GHPR: The dataset of GitHub PRs and issues

GHPR contains data about pull requests that have fixed one or more issues on GitHub.

The dataset can be found at /ghpr.csv. A small sample of the dataset is also available at /ghpr-sample.csv.

Structure

Each instance of GHPR contains data about an issue and a pull request, where the pull request has fixed the issue. Note that in some cases, a single pull request is linked to multiple issues or vice versa.

The dataset is a CSV file with the following columns:

• repo_id - Integer
• issue_number - Integer
• issue_title - Text
• issue_body_md - Text, in Markdown format, can be empty
• issue_body_plain - Text, in plain text, can be empty
• issue_created_at - Integer, in Unix time
• issue_author_id - Integer
• issue_author_association - Integer enum (see values below)
• issue_label_ids - Comma-separated integers, can be empty
• pull_number - Integer
• pull_created_at - Integer, in Unix time
• pull_merged_at - Integer, in Unix time
• pull_comments - Integer
• pull_review_comments - Integer
• pull_commits - Integer
• pull_additions - Integer
• pull_deletions - Integer
• pull_changed_files - Integer

The value of issue_body_plain is converted from issue_body_md. The conversion is not always perfect. In some cases, issue_body_plain still contains some Markdown tags.

The value of issue_author_association can be one of the following:

• 0 - Collaborator
• 1 - Contributor
• 2 - First-timer
• 3 - First-time contributor
• 4 - Mannequin
• 5 - Member
• 6 - None
• 7 - Owner

See GitHub docs for more details on author association.

Rows of the dataset are sorted by repository owner username, repository name, pull request number, and then issue number.

Method

The data is collected using the GitHub REST API. The data collection flow is as follows:

• For repository R:
  • For each merged pull request P in R:
    • For each issue I that is linked by P using a GitHub keyword and is in R:
      • (I, P) is a member of GHPR.

The dataset is created using GHPR Tools.

The raw data for this dataset is also available, where you can find a JSON file for each issue and pull request present in the dataset.

Data

This version of GHPR contains 14,384 instances. The data is collected in October 2020 from CNCF graduated projects, specifically, the following repositories:

Repository	# instances
containerd/containerd	351
coredns/coredns	227
envoyproxy/envoy	1,229
fluent/fluentd	161
goharbor/harbor	598
helm/helm	859
jaegertracing/jaeger	455
kubernetes/kubernetes	8,323
prometheus/prometheus	543
rook/rook	946
theupdateframework/specification	13
tikv/tikv	469
vitessio/vitess	210