smilegeng/in5000_oss_forecasting

IN5000 Open-Source Software Evolution — Forecasting (TsFile)

This dataset is the multivariate time-series forecasting data from the TU Delft IN5000 MSc thesis "A Framework for Identifying Evolution Patterns of Open-Source Software Projects", converted from CSV to Apache TsFile format.

Each time series describes the evolution of one open-source GitHub repository over a sequence of observation steps, across 11 software-engineering activity metrics (stars, issues, commits, contributors, deployments, forks, pull requests, workflows, releases, repository size, …).

Source

• Original dataset: MattiaBonfanti-CS/IN5000-MB-TUD-Forecasting
• Original code / analysis: https://github.com/IN5000-MB-TUD/data-analysis
• Author: Mattia Bonfanti — MSc Computer Science (Software Technology), TU Delft, IN5000 master's thesis, academic year 2023/2024 (m.bonfanti@student.tudelft.nl)
• License: MIT (inherited from the original dataset)

Data structure

The source is a single CSV (time_series_data.csv, 109,882 rows × 14 columns) in long format: one row per (repository, step). Key facts derived from the data:

• 1,323 distinct repositories (unique_id, e.g. nlbdev/pipeline), each one an independent time series.
• ds is a per-series integer step index 0, 1, 2, … (ordinal observation step, not a real calendar timestamp). Series lengths vary (≈30–162 steps, median 83).
• cluster ∈ {0, 1, 2} is a fixed evolution-pattern cluster label assigned to each repository (constant within a series).
• repository is the integer encoding of unique_id (1:1 mapping).
• 10 integer activity metrics: stargazers, issues, commits, contributors, deployments, forks, pull_requests, workflows, releases, size. No missing values.

TsFile mapping

The TsFile uses the table model. One repository = one device (time-series).

TsFile role	Column(s)	Type	Notes
TAG (device dimension)	unique_id, cluster	STRING, INT64	repository identity + its evolution cluster
TIME	derived from ds	INT64 (ms)	ds value used directly as the millisecond timestamp
FIELD	repository	INT64	integer encoding of unique_id (redundant, kept on request)
FIELD	stargazers, issues, commits, contributors, deployments, forks, pull_requests, workflows, releases, size	INT64	activity metrics

• Table name: in5000_oss_forecasting. Time precision: ms.
• Within each device, rows are sorted by Time (ascending).

Conversion notes (column handling)

• ds → Time: consumed into the derived Time column and not kept as a separate field. Lossless: Time equals the original ds. Because the dataset carries no real calendar dates (only an ordinal step index), the integer step is written directly as a millisecond value.
• repository is kept as a FIELD even though it is a 1:1 numeric encoding of the unique_id tag (redundant by design choice).
• No rows dropped; the dataset has no train/test split (single CSV).

Files

.
├── README.md
└── data/
    └── in5000_oss_forecasting.tsfile

Reading the TsFile

from tsfile import TsFileReader

reader = TsFileReader("data/in5000_oss_forecasting.tsfile")
for name, table in reader.get_all_table_schemas().items():
    print(name, [c.get_column_name() for c in table.get_columns()])
# Query FIELD/TAG columns of the table 'in5000_oss_forecasting' via reader.query_table(...)

Citation

Please cite the original thesis / dataset author (Mattia Bonfanti, TU Delft IN5000, 2023/2024) and link back to the original dataset.