There are several datasets available for use:
These datasets allow you to analyse the contents of PyPI without having to download and process every package yourself. All of the statistics within the stats page are periodically generated using the datasets below.
This dataset contains information about every file within every release uploaded to PyPi, including:
The dataset should be accessed by downloading the files specified within https://github.com/pypi-data/data/raw/main/links/dataset.txt . The following command downloads the dataset from this URL:
$ curl -L --remote-name-all $(curl -L "https://github.com/pypi-data/data/raw/main/links/dataset.txt")DuckDB is a great tool for processing the dataset. It is very fast and supports SQL queries over Parquet files. The following command uses DuckDB to find the largest file ever uploaded to PyPI:
$ curl -L --remote-name-all $(curl -L "https://github.com/pypi-data/data/raw/main/links/dataset.txt")
$ duckdb -json -s "select * from '*.parquet' order by lines DESC limit 1"
[
{
"project_name": "EvenOrOdd",
"project_version": "0.1.10",
"project_release": "EvenOrOdd-0.1.10-py3-none-any.whl",
"uploaded_on": "2021-02-21 02:25:57.832",
"path": "EvenOrOdd/EvenOrOdd.py",
"size": "514133366",
"hash": "ff7f863ad0bb4413c939fb5e9aa178a5a8855774262e1171b876d1d2b51e6998",
"skip_reason": "too-large",
"lines": "20010001"
}
]
Woah, a whopping 20 million lines of code! Lets confirm it:
$ wget https://files.pythonhosted.org/packages/b2/82/c4265814ed9e68880ba0892eddf1664c48bb490f37113d74d32fe4757192/EvenOrOdd-0.1.10-py3-none-any.whl
$ unzip EvenOrOdd-0.1.10-py3-none-any.whl
$ wc -l EvenOrOdd/EvenOrOdd.py
20010000 EvenOrOdd/EvenOrOdd.py
$ tail -n6 EvenOrOdd/EvenOrOdd.py
elif num == 9999996:
return True
elif num == 9999997:
return False
elif num == 9999998:
return True
elif num == 9999999:
return False
else:
raise Exception("Number is not within bounds")
Very funny, I hope this module is a joke 😅
The dataset contains a skip_reason column. If a file is not present in the git repositories then the reason for skipping is recorded here. On the right is a list of the current skip reasons and the number of files excluded from the git repositories for each reason.
The exact reasons for skipping a file are not fully documented here, but ignored files include virtual environments accidentally uploaded to PyPI. text-long-lines means the file had very few lines, but the total size was large.
| skip reason | count | total size |
|---|---|---|
| virtualenv | 11,699,281 | 112.3 GiB |
| binary | 98,573,205 | 58.2 TiB |
| text-long-lines | 1,626,791 | 166.5 GiB |
| too-large | 8,507,853 | 9.5 TiB |
| empty | 38,282,284 | 0 B |
| version-control | 228,413 | 219.3 MiB |
| Total | 158,917,827 | 68.0 TiB |
[
{
"project_name": "tf-nightly-tpu",
"project_version": "2.17.0.dev20240222",
"project_release": "tf_nightly_tpu-2.17.0.dev20240222-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl",
"uploaded_on": "2024-02-22 17:29:51.001",
"path": "packages/tf-nightly-tpu/tf_nightly_tpu-2.17.0.dev20240222-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl/tensorflow/include/external/com_github_grpc_grpc/include/grpc++/impl/codegen/method_handler_impl.h",
"archive_path": "tensorflow/include/external/com_github_grpc_grpc/include/grpc++/impl/codegen/method_handler_impl.h",
"size": 987,
"hash": "\\x0B\\xDBb\\x06\\x16\\xBD\\x84\\xDD\\x12\\x03\\x093\\x9A\\xE7 \\xAC\\xF1\\xCA\\x98\\x0D",
"skip_reason": "",
