K value selection method for test data similarity division based on K-Means algorithm

摘要

The scale of test data can not be underestimated in the analysis of large-scale integrated circuit test,for the characteristics of K-Means algorithm the similarity of large amounts of data can be processed. It is convenient for subsequent operations to be processed according to the similarity such as compression, encoding, etc.. The selection of its center and the determination of the cluster number (K value) are the core of the K-Means algorithm, the selection of the K value is difficult to estimate. This paper has divided targeted and similarity for the generated exhaustive or pseudo-random test data. The K-means algorithm is combined with the constraints proposed in this paper to constrain the generation of suitable K values, and select appropriate range of K value. The experimental results based on a large number of random data have shown that when using the method in the text, the range of K values is selected in the 4-8 categories, the trend of each index is gentle, and the similarity between data is well.