Partition Data Task

Example: Partitioning the SASHELP.CLASSFIT Data Set

About the Partition Data Task

A partition is all or part of a logical file. The Partition Data task enables you to create up to four partitions created by randomly sampling the input data. Partitions can be used to develop a model. In this case, you want to train the model on part of the data and reserve some of the data for testing. Using the Partition Data task, you can save all the partitions to one output data set or save each partition in a separate data set.

Example: Partitioning the SASHELP.CLASSFIT Data Set

In the Tasks section, expand the Data folder and double-click Partition Data. The user interface for the Partition Data task opens.
On the Data tab, select SASHELP.CLASSFIT as the input data set.
In the Number of partitions box, enter 2.
In the Proportion of cases for partition 1 box, enter .5, which specifies 50% of the values should be in partition 1.
In the Proportion of cases for partition 2 box, enter .3, which specifies 30% of the values should be in partition 2.
From the Partition data sets drop-down list, select All partitions in one data set.
In the ID value for partition 1 data role, enter Test.
In the ID value for partition 2 data role, enter Train.
To run the task, click .

Here is a subset of the results:

The new _Partition_ variable in the output data set specifies the partition (either Train or Test) for the observation. For example, the data for Joyce is in the Train partition. The data for Louise is in the Test partition. This example does not specify a random seed. As a result, the task randomly assigns 50% of the observations to the Test partition and 30% of the observations to the Train partition. If you run this example again, you might see slightly different results.

Creating a Partitioned Data Set

To run the Partition Data task, you must assign values to the Proportion of cases option for each partition.

Role	Description
Roles
Stratify by	specifies separate partitions for each combination of levels. You can specify a maximum of two variables to this role.
Partition Data
Number of partitions	specifies the number of partitions. You can choose to create one, two, three, or four partitions.
Proportion of cases for partition n	specifies the proportion of cases for each partition. The sum of all the partition proportions must be less than or equal to 1.
Output Data Set
Partition data sets	specifies whether to include all partitions in one data set or put each partition in a separate data set. You can specify a unique name for each output data set.
Include non-sampled observations	specifies whether to include non-sampled observations in the output data set. Note: This option applies only if you are saving all the partitions to one data set.
Variable name for partitioned values	specifies the name for the variable that contains the partitioned values. Note: This option applies only if you are saving all the partitions to one data set.
ID value for partition n data	specifies the identifier to use for each value in a partition. Note: This option applies only if you are saving all the partitions to one data set.
Show Output Data
Show output data	specifies whether to include the output data in the results that appear on the Results tab. You can include all or a subset of the output data. The task always creates the output data set that appears on the Output Data tab. This data set is also saved to the specified location.