Method
Mi
Our reproduction of Mi et al.'s character-level ConvNetCR branch, not the complete DeepCRM model. It is trained on the highest and lowest readability quartiles of Buse, the Java subset of Dorn, and Scalabrino. The score is the readable-class probability.
Score scale: 0–1; higher is more readable.
Dataset results
| Dataset | Coverage | Spearman ρ |
|---|---|---|
| MBJP | 17/17 | -0.063 |
| Buse | 100/100 | 0.64 |
| Scalabrino | 200/200 | 0.626 |
| Dorn | 360/360 | 0.482 |
| Schnappinger | 304/304 | 0.17 |
| JetBrains | 119/119 | -0.151 |
| Java transformations | 1400/1400 | — |
| Python transformations | 1400/1400 | — |