The knowledge gained by a trained mining model can be retrieved in the form of a rowset that represents a graph of nodes, where each row corresponds to one node. The rowset can be fetched either by sending a Discover or schema rowset request for DMSCHEMA_MINING_MODEL_CONTENT or by executing the "SELECT * FROM model.CONTENT" DMX query.

 

Here we describe the layout of the CONTENT rowset for the Microsoft_Clustering algorithm.

 

The Microsoft_Clustering algorithm generates one root node and then one node for each cluster found.

The Cluster nodes appear as children of the root node.

 

 

 

The root node NODE_TYPE is 1 as in all other algorithms.

The description is ‘All’ and the default Caption is ‘Cluster Model’.

The CHILDREN_CARDINALITY column contains the number of clusters found by the algorithm.

The NODE_SUPPORT column contains the total number of cases in the training set

            The NODE_DISTRIBUTION contains one row for each attribute/value pair, ordered by the attribute number then by state number.

                        The Support and probability / variance values in the distribution table represent the attribute distribution in the training set (marginal support, probability and variance)

 

 

The Cluster nodes have NODE_TYPE = 5 (in Adomd.Net this corresponds to the MiningNodeType.Cluster enumeration value).

            The default NODE_CAPTION is ‘Cluster N’ where N is the 1-based index of the cluster (it can be changed with UPDATE)

            The NODE_PROBABILITY contains the probability for the current cluster

            The NODE_SUPPORT contains the hard support for the current cluster (number of cases in the training set that were classified as belonging to this cluster)

            The NODE_DISTRIBUTION contains one row for each attribute/value pair, ordered by the attribute number then by state number. For each NODE_DISTRIBUTION row:

-   ATTRIBUTE_NAME is the name of the attribute

-   ATTRIBUTE_VALUE is the state of the attribute

-   VALUETYPE describes the value type

-   SUPPORT is the support for the current state/value in the current cluster

-   PROBABILITY is the probability for the current attribute state in the current cluster

-   VARIANCE – defined only for continuous, variance in side this cluster

            The NODE_DESCRIPTION contains a natural language description of the cluster as an enumeration of attribute states that favor that particular cluster, in the order of importance,  such as:

                       PetalWidth >= 2.09,

    Class = Iris-virginica,

    4.6 <= PetalLength <= 6.64

            The attribute descriptions are separated by a newline character