Split data frame into multiple data frames based on unique column combinations

Question

I have the following data frame:

import pandas as pd

units = [1, 1, 1, 5, 5, 5]
locations = [30, 30, 30, 32, 32, 32]
timestamps = [1, 2, 3, 1, 2, 3]
quantities = [1, 5, 3, 10, 35, 39]
data = {'units': units, 'locations': locations, 'timestamps': timestamps,
        'quantities': quantities}
df = pd.DataFrame(data=data)

that looks like this:

🐍 >>> df
   units  locations  timestamps  quantities
0      1         30           1           1
1      1         30           2           5
2      1         30           3           3
3      5         32           1          10
4      5         32           2          35
5      5         32           3          39

I need to get a list of data frames from all unique combinations of units and locations, ie something that uses df.groupby(['units', 'locations']) . The end result should look something like this:

(1, 30)
   timestamps  quantities
0           1           1
1           2           5
2           3           3

(5, 32)
   timestamps  quantities
3           1          10
4           2          35
5           3          39

is this possible, please?

Answer 1

Run a dictionary comprehension through the groupby. You can read up more on this on the Pandas doc for groupby:split-apply-combine page:

d = {name:group.filter(['timestamps','quantities']) 
     for name, group in df.groupby(['units','locations'])}

#print(d.keys())
#dict_keys([(1, 30), (5, 32)])

print(d[(1,30)])

    timestamps  quantities
0       1           1
1       2           5
2       3           3

 print(d[(5,32)])

  timestamps    quantities
3       1          10
4       2          35
5       3          39

Answer 2

another method is to use dict comp with groupby and concat

d = pd.concat(({combo : data for combo,data in df.groupby(['units','locations'])}))

print(d)

        units  locations  timestamps  quantities
1 30 0      1         30           1           1
     1      1         30           2           5
     2      1         30           3           3
5 32 3      5         32           1          10
     4      5         32           2          35
     5      5         32           3          39

Answer 3

you are right that it is just groupby:

cols = ['units','locations']
for k, d in df.drop(cols, axis=1).groupby([df[c] for c in cols]):
    print(k)
    print(d)

Output:

(1, 30)
   timestamps  quantities
0           1           1
1           2           5
2           3           3
(5, 32)
   timestamps  quantities
3           1          10
4           2          35
5           3          39