"lines": 28,
"repository": 258
},
{
"project_name": "rendercv",
"project_version": "1.5",
"project_release": "rendercv-1.5-py3-none-any.whl",
"uploaded_on": "2024-03-27 18:15:31.604",
"path": "packages/rendercv/rendercv-1.5-py3-none-any.whl/rendercv/tinytex-release/TinyTeX/texmf-dist/fonts/type1/public/amsfonts/cm/cmssi12.pfm",
"archive_path": "rendercv/tinytex-release/TinyTeX/texmf-dist/fonts/type1/public/amsfonts/cm/cmssi12.pfm",
"size": 1154,
"hash": "\\xD1\\xB1\\x11\\x13\\xC2\\xAB\\x94_y6\\x95\\xE6\\x18&\\x19\\xDEr\\xDCJ\\x8D",
"skip_reason": "binary",
"lines": 0,
"repository": 264
}
]This is a SQLite dump of all PyPI metadata fetched from the PyPI API. It is updated daily. It can be accessed directly from the following url:Â https://github.com/pypi-data/pypi-json-data/releases/download/latest/pypi-data.sqlite.gz:
$ curl -L https://github.com/pypi-data/pypi-json-data/releases/download/latest/pypi-data.sqlite.gz | gzip -d > pypi-data.sqlite| URL | Size |
|---|---|
| https://github.com/pypi-data/pypi-json-data/releases/download/latest/pypi-data.sqlite.gz | 1.2 GiB |
| 1 links | 1.2 GiB |
CREATE table projects(
id integer not null primary key,
name text,
version text,
author text,
author_email text,
home_page text,
license text,
maintainer text,
maintainer_email text,
package_url text,
platform text,
project_url text,
requires_python text,
summary text,
yanked int,
yanked_reason text,
classifiers text,
requires_dist text,
UNIQUE(name, version)
);
CREATE table urls(
project_id int,
url text,
upload_time text,
package_type text,
python_version text,
requires_python text,
size int,
yanked int,
yanked_reason text,
foreign key(project_id) REFERENCES projects(id)
);
This dataset contains information about the pypi-data git repositories. The repositories_with_releases.jsonfile contains a list of project names contained within each git repository.
| URL | Size |
|---|---|
| https://github.com/pypi-data/data/raw/main/stats/repositories_with_releases.json | 66.7 MiB |
| https://github.com/pypi-data/data/raw/main/stats/repositories.json | 122.4 KiB |
| 2 links | 66.8 MiB |
[
{
"name": "pypi-mirror-99",
"index": 99,
"stats": {
"earliest_package": "2021-03-18T05:01:02.831403Z",
"latest_package": "2021-03-26T01:30:18.278384Z",
"total_packages": 40000,
"done_packages": 40000
},
"percent_done": 100,
"size": 1441702912,
"url": "https://github.com/pypi-data/pypi-mirror-99",
"packages_url": "https://github.com/pypi-data/pypi-mirror-99/tree/code/packages",
"projects": {}
}
]This dataset contains one row per unique Python file within every release uploaded to PyPI. Only the sha256 hash and a random path to the file is provided. This dataset is useful if you want to parse the Python files yourself, but want to avoid parsing the same file multiple times.
Like the main dataset, the unique files dataset should be accessed by downloading the linksfrom the following file :
$ curl -L --remote-name-all $(curl -L "https://github.com/pypi-data/data/raw/main/links/only_python_files.txt")| URL | Size |
|---|---|
| https://github.com/pypi-data/data/releases/download/2024-04-30-03-04/unique_python_files.parquet | 867.2 MiB |
| 1 links | 867.2 MiB |
[
{
"hash": "\\x9E\\x81\\xB3\\x00\\x063\\x9F\\x8F\\x0CV\\x16^a\\x84Lu\\xE6\\xB8\\xCB*",
"repository": 65,
"uploaded_on": "2020-04-09 08:19:58.608"
},
{
"hash": "\\xEF\\xC9r\\x1A\\xFA<\\x07Y0U~\\x1A\\xB3\\x8BOA\\x92\\x097\\x9D",
"repository": 9,
"uploaded_on": "2015-02-25 16:50:54.479"
}
]