This paper is focused on evaluating the effect of some different techniques in machine learning speed-up, including vector caches, parallel execution, and so on. The following content will include some review of the previous approaches and our own experimental results.