scikit-learn-versions.py

# %%
from pathlib import Path
import re
from packaging.version import Version


def find_python_version():
    content = Path("pyproject.toml").read_text()
    python_versions = re.findall(r"Python :: (\d+\.\d+)", content)
    if not python_versions:
        content = Path("setup.py").read_text()
        python_versions = re.findall(r"Python :: (\d+\.\d+)", content)

    python_versions = [Version(v) for v in python_versions]
    min_python_version = sorted(python_versions)[0]
    return str(min_python_version)


def find_dependencies_version(dep):
    return (
        subprocess.check_output([sys.executable, "sklearn/_min_dependencies.py", dep])
        .decode()
        .strip()
    )


# %%
import sys
import subprocess
import pandas as pd

# older versions done by hand
min_version_data = [
    {
        "scikit-learn": "0.21",
        "python": "3.5",
        "numpy": "1.11.0",
        "scipy": "0.17.0",
        "joblib": "0.11",
    },
    {
        "scikit-learn": "0.22",
        "python": "3.5",
        "numpy": "1.11.0",
        "scipy": "0.17.0",
        "joblib": "0.11",
    },
    {
        "scikit-learn": "0.23",
        "python": "3.6",
        "numpy": "1.13.3",
        "scipy": "0.19.1",
        "joblib": "0.11",
    },
]

version_list = [f"1.{minor}" for minor in range(7)]
dependency_list = ["numpy", "scipy", "pandas", "joblib", "threadpoolctl"]

for version in version_list:
    d = {"scikit-learn": version}
    subprocess.check_call(f"git checkout {version}.X".split())
    d["python"] = find_python_version()

    for dep in dependency_list:
        d[dep] = find_dependencies_version(dep)
    min_version_data.append(d)

min_version_df = pd.DataFrame(min_version_data)
# %%
# get release dates from Python and scikit-learn and then
# https://devguide.python.org/versions/
# https://pypi.org/project/scikit-learn/1.1.0/#history
import pandas as pd

df_list = pd.read_html("https://devguide.python.org/versions/")
df = pd.concat(df_list).astype({"Branch": str})
release_dates = {}
python_version_info = {
    version: release_date
    for version, release_date in zip(df["Branch"], df["First release"])
}
python_version_info = {
    version: pd.to_datetime(release_date)
    for version, release_date in python_version_info.items()
}
release_dates["python"] = python_version_info

# %%
import requests


def get_release_time(package, version):
    r = requests.get(f"https://pypi.org/pypi/{package}/{version}/json")
    json_content = r.json()
    sdist_list = [
        each for each in json_content["urls"] if each["packagetype"] == "sdist"
    ]
    # for some reason 0.21 doesn't have a sdist ...
    if len(sdist_list) == 0:
        sdist_list = json_content["urls"]
    return pd.to_datetime(sdist_list[0]["upload_time"])


for dep in ["scikit-learn"] + dependency_list:
    release_dates[dep] = {
        ver: get_release_time(dep, ver)
        for ver in min_version_df[dep]
        if isinstance(ver, str)
    }

release_dates


# %%
def previous_minor_release(ver):
    major, minor, *_ = ver.split(".")
    previous_minor = int(minor) - 1
    return f"{major}.{previous_minor}"


# %%

for each in ["scikit-learn", "python"] + dependency_list:
    min_version_df[f"{each}-date"] = min_version_df[each].map(release_dates[each])

for each in ["python"] + dependency_list:
    min_version_df[f"{each}-date-diff"] = (
        min_version_df["scikit-learn-date"] - min_version_df[f"{each}-date"]
    ).dt.days / 365

# %%
print(min_version_df[["scikit-learn", "python", "python-date-diff"]])
print(min_version_df[["scikit-learn", "python", "numpy", "numpy-date-diff"]])
print(min_version_df[["scikit-learn", "python", "scipy", "scipy-date-diff"]])
print(min_version_df[["scikit-learn", "python", "pandas", "scipy-date-diff"]])
print(min_version_df[["scikit-learn", "python", "joblib", "joblib-date-diff"]])
print(
    min_version_df[
        ["scikit-learn", "python", "threadpoolctl", "threadpoolctl-date-diff"]
    ]
)

