Research Question 6

RQ6: To what extent have the co-changed clones in deep learning projects been involved in bug fixes?

Glossary

  • NCC: the total number of co-changed clones detected.

  • NBCC: the number of bug-prone co-changed clones

  • NBC: the number of bug-fixing commits

  • NCCBC: the number of bug-fixing commits related to co-changed clones

  • the Pt column shows the percentage of NBCC to NCC or the percentage of NCCBC to NBC.

Numbers of bug-prone co-changed clones

Subject NCC NBCC Pt
deeplake 5 5 100.0%
transformers 773 739 95.6%
stellargraph 22 21 95.5%
texar 22 21 95.5%
ignite 54 51 94.4%
tflearn 41 37 90.2%
autokeras 9 8 88.9%
tfx 21 18 85.7%
ludwig 227 186 81.9%
catalyst 70 53 75.7%
deepchem 39 27 69.2%
MONAI 51 34 66.7%
TensorLayer 89 56 62.9%
tianshou 272 168 61.8%
coach 10 6 60.0%
keras 313 180 57.5%
horovod 6 3 50.0%
torch-points3d 13 6 46.2%
ray 32 11 34.4%
imgaug 366 114 31.1%
DeepLabCut 17 5 29.4%
deepvariant 15 3 20.0%
TTS 5 1 20.0%
allennlp 25 5 20.0%
chainer 19 3 15.8%
nni 29 3 10.3%
clearml 396 11 2.8%
pyod 140 3 2.1%
addons 4 0 0.0%
DeepPavlov 1 0 0.0%
tensor2tensor 7 0 0.0%
tensorpack 1 0 0.0%
luminoth 1 0 0.0%
DIG 18 0 0.0%
Median 22 7 48.1%
Avg 92 52 57.1%
Total 3,113 1,778 -

Numbers of bug-fixing commits of 34 projects

Subject NBC NCCBC Pt
tianshou 17 7 41.2%
autokeras 6 1 16.7%
coach 50 6 12.0%
ignite 230 27 11.7%
TensorLayer 111 12 10.8%
tflearn 205 22 10.7%
keras 292 31 10.6%
stellargraph 119 12 10.1%
transformers 313 26 8.3%
texar 97 8 8.2%
horovod 85 7 8.2%
catalyst 37 3 8.1%
MONAI 305 23 7.5%
torch-points3d 141 9 6.4%
chainer 49 3 6.1%
ludwig 128 7 5.5%
tfx 42 2 4.8%
imgaug 614 29 4.7%
clearml 45 2 4.4%
allennlp 77 3 3.9%
deepchem 1,004 36 3.6%
pyod 82 2 2.4%
deepvariant 92 2 2.2%
nni 147 3 2.0%
ray 450 8 1.8%
DeepLabCut 72 1 1.4%
deeplake 89 1 1.1%
TTS 146 1 0.7%
addons 50 0 0.0%
DeepPavlov 14 0 0.0%
tensor2tensor 19 0 0.0%
tensorpack 50 0 0.0%
luminoth 23 0 0.0%
DIG 37 0 0.0%
Median 87 3 4.7%
Avg 154 9 5.6%
Total 5,238 294 -
0%