get_bugbug_labels no longer adds nobug type to regression training data

[avinashselvam]

#539

Modified get_bugbug_labels in defect.py to include only those data points that are labelled either regression or bug_no_regression in the training set.

Training the model without changes

72486 non-regression bugs

Cross Validation scores:
Accuracy: f0.9731263445549161 (+/- 0.0012810455820845609)
Precision: f0.9560802008310938 (+/- 0.006503421458310747)
Recall: f0.9316432362619518 (+/- 0.0042866900183067425)

Training the model after changes

71597 non-regression bugs (889 dropped)

Cross Validation scores:
Accuracy: f0.9739072259525028 (+/- 0.0019480324611321944)
Precision: f0.9561803892880535 (+/- 0.006928496874119621)
Recall: f0.9358629670750973 (+/- 0.0045683573571298)

Minor improvement in precision and recall.

Should categories task, enhancement, feature also be removed from the training data for regression?

Please let me know if I have misunderstood the task.

[marco-c]

@avinashselvam this is part of the request from #539. The other part is not to consider bugs with type "enhancement" or "task" as label 0 for the regression model.

[suhaibmujahid]

@avinashselvam are you still interested in working on this? If so, I will be glad to help.