# %%

# Plan with minor versions X.Y
plan = [
    {
        "scikit-learn": "1.7",
        "scikit-learn-date": pd.to_datetime("2025-06-01"),
        "python": "3.10",
        "numpy": "1.22.0",
        "scipy": "1.8.0",
        "pandas": "1.4.0",
        "joblib": "1.2.0",
        "threadpoolctl": "3.1.0",
    },
    {
        "scikit-learn": "1.8",
        "scikit-learn-date": pd.to_datetime("2025-12-01"),
        "python": "3.11",
        "numpy": "1.24.0",
        "scipy": "1.11.0",
        "pandas": "2.0.0",
        "joblib": "1.3.0",
        "threadpoolctl": "3.2.0",
    },
    {
        "scikit-learn": "1.9",
        "scikit-learn-date": pd.to_datetime("2026-06-01"),
        "python": "3.11.0",
        "numpy": "1.24.0",
        "scipy": "1.11.0",
        "pandas": "2.0.0",
        "joblib": "1.4.0",
        "threadpoolctl": "3.5.0",
    },
    {
        "scikit-learn": "1.10",
        "scikit-learn-date": pd.to_datetime("2026-12-01"),
        "python": "3.12",
        "numpy": "1.26.0",
        "scipy": "1.12.0",
        "pandas": "2.2.0",
        "joblib": "1.4.0",
        "threadpoolctl": "3.5.0",
    },
]

# Plan with bugfix versions X.Y.Z
# plan = [
#     {
#         "scikit-learn": "1.7",
#         "scikit-learn-date": pd.to_datetime("2025-06-01"),
#         "python": "3.10",
#         "numpy": "1.21.2",
#         "scipy": "1.8.0",
#         "pandas": "1.3.4",
#         "joblib": "1.2",
#         "threadpoolctl": "3.1",
#     },
#     {
#         "scikit-learn": "1.8",
#         "scikit-learn-date": pd.to_datetime("2025-12-01"),
#         "python": "3.11",
#         "numpy": "1.23.3",
#         "scipy": "1.10.1",
#         "pandas": "1.5.2",
#         "joblib": "1.3",
#         "threadpoolctl": "3.2",
#     },
#     {
#         "scikit-learn": "1.9",
#         "scikit-learn-date": pd.to_datetime("2026-06-01"),
#         "python": "3.11",
#         "numpy": "1.23.3",
#         "scipy": "1.10.1",
#         "pandas": "1.5.2",
#         "joblib": "1.4",
#         "threadpoolctl": "3.5",
#     },
#     {
#         "scikit-learn": "1.10",
#         "scikit-learn-date": pd.to_datetime("2026-12-01"),
#         "python": "3.12",
#         "numpy": "1.26.0",
#         "scipy": "1.11.3",
#         "pandas": "2.2.0",
#         "joblib": "1.4",
#         "threadpoolctl": "3.5",
#     },
# ]
plan = pd.DataFrame(plan)

for each in ["numpy", "scipy", "pandas", "joblib", "threadpoolctl"]:
    for ver in plan[each]:
        release_dates[each][ver] = get_release_time(each, ver)

for each in ["python", "numpy", "scipy", "pandas", "joblib", "threadpoolctl"]:
    diff_days = (
        plan["scikit-learn-date"] - plan[each].map(release_dates[each])
    ).dt.days.apply(
        lambda days: (
            pd.NaT
            if pd.isna(days)
            else f"{days // 365} years {round(days % 365 / 30)} months"
        )
    )
    plan[f"{each}-date-diff"] = diff_days


print(
    plan[
        ["scikit-learn", "scikit-learn-date", "python", "python-date-diff"]
    ].to_markdown(index=False)
)
print(
    plan[
        [
            "scikit-learn",
            "scikit-learn-date",
            "numpy",
            "numpy-date-diff",
            "scipy",
            "scipy-date-diff",
            "pandas",
            "pandas-date-diff",
        ]
    ].to_markdown(index=False)
)
print(
    plan[
        [
            "scikit-learn",
            "scikit-learn-date",
            "joblib",
            "joblib-date-diff",
            "threadpoolctl",
            "threadpoolctl-date-diff",
        ]
    ].to_markdown(index=False)
)