Published September 16, 2026 | Version 1.0.0

Replication package for "Detecting silent deprecation in PyPI: metadata-based identification of unmaintained packages without explicit announcements"

  • 1. Universidad Nacional del Altiplano, Puno, Peru

Description

# Silent deprecation detection in PyPI

Replication package for the paper:

**Detecting silent deprecation in PyPI: metadata-based identification of
unmaintained packages without explicit announcements**
K.X. Colquehuanca Zapana. Submitted to *Informatica* (2026).

## What this is

Code and derived results for a study on whether package-level
deprecation in PyPI can be inferred from metadata alone, in the case
where the maintainer gave no explicit signal.

## Source data

This package does **not** redistribute the underlying dataset. It is
publicly available and should be obtained from:

> Z. Zhong, S. He, H. Wang, B. Yu, H. Yang, P. He.
> ICSE 2025 artifact for "An empirical study on package-level
> deprecation in Python ecosystem" [data set]. Zenodo, 2024.
> https://doi.org/10.5281/zenodo.13335360

Download `supplementary_material.zip` and extract it so that this path
exists:

```
dataset/supplementary_material/inactive_task/deprecation_dataset/refined_new_final_data.csv
```

## Analysis sample

Of the 106,323 records in the source artifact, we retain the 1,885
labelled packages whose GitHub repository is **not** archived (547
deprecated, 1,338 not). Archived repositories are excluded because the
archived flag is the signal from which 94% of the positive labels were
derived; including it would constitute target leakage.

Six further fields (`in_readme`, `in_issues`, `in_setup`, `rationale`,
`replacement`, `rationale_type`) are excluded because they exist only
once a package is already known to be deprecated.

## Environment

Python 3.12. Install with:

```
pip install -r requirements.txt
```

## Scripts

Run in order from the repository root.

| Script | Purpose |
|---|---|
| `01_train_baseline.py` | First pass: model comparison against trivial baselines |
| `02_heuristic_vs_model.py` | Comparison against the inactivity heuristic, precision@k, calibration check |
| `03_calibration_and_figures.py` | Platt and isotonic calibration, permutation importance |
| `04_significance_and_ablation.py` | Bootstrap confidence intervals, ablation by attribute family |
| `05_figures_publication.py` | Publication figures (vector PDF, greyscale-safe) |
| `06_consolidate_results.py` | **Single source of truth.** Recomputes every reported number under one criterion and emits the LaTeX tables |

Scripts 01 to 04 compute average precision in three different ways
(mean across folds, pooled out-of-fold, bootstrap mean). Script 06
consolidates everything under the pooled out-of-fold criterion. **The
numbers reported in the paper come from script 06.**

## Reproducibility

Random seed fixed at 42 throughout. Stratified five-fold
cross-validation; all reported metrics are computed on pooled
out-of-fold predictions, so no observation contributes to the model
that scores it. Bootstrap uses 2,000 stratified replicates, resampling
positives and negatives separately to preserve prevalence.

Expected runtime on a laptop CPU: under a minute for script 01, a few
minutes for scripts 02 to 06.

## Outputs

- `data/*.csv` — result tables
- `data/tabla1.tex`, `tabla2.tex`, `tabla3.tex` — LaTeX tables as used in the paper
- `data/todos_los_numeros.txt` — every figure cited in the manuscript
- `figures/*.pdf` — publication figures

## Licence

Code released under the MIT licence. The source dataset retains the
licence of its original deposit.

## Contact

75453117@est.unap.edu.pe
ORCID: 0009-0004-2666-3598

Files

README.md

Files (54.2 kB)

Name Size
md5:53b094581c2f1327430a8ccb129984ab
5.6 kB Download
md5:f774cf54a289898a691045950132938e
7.4 kB Download
md5:73d5fc9c7455c78256ac91e39000d30a
9.3 kB Download
md5:28ec9f84c60ceb3f4315d896bf01d9eb
7.2 kB Download
md5:bfe3d35c7511728a4c7ab3f08b7ca23c
7.6 kB Download
md5:329e9a56ea3c47780c7b0e24a2407667
13.5 kB Download
md5:c570ea6fcfba12f1c6747ceca4377953
3.4 kB Preview Download
md5:9667cc0b36c24c6b279d3202f07f45f3
66 Bytes Preview Download

Additional details

Related works

Is derived from
Dataset: 10.5281/zenodo.13335360 (DOI)