Application Overview
In this wiki page, we highlight the overall design, purpose, and intentions behind the provided default application. To look at the relevant class implementations, see the DefaultToolBox.class which contains all of the default implementations we add to the DefaultToolBox.
Upon running the default application you can expect a screen such as this:
You may notice three primary sections of the application. On the left, we have a number of JPanels grouped into a Glimpse Tile responsible for various CRUD operations and general configurations determining which curves to calculate and display. In the bottom section we have a collection of JPanels grouped into a Glimpse Tile handling analytical information about particular data. In the remaining section, a Glimpse Graph is used to display any calculated curves from the panels on the left.
The data sheet panel, at first glance, can seem to be the least straightforward of the panels. From the top down, the first JTextArea provides a location for users to name the Data Set to be added.
Clicking the “Create New Classifier Set” button will calculate a ROC Curve for all of the data read in as input during start-up of the application.
To apply a filter and calculate a ROC Curve for a subsequent subset of the data, the JTable below the “Create New Classifier Set” button is houses a collection of attributes each with their own collection of values. These are denoted as TagHeaders and Tags, respectively. For instance, the “classifier” TagHeader contains potential Tags “classifier0” and “classifier1.” Users can select one or more of the available Tags to build up a positive whitelist of Tags to be included specific to a particular TagHeader. Using CTRL + Left Click allows for selecting or deselecting a specific Tag while maintaining the remaining list of currently selected Tags. Holding Left Click and dragging the mouse across desired Tags is also an option for selecting multiple values.
The “Union with Classifier Set” button alongside the corresponding dropdown menu of existing data sets will perform a union operation between the set generated by the currently selected values in the Tags Table and the data set currently selected in the drop down. The Set Operations JLabel describes the Tag & Set operation combinations that created the currently selected data set from the dropdown menu.
The Data Set Panel contains a list of all data sets created by the Data Sheet Panel. These data sets can be freely renamed by Double Left Click, modifying the name, and hitting enter. To toggle whether the data set is displayed on the graph, click the check box in the far right of the Data Set Panel. By default, newly created data sets will display on the graph.
The Options Panel provides the ability to modify particular settings on the Graph. The X-Axis and Y-Axis drop-down menus allow users to change the particular axes of the graph. The Vertical Average and Variance checkboxes will add additional visual representations to the graph, the first computing the vertical average of all displayed data sets, and the second displaying the variance between them. Lastly, the Export Graph File Path and Export Graph buttons allow users to store the graph visuals as a .png file and persist it to the local file system.
TODO: Add component on LAF after that is merged in.
When selecting a particular set of points by clicking along the x-axis, various data at the point level is displayed in their respective panels. When multiple curves are displayed, the confusion matrices and point info values will be displayed in a horizontally expanding list, labeled by the data set’s name.
The Graph Panel houses the Graph on which the ROC Curves among other functions (Vertical Average, for instance) are plotted. Currently a Single Left Click selects the collection of points at a given x-value for all evaluation sets. The analytical data about these points is then displayed in the Analytics Tile. A Double Left Click allows the user to drag the mouse pointer across a domain of x-values, Left Clicking once to close the domain. Once selected, the domain is used to calculate the area under the curve.