| 摘要 |
In systems and methods for computer-based selection of identifying input for differentiating classes, training regions (each of which is associated with a defined class) are specified in a training space that is organized by data bands according to selected definitions. Windows are defined in training elements associated with data locations in the training regions. Multiple training windows are defined in the training elements in a known band in the training data. Relevance measures for training windows represent an extent of likelihood of correctly identifying class for a test location based on data band, window position within the training element, and the frequency of occurrence of data symbols in training windows at the window position. The window having the highest value relevance measure is selected as the most relevant window. Multiple most relevant windows, together with their parameters, are selected as identifying input to facilitate class differentiation in test spaces.
|
