Systems Architects and Engineers specialising in the effective use of Systems Architecture, Methods and Tools since 1982

Matrix Types

To perform clustering and the subsequent processes on a matrix use the clustering manager form. You can enter the clustering manager either from one of the relation review forms or from the facilities page of the main menu. In the latter case, click one of the radio buttons of the Cluster Matrix button. Available matrix types are shown by the appearance of the button labels. Labels for existing matrices are shown in black while other buttons are disabled with labels shown in grey.

MXC - the matrix using relation value codes and limited to sphere items for which there are existing relations

MXD - the clustered matrix using relation value codes and limited to sphere items for which there are existing relations; it has been created by clustering the precursor MXC version

MXN - the matrix using relation value names and limited to sphere items for which there are existing relations

MXE - the matrix using relation value codes and including all sphere items which met the filtering criteria whether or not there were existing relations.

The clustering manager form will be displayed. Note that if none of the matrices exists, no choice will be available and the form cannot be used.

The Clustering Manager includes the facilities to:

cluster an existing matrix and report on any unclusterable items;

transfer the newly defined clusters to the appropriate sphere and the relations between the columns and rows of the matrix to these clusters to their corresponding relation tables.

Clustering

To cluster the current matrix table you should ensure that each relation value which you wish to include in the clustering process is checked. (See Clustering Method for more details of the clustering algorithm used.) Now click the Cluster button to initiate the process. When it is complete, the button will come back up. The new clustered matrix table "MXD" is now available for export etc. Any earlier clustered matrix table in the current project will have been overwritten. If no relationship value is chosen, the clustering process will abort. As clustering is carried out, the matrix file is validated in respect of the presence of "clusterable" items: it is normally the case that each column of the matrix contains at least one such relation. If this is not so, a message will be presented and the non-compliant column items will be listed in the table on the left of the form. The message also gives the option of printing a report which identifies these items. If you have changed the set of clusterable relation values, you will be prompted to decide whether to record your new choice as the default for the relationship. As a matter of methodology, it is not unusual to find that a significant number of rows of a matrix are not included under the commutative clustering algorithm and you may wish to attach them to the clusters discovered by using a subsidiary clusterable relation value. This can be accomplished by adding or modifying relationships after the first clustering pass and then changing the set of clusterable ids.

Cluster to Sphere Transfer

When clustering is complete, you may opt to transfer the clusters to an existing sphere or to a new sphere and to create a new set of relations between the original matrix items and the newly created clusters. To do this, identify a sphere in the Cluster Sphere list box and an initial sequence number. Then click the Set Struc button.

If the standard cluster to row or column sphere relationship values are not already defined they will be inserted into the relation table spheres. Clusters will be assigned names constructed from the name of the level to which they are assigned in the target sphere. E.g. a level name of "Cluster" will give "Cluster1", Cluster2" etc. You can change these names to more appropriate values later if you wish using the sphere manipulation forms. Cluster items will be given sequence keys starting from the first value which you supply in accordance with the rules for sequencing given in Sequence Codes. For example if you choose the initial sequence number, "BA01", the cluster items will be given numbers, "BA01", "BA02", "BA03" etc. By careful choice of these numbers it is possible to cluster and identify clusters at different levels and to preserve hierarchic relationships between them. The relationship values applied are;

for a row or column which is part of a cluster - "Includes" for the cluster to row or column item and "Included In" for its inverse cluster to item;

for a row which uses the product of a cluster of which it is not a part -"Supplied By" for the row to cluster and "Supplies" for the inverse cluster to row;

for a column which is used by a cluster of which it is not a part - "Used By" for the column to cluster and "Uses" for its inverse cluster to column.

These values are independent of the existing setup of the relationships.