Viewing a Mining Model with the Microsoft Neural Network Viewer
The Microsoft Neural Network Viewer in Microsoft SQL Server Analysis Services displays mining models that are built with the Microsoft Neural Network algorithm. The Microsoft Neural Network algorithm creates classification and regression mining models by constructing a multilayer perceptron network of neurons. For more information about this algorithm, see Microsoft Neural Network Algorithm.
You can use the Microsoft Neural Network Viewer to select specific states of input attributes, and to investigate how the other input attributes in the model affect the state of the output attribute, also known as the predictable attribute. For example, you may know that a potential customer is middle aged, 40 to 50 years old, owns a home, and has two children who still live at home. However, you might not know what other specific traits about that person you could use to determine whether they will buy a bicycle from Adventure Works. You can explore the TM_Neural_Net model in the Adventure Works DW sample database to discover that if a person also has a high income, they will likely buy a bicycle, according to the model. Conversely, if they live more than 10 miles from their place of employment, they probably will not buy a bicycle.
To view detailed information about the equations used in the model and the patterns that were discovered, use the Microsoft Generic Content Tree viewer. For more information, see Viewing Model Details with the Microsoft Generic Content Tree Viewer or Microsoft Generic Content Tree Viewer (Data Mining Designer).
When you browse a mining model in Analysis Services, the model is displayed on the Mining Model Viewer tab of Data Mining Designer in the appropriate viewer for the model. The Microsoft Neural Network Viewer provides the following tabs for use in exploring neural network mining models:
With the Inputs tab, you can select the attributes and attribute values that the neural network model will use as inputs. When the viewer opens, the default is to include all attributes. This implies that you are asking the model which attribute values are the most important to determine the value of the selected output attribute.
To select an input attribute, click inside the Attribute column of the Input grid, and select an attribute from the drop-down list. Only attributes that are included in the model are included in the list. The first distinct value appears under the Value column. Clicking the default value reveals a list that contains all the possible states of the associated attribute. You can select the state that you want to investigate. You can select as many attributes as you want.
You can use the Outputs tab to designate the attribute for the neural network model to use an output, and the two states that you want to compare. You can only select attributes from the model that are defined as predict or predict only columns.
Use the OutputAttribute list to select an attribute. You can then select two states that are associated with the attribute from the Value 1 and Value 2 lists. These two states of the output attribute will be compared in the Variables pane.
The grid in the Variables tab contains the following columns: Attribute, Value, Favors [value 1], and Favors [value 2]. By default, the columns are sorted by the strength of Favors [value 1]. Clicking a column heading changes the sort order to the selected column.
A bar to the right of the attribute shows which state of the output attribute the specified input attribute state favors. The size of the bar shows how strongly the output state favors the input state.