Association Tree Planning

[rahlk]

Generating Plans with Assoc. Rule + XTREE + Random Walk

Reviewers comment on causality

The majority of the reviewers do not agree with the key underlying assumption that there is a causal relationship between metrics and defects. In defect prediction we always rely on correlations and pretty much never claim causality (the majority of reviewers stand fairly strong grounds on this point).

Newer XTREE

Goal: Use previous changes to guide search

1. Mine association rule using FP-Growth
2. Build XTREE (as usual)
3. Perform a random walk on the tree to (a) maximize overlap of change, and (b) minimize defects

Ant

Camel

Ivy

Jedit

Log4j

Lucene

Poi

Velocity

Xalan

Xerces

General Review Comments

Reviewers also point out some other issues that require attention, which I hope would be useful for future revisions of this paper:

1. The paper needs to be positioned as a new planning approach (R2).

2. A discussion on why the proposed approach is better as compared to existing approach is needed (R2).

3. A more clear justification on why we are going back to product metrics is needed (R2), especially in the context of recent related work.

4. The description of the approach lacks details (R1).

5. The small-scale study needs to be justified (R1).

6. The dataset used could be potentially biased (see R1, R3).

7. Better Evaluation Metric (R2)

8. Correlation (not causation).