Silicon Graphics, Incorporated

Barry Glenn Becker

Joel D Tesler

Ron Kohavi

Daniel A Sommerfield

Roger Crawfis

A method, system, and computer program product visualizes the structure of an evidence classifier. An evidence inducer generates an evidence classifier based on a training set of labeled records. A mapping module generates visualization data files. An evidence visualization tool uses the visualization data files to display an evidence pane and/or a label probability pane. A first evidence pane display view shows a normalized conditional probability of each label value, for each attribute value. The first evidence pane display view can be a plurality of rows of pie charts. Each pie slice in a pie chart has a size which is a function of the normalized conditional probability of each label value for the respective attribute value. For each pie chart, the mapping module maps a height that is a function of the number of records in the training set associated with the evidence classifier. A second evidence pane display view shows relative conditional probabilities of a selected label value, for each attribute value. The second evidence pane display view can be a plurality of rows of bars. Bar height is a function of a conditional probability of a respective attribute value conditioned on the selected label value. Bar heights can represent Evidence For a selected label value or Evidence Against a selected label. A first label probability pane display view shows a pie chart of prior probabilities of each label value based on the training set. A second label probability pane display view shows a pie chart of posterior probabilities of each label value based on at least one selected attribute value. An importance slider controls filtering of attributes based on the importance of the attributes to a classification of unlabeled records. A count slider filters out attribute values having relatively low record counts. The evidence classifier visualization tool further provides sorting of attributes and/or attribute values. A subtracting minimum evidence capability is provided.

Method system and computer program product for visualizing an evidence classifier

A method, system, and computer program product for a new data visualization tool for determining distribution weights that represent values of a categorical variable and then mapping a distinct color to each of the weights so as to visually represent the different values of the categorical variable (or data attribute) in a scatter plot. The distinct colors of a splat are based on the distribution of categorical variable values in a corresponding bin, the distribution of which is represented by a vector. The vector contains as many locations as the number of different values for the categorical variable. The value stored in each location is typically a weight or percentage for that particular value of the categorical variable. Each location in the vector is also associated with a distinct color. The coloring of a single splat with multiple colors involves the rendering of each vector by looping through each vector location, and then based on the weight stored in that location, randomly selecting the same percentage of triangles in the splat for the color associated with that vector location. A threshold is used to help reduce confusion and decrease processing time by summing all weights below the threshold and assigning to it a single neutral color. A slider or other controller can be used to vary the value of the threshold.

Method, system and computer program product for visually approximating scattered data using color to represent values of a categorical variable

A data structure visualization tool visualizes a data structure such as a decision table classifier. A data file based on a data set of relational data is stored as a relational table, where each row represents an aggregate of all the records for each combination of values of the attributes used. Once loaded into memory, an inducer is used to construct a hierarchy of levels, called a decision table classifier, where each successive level in the hierarchy has two fewer attributes. Besides a column for each attribute, there is a column for the record count (or more generally, sum of record weights), and a column containing a vector of probabilities (each probability gives the proportion of records in each class). Finally, at the top-most level, a single row represents all the data. The decision table classifier is then passed to the visualization tool for display and the decision table classifier is visualized. By building a representative scene graph adaptively, the visualization application never loads the whole data set into memory. Interactive techniques, such as drill-down and drill-through are used view further levels of detail or to retrieve some subset of the original data. The decision table visualizer helps a user understand the importance of specific attribute values for classification.

Method, system, and computer program product for visualizing a data structure

A method, system, and computer program product are provided for a new data visualization tool that visually approximates a scatter plot and provides smooth animation of graphics by using interpolated data. An interpolation process is performed over pre-processed bins of scattered data points. A user can smoothly animate a plot along several dimensions by using an external query device to vary a selected data attribute. If the external query device (or slider) control is positioned in between discrete positions of the slider, the displayed plot corresponds to interpolated data. Adjacent data structures are determined corresponding to the position of the external query device. The adjacent data structures are merged together, then aggregated the using the spatial columns of the data structure as a unique key. An interpolated bin is generated, where the weight of the bin is interpolated and the dependent attribute is also interpolated and weighted. The interpolated dependent attribute is mapped to color in the visualized scatter plot. The plot appears as rendered splats corresponding to bin positions of the interpolated bins, where each splat has an opacity that is a function of the interpolated weight of data points in the corresponding bin.

Interpolation between relational tables for purposes of animating a data

A method, system, and computer program product visualizes the structure of an evidence classifier. An evidence inducer generates an evidence classifier based on a training set of labeled records. A mapping module generates visualization data files. An evidence visualization tool uses the visualization data files to display an evidence pane and/or a label probability pane. A first evidence pane display view shows a normalized conditional probability of each label value, for each attribute value. The first evidence pane display view can be a plurality of rows of pie charts. Each pie slice in a pie chart has a size which is a function of the normalized conditional probability of each label value for the respective attribute value. A second evidence pane display view shows relative conditional probabilities of a selected label value, for each attribute value. The second evidence pane display view can be a plurality of rows of bars. Bar height is a function of a conditional probability of a respective attribute value conditioned on the selected label value. A first label probability pane display view shows a pie chart of prior probabilities of each label value based on the training set. A second label probability pane display view shows a pie chart of posterior probabilities of each label value based on at least one selected attribute value. An importance slider controls filtering of attributes based on the importance of the attributes to a classification of unlabeled records. A count slider filters out attribute values having relatively low record counts. The evidence classifier visualization tool further provides sorting of attributes and/or attribute values. A subtracting minimum evidence capability is provided.

Method, system, and computer program product for visualizing an evidence

A method, system, and computer program product are provided for visually approximating a scatter plot. Bins of scattered data points are formed. Each axis of a scatter plot is discretized according to a binning resolution. Bin positions along each discretized scatter plot axis are determined from the bin numbers. The bins, which represent a cloud of scattered data points, are volume rendered as splats. The opacity of each splat is a function of the number (count) of data points in a corresponding bin. In one example, the opacity of a splat is determined by the following equation:

Growing community of inventors

Mountain View, CA, United States of America

Barry Glenn Becker