To train and evaluate the DQN-based agents, we have used the byte-codes of these 6 open-source java softwares (as shown in the table below). In our research, the first one is used as the traning dataset and the rest is used to evaluate trained system.
Software | Number of classes | Number of lines | Number of possible refactoring actions (at the beginning of the process) |
---|---|---|---|
JSON 1.1 | 12 | 2196 | 173 |
JFlex 1.4.1 | 56 | 15094 | 1101 |
Apache-XmlRpc 3.1.1 | 100 | 6532 | 750 |
Mango | 78 | 3470 | 604 |
Beaver 0.9.8 | 81 | 7851 | 819 |
JHotDraw 5.3 | 241 | 27824 | 3359 |