Sampling method, system, storage medium and equipment of multi-class scatter diagram based on recursive partition

Abstract

The present disclosure provides a multi-class scatter diagram sampling method, system, storage medium and device based on recursive division, by converting the received multi-class scatter diagram into a multi-class density map; according to the multi-class density map of each adjacent area The density difference recursively divides the space into a KD tree structure; based on the KD tree structure, we start backtracking from all leaf nodes to find the node that can maintain both the rare class point and the relative class density. From this node, we recursively assign class labels to ensure that every At least one point of the class and the class density order is consistent with that before sampling. Finally, a point with the assigned class label is selected from each leaf node as the sampling result output, which can be quickly calculated while faithfully maintaining the relative data density and relative class density. , and displays the sampling results of major outlier points and rare class points, which helps improve the efficiency of analyzing data visualized with multi-class scatter plots.

Description

technical field [0001] The disclosure belongs to the field of image information processing, and in particular relates to a method, system, storage medium and equipment for sampling multi-class scatter diagrams based on recursive division. Background technique [0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art. [0003] A scatter plot (scatter plot) in regression analysis refers to the distribution of data points on a Cartesian coordinate plane. A scatter plot can effectively present variables by encoding data points into visual markers (for example, dots). correlations and outliers, as well as other patterns in the data. Also, by color-coding visual markers based on class labels, multiclass scatterplots are effective in visualizing data with class labels and observing correlations among multiple classes. In addition, many times, high-dimensional data reduced to 2D space...

AI Technical Summary

Problems solved by technology

Because large jitter may introduce false patterns that do not exist, which may lead to false perception by the observer, it has great limitations, and such methods cannot handle severe overdrawing

[0008] Do Density Estimation: This class of methods is an alternative to scatterplots that display discrete scatterplots as either a color-coded density map or a set of contour lines, in this way denser regions can be better characterized, However, outliers and sparse regions may be missed; moreover, visual observation of color-coded multi-class density fields is difficult due to color mixing, especially for overlapping regions.

[0009] Sampling optimization, this type of technology mainly solves the problem of overdrawing by optimizing the sampling process, mainly including random sampling, non-uniform sampling, blue noise sampling, etc. Random sampling can preserve dense areas and relative density differences, but it is easy to lose outliers Points and sparse areas; non-uniform sampling is easy to ignore the difference in different density areas; blue noise sampling imports special local patterns, and at the same time the processing speed is slow when the number of original data points is very large, and it will be unreasonable when the constructed density field is not good. the result of

[0010] To sum up, there are currently several ways to deal with overdrawing problems that have their own limitations.

Images