Frame group_by

---

group_by(self, group_by_columns, aggregation_arguments=None)

[BETA] Create summarized frame.

Parameters:

group_by_columns : list

Column name or list of column names

aggregation_arguments : dict (default=None)

Aggregation function based on entire row, and/or dictionaries (one or more) of { column name str : aggregation function(s) }.

Returns:

: Frame

A new frame with the results of the group_by

Creates a new frame and returns a Frame object to access it. Takes a column or group of columns, finds the unique combination of values, and creates unique rows with these column values. The other columns are combined according to the aggregation argument(s).

Notes

• Column order is not guaranteed when columns are added
• The column names created by aggregation functions in the new frame are the original column name appended with the ‘_’ character and the aggregation function. For example, if the original field is a and the function is avg, the resultant column is named a_avg.
• An aggregation argument of count results in a column named count.
• The aggregation function agg.count is the only full row aggregation function supported at this time.
• Aggregation currently supports using the following functions:
  • avg
  • count
  • count_distinct
  • max
  • min
  • stdev
  • sum
  • var (see glossary Bias vs Variance)

Examples

For setup, we will use a Frame my_frame accessing a frame with a column a:

>>> my_frame.inspect()

  a:str
/-------/
  cat
  apple
  bat
  cat
  bat
  cat

Create a new frame, combining similar values of column a, and count how many of each value is in the original frame:

>>> new_frame = my_frame.group_by('a', agg.count)
>>> new_frame.inspect()

  a:str       count:int
/-----------------------/
  cat             3
  apple           1
  bat             2

In this example, ‘my_frame’ is accessing a frame with three columns, a, b, and c:

>>> my_frame.inspect()

  a:int   b:str   c:float
/-------------------------/
  1       alpha     3.0
  1       bravo     5.0
  1       alpha     5.0
  2       bravo     8.0
  2       bravo    12.0

Create a new frame from this data, grouping the rows by unique combinations of column a and b. Average the value in c for each group:

>>> new_frame = my_frame.group_by(['a', 'b'], {'c' : agg.avg})
>>> new_frame.inspect()

  a:int   b:str   c_avg:float
/-----------------------------/
  1       alpha     4.0
  1       bravo     5.0
  2       bravo    10.0

For this example, we use my_frame with columns a, c, d, and e:

>>> my_frame.inspect()

  a:str   c:int   d:float e:int
/-------------------------------/
  ape     1       4.0     9
  ape     1       8.0     8
  big     1       5.0     7
  big     1       6.0     6
  big     1       8.0     5

Create a new frame from this data, grouping the rows by unique combinations of column a and c. Count each group; for column d calculate the average, sum and minimum value. For column e, save the maximum value:

>>> new_frame = my_frame.group_by(['a', 'c'], agg.count, {'d': [agg.avg, agg.sum, agg.min], 'e': agg.max})

  a:str   c:int   count:int  d_avg:float  d_sum:float   d_min:float   e_max:int
/-------------------------------------------------------------------------------/
  ape     1       2          6.0          12.0          4.0           9
  big     1       3          6.333333     19.0          5.0           7

For further examples, see Group by (and aggregate):.