基于变值图示判定RNN基因序列分类器的稳定性

Stability of RNN Classified Gene Sequence on Variant Maps

循环神经网络(RNN)在分析整数序列的数据特征中具有优异的作用。利用其特征分类的原理,本文将预分类和分类后的基因序列进行包括移位等的序列操作,按照文中框架下模块操作得到一系列的变值概率统计图示可视化结果,通过变值图示以及其他的图示比较与分析,对RNN分类器的稳定性进行分析。在替换操作中多种类的替换关系和移位操作中长度的变化,提供丰富的可视化结果,综合交叉比较结果,有利于对RNN分类器稳定性问题进行分析和深入探索。

The recurrent neural network (RNN) has an excellent role to analyze the data characteristics of integer sequences. Using the principle of feature classification, the selected gene sequences are pre-classified, and subsequent processes are performed on the classified sequences. In this paper, the pre-classified gene sequences are segmented and shifted. For the sequence operation after classification, the gene sequence data sets with different shift-length values are obtained, and the different data sets are respectively used as the inputs of the RNN classifier. The sequences of different detection data sets are replaced to obtain the final visualized sequence data set. The variant maps are provided to show a series of visualization results of the variant probability statistics. The stability of the RNN classifier is analyzed through comparison and analysis of the variant maps and other diagrams. Multiple substitution relationships in the replacement operation change lengths of the shift operations to provide a variety of visualization and comprehensive cross-comparisons to support the analysis and in-depth exploration of the stability problems of the RNN classifier.