29

What is the most pythonic way to group by multiple keys and summarize/average values of a list of dictionaries in Python please? Say I have a list of dictionaries as below:

input = [
{'dept': '001', 'sku': 'foo', 'transId': 'uniqueId1', 'qty': 100},
{'dept': '001', 'sku': 'bar', 'transId': 'uniqueId2', 'qty': 200},
{'dept': '001', 'sku': 'foo', 'transId': 'uniqueId3', 'qty': 300},
{'dept': '002', 'sku': 'baz', 'transId': 'uniqueId4', 'qty': 400},
{'dept': '002', 'sku': 'baz', 'transId': 'uniqueId5', 'qty': 500},
{'dept': '002', 'sku': 'qux', 'transId': 'uniqueId6', 'qty': 600},
{'dept': '003', 'sku': 'foo', 'transId': 'uniqueId7', 'qty': 700}
]

Desired output for aggregation:

output=[
{'dept': '001', 'sku': 'foo', 'qty': 400},
{'dept': '001', 'sku': 'bar', 'qty': 200},
{'dept': '002', 'sku': 'baz', 'qty': 900},
{'dept': '002', 'sku': 'qux', 'qty': 600},
{'dept': '003', 'sku': 'foo', 'qty': 700}
]

or average:

output=[
{'dept': '001', 'sku': 'foo', 'avg': 200},
{'dept': '001', 'sku': 'bar', 'avg': 200},
{'dept': '002', 'sku': 'baz', 'avg': 450},
{'dept': '002', 'sku': 'qux', 'avg': 600},
{'dept': '003', 'sku': 'foo', 'avg': 700}
]

I have found this: Group by and aggregate the values of a list of dictionaries in Python but it doesn't seem to give me what I want.

Community
  • 1
  • 1
Can Lu
  • 736
  • 2
  • 8
  • 17

8 Answers8

47

To get the aggregated results

from itertools import groupby
from operator import itemgetter

grouper = itemgetter("dept", "sku")
result = []
for key, grp in groupby(sorted(input_data, key = grouper), grouper):
    temp_dict = dict(zip(["dept", "sku"], key))
    temp_dict["qty"] = sum(item["qty"] for item in grp)
    result.append(temp_dict)

from pprint import pprint
pprint(result)

Output

[{'dept': '001', 'qty': 200, 'sku': 'bar'},
 {'dept': '001', 'qty': 400, 'sku': 'foo'},
 {'dept': '002', 'qty': 900, 'sku': 'baz'},
 {'dept': '002', 'qty': 600, 'sku': 'qux'},
 {'dept': '003', 'qty': 700, 'sku': 'foo'}]

And to get the averages, you can simply change the contents inside the for loop, like this

temp_dict = dict(zip(["dept", "sku"], key))
temp_list = [item["qty"] for item in grp]
temp_dict["avg"] = sum(temp_list) / len(temp_list)
result.append(temp_dict)

Output

[{'avg': 200, 'dept': '001', 'sku': 'bar'},
 {'avg': 200, 'dept': '001', 'sku': 'foo'},
 {'avg': 450, 'dept': '002', 'sku': 'baz'},
 {'avg': 600, 'dept': '002', 'sku': 'qux'},
 {'avg': 700, 'dept': '003', 'sku': 'foo'}]

Suggestion: Anyway, I would have added both the qty and avg in the same dict like this

temp_dict = dict(zip(["dept", "sku"], key))
temp_list = [item["qty"] for item in grp]
temp_dict["qty"] = sum(temp_list)
temp_dict["avg"] = temp_dict["qty"] / len(temp_list)
result.append(temp_dict)

Output

[{'avg': 200, 'dept': '001', 'qty': 200, 'sku': 'bar'},
 {'avg': 200, 'dept': '001', 'qty': 400, 'sku': 'foo'},
 {'avg': 450, 'dept': '002', 'qty': 900, 'sku': 'baz'},
 {'avg': 600, 'dept': '002', 'qty': 600, 'sku': 'qux'},
 {'avg': 700, 'dept': '003', 'qty': 700, 'sku': 'foo'}]
thefourtheye
  • 233,700
  • 52
  • 457
  • 497
6

Inspired by Eelco Hoogendoorn's answer. Here is another way to resolve this using Pandas package. The code is more readable.

import numpy as np
import pandas as pd

def sum_by_cusip_and_dept(data):
    df = pd.DataFrame(data)
    grouped = df.groupby(['sku', 'dept'])    
    sum = grouped.sum()
    return [{'sku': r[0], 'dept': r[1], 'qty': kv.to_dict().get('qty')} for r, kv in sum.iterrows()]
Can Lu
  • 736
  • 2
  • 8
  • 17
2

Using the numpy EP you can find here, you could write:

inputs = dict( (k, [i[k] for i in input ]) for k in input[0].keys())
print group_by((inputs['dept'], inputs['sku'])).mean(inputs['qty'])

However, you may want to consider using the pandas package if you are doing a lot of relational operations of this kind.

