(AI Studio Core)
Synopsis
This operator creates a stratified sample from an ExampleSet. Stratified sampling builds random subsets and ensures that the class distribution in the subsets is the same as in the whole ExampleSet. This operator cannot be applied on data sets without a label or with a numerical label. The size of the sample can be specified on absolute and relative basis.Description
The stratified sampling builds random subsets and ensures that the class distribution in the subsets is the same as in the whole ExampleSet. For example in the case of a binominal classification, Stratified sampling builds random subsets such that each subset contains roughly the same proportions of the two values of class labels.
When there are different classes in an ExampleSet, it is sometimes advantageous to sample each class independently. Stratification is the process of dividing examples of the ExampleSet into homogeneous subgroups (i.e. classes) before sampling. The subgroups should be mutually exclusive i.e. every examples in the ExampleSet must be assigned to only one subgroup (or class). The subgroups should also be collectively exhaustive i.e. no example can be excluded. Then random sampling is applied within each subgroup. This often improves the representativeness of the sample by reducing the sampling error.
A real-world example of using stratified sampling would be for a political survey. If the respondents needed to reflect the diversity of the population, the researcher would specifically seek to include participants of various minority groups such as race or religion, based on their proportionality to the total population as mentioned above. A stratified survey could thus claim to be more representative of the population than a survey of simple random sampling or systematic sampling.
In contrast to the simple sampling operator (the Sample operator), this operator performs a stratified sampling of the data sets with nominal label attributes, i.e. the class distributions remains (almost) the same after sampling. Hence, this operator cannot be applied on data sets without a label or with a numerical label. In these cases a simple sampling without stratification should be performed through the Sample operator.
This operator is similar to the Filter Examples operator in principle that it takes an ExampleSet as input and delivers a subset of the ExampleSet as output. The difference is this that the Filter Examples operator filters examples on the basis of specified conditions. But the Sample operator focuses on the number of examples and class distribution in the resultant sample. Moreover, the samples are generated randomly. The number of examples in the sample can be specified on absolute and relative basis depending on the setting of the sample parameter.
Input
example set input (Data table)This input port expects an ExampleSet. It is output of the Filter Examples operator in the attached Example Process.
Output
- example set output (Data table)
A randomized sample of the input ExampleSet is output of this port. The class distributions of the sample is (almost) the same as the class distribution of the complete ExampleSet.
- original (Data table)
The ExampleSet that was given as input is passed without changing to the output through this port. This is usually used to reuse the same ExampleSet in further operators or to view the ExampleSet in the Results Workspace.
Parameters
- sampleThis parameter determines how the amount of data is specified.
- absolute: If the sample parameter is set to 'absolute' then the sample is created of an exactly specified number of examples. The required number of examples is specified in the sample size parameter.
- relative: If the sample parameter is set to 'relative' then the sample is created as a fraction of the total number of examples in the input ExampleSet. The required ratio of examples is specified in the sample ratio parameter.
- sample sizeThis parameter specifies the exact number of examples which should be sampled. This parameter is only available when the sample parameter is set to 'absolute'.
- sample ratioThis parameter specifies the fraction of examples which should be sampled. This parameter is only available when the sample parameter is set to 'relative'.
- use local random seedThis parameter indicates if a local random seed should be used for randomizing examples of the sample. Using the same value of local random seed will produce the same sample. Changing the value of this parameter changes the way the examples are randomized, thus sample will have a different set of examples.
- local random seedThis parameter specifies the local random seed. This parameter is only available if the use local random seed parameter is set to true.
Tutorial Processes
Stratified Sampling of the Golf data set
The 'Golf' data set is loaded using the Retrieve operator. The Filter Example Range operator is applied on it to select the first 10 examples. This is done to simplify the Example Process otherwise the filtering was not necessary here. A breakpoint is inserted here so that you can view the ExampleSet before the application of the Sample (Stratified) operator. As you can see, the ExampleSet has 10 examples. 6 examples (i.e. 60%) belong to class 'yes' and 4 examples (i.e. 40%) belong to class 'no'. The Sample (Stratified) operator is applied on the ExampleSet. The sample parameter is set to 'absolute' and the sample size parameter is set to 5. Thus the resultant sample will have only 5 examples. The sample will have the same class distribution as the class distribution of the input ExampleSet i.e. 60% examples with class 'yes' and 40% examples with class 'no'. You can verify this by viewing the results of this process. 3 out of 5 examples (i.e. 60%) have class 'yes' and 2 out of 5 examples (i.e. 40%) have class 'no'.
