ai-se/bellwether_community

Bellwether Community detection with JS projects using r2c

bellwether_community's Issues

Create a class for Bellwether

For every node:

• Cluster IDs
• Be able to query all the projects at and below the node

r2c prediction

Preprocess data

• Remove all the commits that are not bugfix related
• Remove rows that are not involved in bugfix
• Create one big csv for each project
  • Make sure the rows in the file are chronological
  • Keep track of the commit hash (to separate into releases)

using bellwether for small projects

Compare bubble vs total

Bubble Algo:

def bubble(projects):
    project_attrs = get_attr(projects)
    birch_tree = birch(project_attrs)
    max_level = birch_tree.max_level
    for level in range(max_level,0):
        level_cluster_ids = birch_tree.get_cluster_ids[level]
        if level == max_level:
            bellwethers = get_bellwether(level_cluster_ids)
        else:
            for bellwether in bellwethers:
                parent_cluster = bellwether.parent_cluster_id
                s_project = bellwether.project
                d_projects = birch_tree.cluster[parent_cluster].projects
                score = apply_bellwether(s_project,d_projects)
                if parent_cluster not in bellwether_score.keys():
                    bellwether_score[parent_cluster] = score
                else:
                    if score > bellwether_score[parent_cluster]:
                        bellwether_score[parent_cluster] = score
            for parent_cluster_ids in bellwether_score.keys():
                bellwethers[parent_cluster_ids] = bellwether_score[parent_cluster_ids][project_name]
            for level_cluster_id in level_cluster_ids:
                if level_cluster_id not in bellwethers.keys():
                    remaining_cluster.append(level_cluster_id)
            bellwethers = bellwethers + get_bellwether(remaining_cluster)

1385 defect prediction datasets

todo

• if we apply these sanity checks to GH, would we prune much?
• results with any tuning?
• sort learned by tenability. LR, linear SVM. FFT*16 (you'll have to discretize with the goal), NB (m-estiate, laplacian, how to discretize).
  - http://robotics.stanford.edu/users/sahami/papers-dir/disc.pdf
• start reporting how long it takes to tune
• add in SMO: that's hard!!!!!!
• generality experiment : if u do anything, justify it!!!!
• tune for IFA reduction (just using FFT)

expectations

The expectation from results:

1. adequacy of predictors (Pd > 66, pf < 33)
2. FSS Is useful
3. Hyperparameter optimization is useful
4. it all scales
5. stable conclusion across
6. stable conclusion locally

CommitGruy15): does your cfs work?

for 10 data sets, showme classification results before and after cfs

Commit guru metrics

Manually group projects

Read the project README files and manually cluster them

Projects that fail

• lodash
• ember-cli
• webpack
• expect.js
• shelljs
• chai
• babel
• karma-mocha
• node-optimist
• react
• TypeScript

Rank all the top features in the projects

Use feature selection

Commit name mismatch in Typescript?

Not all commit ids showup in _commit.pkl and commited_file.pkl

Check against unsupervised

https://www.sciencedirect.com/science/article/abs/pii/S0950584918302076

New datasets with fewer keywords

Plato filenames are explicit

INCORRECT

                                          Unnamed: 0  E001  E002  E003  E004  ...  W144  W145  W146  W147  W148
0       /analysis/inputs/public/source-code/index.js     0     0     0     0  ...     0     0     0     0     0
1   /analysis/inputs/public/source-code/test/main.js     0     0     0     0  ...     0     0     0     0     0
2  /analysis/inputs/public/source-code/test/submo...     0     0     0     0  ...     0     0     0     0     0
3  /analysis/inputs/public/source-code/test/simpl...     0     0     0     0  ...     0     0     0     0     0

CORRECT

            Unnamed: 0  E001  E002  E003  E004  E005  E006  ...  W142  W143  W144  W145  W146  W147  W148
0            /index.js     0     0     0     0     0     0  ...     0     0     0     0     0     0     0
1        /test/main.js     0     0     0     0     0     0  ...     0     0     0     0     0     0     0
2   /test/submodule.js     0     0     0     0     0     0  ...     0     0     0     0     0     0     0
3  /test/simpleTask.js     0     0     0     0     0     0  ...     0     0     0     0     0     0     0

Do LDA (topic modelling) on READMEs of all the projects

Replace keywords with commit guru's words

Commit Guru 150

TODO

• do the sanity checks of 1385 help here?

Expectation from results:

1. adequacy of predictors (Pd > 66, pf < 33)
2. FSS Is useful
3. Hyerparameter optimization is useful
4. it all scales
5. stable conclusion across
6. stable conclusion locally

FILES

• results spreadsheet:
  • https://docs.google.com/spreadsheets/d/1_v_oyN3-_JuJkJYC7iA53TEs1frmm9po/edit#gid=1000346812