Deep Neural Embedding for Software Vulnerability Discovery: Comparison and Optimization

Table 4

The statistics on code lengths for the total real-world dataset and test set involved in experiments. The functions are divided into five categories according to length.

Length of functions	<128	≥128 and <256	≥256 and <384	≥384 and <512	≥512
No. of samples (% of total sets)	58,556 (44.4%)	32,140 (24.3%)	15,052 (11.4%)	8,065 (6.1%)	18,205 (13.8%)
No. of samples (% of test set)	11,735 (44.4%)	6,446 (24.4%)	2,944 (11.1%)	1,630 (6.2%)	3944 (13.9%)