and paste this code... it should run. Please try it and report back that it did run.

import pandas as pd

datastore = {
    "user1":{
        "time1":[1,2,3,4], 
        "time2":[5,6,7,8], 
        "time3":[1,2,3,4] },
    "user2":{ 
        "time1":[1,2,3,4], 
        "time2":[5,6,7,8]}
}

pd.DataFrame.from_dict(datastore, 'index').stack() \
    .rename_axis(['users', 'times']) \
    .apply(pd.Series, index=list('xyzk')).reset_index()

This gives me an error. TypeError: 'numpy.ndarray' object is not callable — someRandomGuy, Jul 24 '17 at 14:53
@someRandomGuy I added to my post. Please run that exact code and tell me if it works or not. — piRSquared, Jul 24 '17 at 15:13

score 2 · Accepted Answer · answered Jul 24 '17 at 02:33

Here's an approach

datastore = {
"user1":{
    "time1":[1,2,3,4], 
    "time2":[5,6,7,8], 
    "time3":[1,2,3,4] },
"user2":{ 
    "time1":[1,2,3,4], 
    "time2":[5,6,7,8]}
}

We can use pd.DataFrame() with the dict then stack() it then reset_index() it

df = pd.DataFrame(datastore).stack().reset_index()
print(df)
  level_0 level_1             0
0   time1   user1  [1, 2, 3, 4]
1   time1   user2  [1, 2, 3, 4]
2   time2   user1  [5, 6, 7, 8]
3   time2   user2  [5, 6, 7, 8]
4   time3   user1  [1, 2, 3, 4]

Now we 'split' the list in 0 with an apply of pd.Series and then join that back to level_1 and level_2. Some column renaming and we're done

df = df[['level_1', 'level_0']].join(df[0].apply(pd.Series))
df.columns = ['users', 'times', 'x', 'y', 'z', 'k']
print(df)
   users  times  x  y  z  k
0  user1  time1  1  2  3  4
1  user2  time1  1  2  3  4
2  user1  time2  5  6  7  8
3  user2  time2  5  6  7  8
4  user1  time3  1  2  3  4

Pandas create a df from dict of dict of lists

2 Answers2