Metadata: Information about data

Because nodes are connected together in a stream, information about the columns or fields that are available at each node is available. For example, in the Modeler UI, this allows you to select which fields to sort or aggregate by. This information is called the data model.

Scripts can also access the data model by looking at the fields coming into or out of a node. For some nodes, the input and output data models are the same, for example a Sort node simply reorders the records but doesn't change the data model. Some, such as the Derive node, can add new fields. Others, such as the Filter node can rename or remove fields.

In the following example, the script takes the standard IBM® SPSS® Modeler druglearn.str stream, and for each field, builds a model with one of the input fields dropped. It does this by:

Accessing the output data model from the Type node.
Looping through each field in the output data model.
Modifying the Filter node for each input field.
Changing the name of the model being built.
Running the model build node.

Note: Before running the script in the druglean.str stream, remember to set the scripting language to Python (the stream was created in a previous version of IBM SPSS Modeler so the stream scripting language is set to Legacy).

import modeler.api

stream = modeler.script.stream()
filternode = stream.findByType("filter", None)
typenode = stream.findByType("type", None)
c50node = stream.findByType("c50", None)
# Always use a custom model name
c50node.setPropertyValue("use_model_name", True)

lastRemoved = None
fields = typenode.getOutputDataModel()
for field in fields:
    # If this is the target field then ignore it
    if field.getModelingRole() == modeler.api.ModelingRole.OUT:
        continue

    # Re-enable the field that was most recently removed
    if lastRemoved != None:
        filternode.setKeyedPropertyValue("include", lastRemoved, True)

    # Remove the field
    lastRemoved = field.getColumnName()
    filternode.setKeyedPropertyValue("include", lastRemoved, False)

    # Set the name of the new model then run the build
    c50node.setPropertyValue("model_name", "Exclude " + lastRemoved)
    c50node.run([])

The DataModel object provides a number of methods for accessing information about the fields or columns within the data model. These methods are summarized in the following table.

Table 1. DataModel object methods for accessing information about fields or columns
Method	Return type	Description
`d.getColumnCount()`	int	Returns the number of columns in the data model.
`d.columnIterator()`	Iterator	Returns an iterator that returns each column in the "natural" insert order. The iterator returns instances of Column.
`d.nameIterator()`	Iterator	Returns an iterator that returns the name of each column in the "natural" insert order.
`d.contains(name)`	Boolean	Returns `True` if a column with the supplied name exists in this DataModel, `False` otherwise.
`d.getColumn(name)`	Column	Returns the column with the specified name.
`d.getColumnGroup(name)`	ColumnGroup	Returns the named column group or `None` if no such column group exists.
`d.getColumnGroupCount()`	int	Returns the number of column groups in this data model.
`d.columnGroupIterator()`	Iterator	Returns an iterator that returns each column group in turn.
`d.toArray()`	Column[]	Returns the data model as an array of columns. The columns are ordered in their "natural" insert order.

Each field (Column object) includes a number of methods for accessing information about the column. The table below shows a selection of these.

Table 2. Column object methods for accessing information about the column
Method	Return type	Description
`c.getColumnName()`	string	Returns the name of the column.
`c.getColumnLabel()`	string	Returns the label of the column or an empty string if there is no label associated with the column.
`c.getMeasureType()`	MeasureType	Returns the measure type for the column.
`c.getStorageType()`	StorageType	Returns the storage type for the column.
`c.isMeasureDiscrete()`	Boolean	Returns `True` if the column is discrete. Columns that are either a set or a flag are considered discrete.
`c.isModelOutputColumn()`	Boolean	Returns `True` if the column is a model output column.
`c.isStorageDatetime()`	Boolean	Returns `True` if the column's storage is a time, date or timestamp value.
`c.isStorageNumeric()`	Boolean	Returns `True` if the column's storage is an integer or a real number.
`c.isValidValue(value)`	Boolean	Returns `True` if the specified value is valid for this storage, and `valid` when the valid column values are known.
`c.getModelingRole()`	ModelingRole	Returns the modeling role for the column.
`c.getSetValues()`	Object[]	Returns an array of valid values for the column, or `None` if either the values are not known or the column is not a set.
`c.getValueLabel(value)`	string	Returns the label for the value in the column, or an empty string if there is no label associated with the value.
`c.getFalseFlag()`	Object	Returns the "false" indicator value for the column, or `None` if either the value is not known or the column is not a flag.
`c.getTrueFlag()`	Object	Returns the "true" indicator value for the column, or `None` if either the value is not known or the column is not a flag.
`c.getLowerBound()`	Object	Returns the lower bound value for the values in the column, or `None` if either the value is not known or the column is not continuous.
`c.getUpperBound()`	Object	Returns the upper bound value for the values in the column, or `None` if either the value is not known or the column is not continuous.

Note that most of the methods that access information about a column have equivalent methods defined on the DataModel object itself. For example the two following statements are equivalent:

dataModel.getColumn("someName").getModelingRole()
dataModel.getModelingRole("someName")