Non-Negative Matrix Factorization Model Viewer
This selection displays a successfully built feature extraction model that uses the Non-Negative Matrix Factorization (NMF) algorithm.
NMF decomposes multivariate data by creating a user-defined number of features, which results in a reduced representation of the original data.
The model viewer has three tabs as follows:
Build Settings
This tab displays the Type of the model (Feature Extraction),
the Model algorithm (Non-Negative Matrix Factorization), and Algorithm settings.
The following algorithm settings are displayed:
- Min Convergence Tolerance: The number representing the minimum convergence tolerance.
- Max Number of Iterations: The integer value representing the maximum number of iterations.
The attributes are displayed in the Attributes
grid; the following information is displayed for each attribute:
- Name
- Sparsity The sparsity of the attribute; either Sparse or Dense. By default, all attributes are Dense; this can be overwritten by users. An attribute must be Dense if its input data set is in multi-record case (transactional) data format.
- Data type: The Oracle data type of the attribute:
NUMBER
, FLOAT
, CHAR
, VARCHAR2
, etc.
- Mining type: Categorical, Numerical, Text, or Not Applicable
- Usage (See Attribute Usage.)
To save the attributes as a file, click the Save As icon (a yellow arrow above a graphic of a page) above the Attributes grid.
Features
To see the features, click the Features tab.
The following summary information is displayed:
- Error Rate: The overall error rate.
- Number Of Iterations: The actual number of iterations required for convergence.
The Features are displayed in a grid with the following columns:
- Feature: The feature identifier. Each feature is an entity that consists of multiple attributes. Each feature has an integer identifier that uniquely identify the resultant features. Each resultant feature is composed of m entries, where m is the number of attributes in the original data set. The difference among the features is the predominance of each of the attributes in the feature that is represented by the coefficient. If an attribute is highly predominant, its coefficient will be high.
- Attribute Name: Mining attribute name.
- Value: A character string value associated with an entry. The value differentiates between numeric and categorical data. The value for numeric data is NULL; for categorical data, the value is the value from the original data set.
- Coefficient: The non-negative floating-point coefficient.
The following buttons are displayed:
- Expand All: Click to expand all class and attribute nodes in the list.
- Collapse All: Click to
collapse all class and attribute nodes in the list.
To save the features as a file, click the Save As icon (a yellow arrow above a graphic of a page) above the Features grid.
Results
To see results, click the Results tab.
The tab contains a tree view of results. To view specific results,
select the result and click View. To delete a result, select one or more results and click Delete.
Task
This tab describes the task used to build the model. It shows when the task ran and what input data was used.
Copyright © 2005, Oracle. All rights reserved.