Selects a subset of data instances from the input data set.
- Examples (ExampleTable)
Attribute-valued data set.
- Sample (ExampleTable)
Attribute-valued data set as sampled from the input data.
- Remaining Examples (ExampleTable)
Data instances from input data set that are not included in the sampled data.
Data Sampler supports provides support for several means of sampling of the data from the input channel and outputs the sampled data set and complementary data set (with instances from the input set that are not included in the sampled data set). Output is set when the input data set is set to the widget or after Sample Data is pressed.
Sampling may be stratified: if input data contains a class, sampling will try to match its class distribution in the output data sets.
Several types of sampling are supported. Random sampling can draw a fixed number of instances or create a data set with a size set as a proportion of instances from the input data set. In repeated sampling, an data instance may be included in a sampled data several times (like in bootstrap).
Cross validation, Leave-one-out or sampling that creates Multiple subsets of preset sample sizes relative to the input data set (like random sampling) all create several data samples. Which one is send to the output is determined by the data set index in Fold/Group (indices start with 1).
Schema where we have sampled 10 data instances from Iris data set and presented this selection in Scatterplot widget is shown below.