Eelco Hoogendoorn
  • 10,459
  • 1
  • 44
  • 42
1

You can put the quantities and the number of their occurrences in one big default dict:

from collections import defaultdict

counts = defaultdict(lambda: [0, 0])
for line in input_data:
    entry = counts[(line['dept'], line['sku'])]
    entry[0] += line['qty']
    entry[1] += 1

Now it is only the question to get the numbers into a list of dicts:

sums_dict = [{'dept': k[0], 'sku': k[1], 'qty': v[0]} 
              for k, v in counts.items()]
avg_dict = [{'dept': k[0], 'sku': k[1], 'avg': float(v[0]) / v[1]} for 
             k, v in counts.items()]

The results for the sums:

sums_dict

[{'dept': '002', 'qty': 600, 'sku': 'qux'},
 {'dept': '001', 'qty': 400, 'sku': 'foo'},
 {'dept': '003', 'qty': 700, 'sku': 'foo'},
 {'dept': '002', 'qty': 900, 'sku': 'baz'},
 {'dept': '001', 'qty': 200, 'sku': 'bar'}]

and for the averages:

avg_dict

[{'avg': 600.0, 'dept': '002', 'sku': 'qux'},
 {'avg': 200.0, 'dept': '001', 'sku': 'foo'},
 {'avg': 700.0, 'dept': '003', 'sku': 'foo'},
 {'avg': 450.0, 'dept': '002', 'sku': 'baz'},
 {'avg': 200.0, 'dept': '001', 'sku': 'bar'}]

An alternative version without the default dict:

counts = {}
for line in input_data:
    entry = counts.setdefault((line['dept'], line['sku']), [0, 0])
    entry[0] += line['qty']
    entry[1] += 1

The rest is the same:

sums_dict = [{'dept': k[0], 'sku': k[1], 'qty': v[0]} 
              for k, v in counts.items()]
avg_dict = [{'dept': k[0], 'sku': k[1], 'avg': float(v[0]) / v[1]} for 
             k, v in counts.items()]
Mike Müller
  • 82,630
  • 20
  • 166
  • 161
0

Like always there are lots of valid solutions, I like the defaultdict one, since I find it easier to understand.

from collections import defaultdict as df
food = df(lambda:df(lambda:df(int)))
for dct in input:  food[dct['transId']][dct['sku']][dct['dept']]=dct['qty']
output_tupl=[(d1,d2,sum(food[d1][d2][d3] for d3 in food[d1][d2]) )for d1 in food for d2 in food[d1]]
tk.
  • 626
  • 4
  • 14
0

I had some extra requirements on top of the original question. I wanted to pass the grouper around and not have to pass around the original order of the fields if you need to reconstruct the grouping key as a dict.

namedtuple() works quite well in that it allows you to sort and use ._asdict()

from collections import namedtuple

def get_grouper(fields):

    key = namedtuple('GroupingKey', fields)

    def get_key(row):
        return key(**{field: row[field] for field in fields})

    return get_key

rows = [
    {'a': 1, 'b': 1, 'c': 1},
    {'a': 1, 'b': 2, 'c': 3},
    {'a': 1, 'b': 1, 'c': 2},
    {'a': 1, 'b': 0},
    {'a': 1, 'b': 2, 'c': 4}
]

grouper = get_grouper(['a','b'])

rows = sorted(rows, key=grouper)

for k, g in groupby(rows, key=grouper):
    print(k, list(g))
Kevin Hill
  • 473
  • 5
  • 16
0

@thefourtheye If we use groupby only one key, we should check the type of key after group, if not a tuple, return a list. 

for key, grp in groupby(sorted(input_data, key = grouper), grouper):
  if not isinstance(key, tuple):
    key = [key]
ElSheikh
  • 321
  • 6
  • 28
Toanalien
  • 659
  • 1
  • 7
  • 15
0

Using pandas and duckdb you can easily query over datasets using SQL:

import pandas as pd
import duckdb

data = [
    {'dept': '001', 'sku': 'foo', 'transId': 'uniqueId1', 'qty': 100},
    {'dept': '001', 'sku': 'bar', 'transId': 'uniqueId2', 'qty': 200},
    {'dept': '001', 'sku': 'foo', 'transId': 'uniqueId3', 'qty': 300},
    {'dept': '002', 'sku': 'baz', 'transId': 'uniqueId4', 'qty': 400},
    {'dept': '002', 'sku': 'baz', 'transId': 'uniqueId5', 'qty': 500},
    {'dept': '002', 'sku': 'qux', 'transId': 'uniqueId6', 'qty': 600},
    {'dept': '003', 'sku': 'foo', 'transId': 'uniqueId7', 'qty': 700}
]

df = pd.DataFrame(data)
result = duckdb.query("""
    SELECT dept, sku, SUM(qty), AVG(qty)
    FROM df
    GROUP BY dept, sku;
""").to_df()
print(result)

Output:

  dept  sku  sum(qty)  avg(qty)
0  001  foo     400.0     200.0
1  001  bar     200.0     200.0
2  002  baz     900.0     450.0
3  002  qux     600.0     600.0
4  003  foo     700.0     700.0
orlp
  • 112,504
  • 36
  • 218
  • 315