1.3.4.9.19 Match Transformation: Round Extra

The Round Extra transformation allows numbers to be rounded, and extra numbers either side of the rounded number created, so that numbers that are close to one another can be brought into the same clusters.

Note that Round Extra cannot be used in comparisons.

Use the Round Extra transformation if you have numerical data with varying degrees of accuracy, and you want to cluster using an approximation of each numeric value, adding extra cluster values either side of the rounded value.

Round Extra is useful in order to avoid problems with rounding to single numeric values. For example, if you want all records with a numeric value within a given difference from another numeric value to be in the same cluster, this can be difficult to achieve when rounding all numbers to a single value.

For example, you may have a rule that all numbers that are within 10 of each other must be in the same cluster, so the numbers 32 and 41 should be in the same cluster. However, if using the conventional Round transformation, rounding down to the nearest 10 will place them in groups 30 and 40 respectively, as will rounding nearest, and rounding up will place them in groups 40 and 50. If you use Round Extra however, you can ensure they are in the same groups by generating multiple cluster values for each input value. In this case, if the values 32 and 41 are rounded to the nearest 10, then 32 will be placed in groups 20, 30 and 40, and 41 will be placed in groups 30, 40 and 50.

The following table describes the configuration options:

Configuration

Description

Options

Specify the following options: Round to nearest: allows integers to be rounded to the nearest aggregation of a given integer, such as the nearest 10, or the nearest 100.. Type: Integer. Default value: 10. Round type: drives how rounding is performed; that is, whether to round up, down or to the nearest whole value. Type: Selection (Up / Down / Nearest). Default value: Nearest. Number of tokens: determines how many distinct cluster values (tokens) will be created. Must be an odd number, as the rounded number is always output, in addition to extra numbers either side of it. For example, if this is set to 3, rounding to the nearest 10, the value 99 would output the rounded value 100, and the extra values 90 and 110. Type: Integer (Must be an odd number). Default value: 1.

Example

In this example, the Round Extra transformation is used to round integer values so that all numeric values within 10 of one another will always be in the same clusters.

Example configuration

Round to nearest: 5

Round type: Nearest

Number of tokens: 3

Example transformations

The following table shows examples of transformations using the above configuration:

Table 1-93 Example Transformations for Round Extra

Value	Values after transformation
1	1 - 0 2 - 5 3 - -5
11	1 - 10 2 - 5 3 - 15
14	1 - 15 2 - 10 3 - 20
26	1 - 25 2 - 20 3 - 30
36	1 - 35 2 - 30 3 - 40
44	1 - 45 2 - 40 3 - 50
61	1 - 60 2 - 55 3 - 65
70	1 - 70 2 - 65 3 - 75

All records that share a common cluster value after transformation will be in the same cluster, for example, the first two records above will be in the '5' cluster, and the last two records will be in the '65' cluster group.