datasets

pycatdap.datasets

Sample datasets for CATDAP analysis.

Bundled datasets from the R catdap package for testing and examples.

load_titanic

load_titanic() -> pd.DataFrame

Load the Titanic dataset (2201 observations, 4 categorical variables).

Long-form expansion of R datasets::Titanic (4-way contingency table of Class × Sex × Age × Survived). Each cell of the original table is expanded to its frequency in row-level form.

Variables

Class : str ('1^st', '2^nd', '3^rd', 'Crew') Passenger or crew class. Sex : str ('Male', 'Female') Passenger sex. Age : str ('Adult', 'Child') Passenger age category. Survived : str ('Yes', 'No') Whether the passenger survived.

Returns:

Type	Description
DataFrame	2201 rows, 4 columns. All columns are categorical strings.

References

R Core Team (R datasets::Titanic): contingency table of survival on the Titanic. Public domain.

Examples:

>>> from pycatdap.datasets import load_titanic
>>> df = load_titanic()
>>> df.shape
(2201, 4)
>>> int(df["Survived"].value_counts()["Yes"])
711

Source code in src/pycatdap/datasets.py

def load_titanic() -> pd.DataFrame:
    """Load the Titanic dataset (2201 observations, 4 categorical variables).

    Long-form expansion of R ``datasets::Titanic`` (4-way contingency table
    of Class × Sex × Age × Survived). Each cell of the original table is
    expanded to its frequency in row-level form.

    Variables
    ---------
    Class : str ('1st', '2nd', '3rd', 'Crew')
        Passenger or crew class.
    Sex : str ('Male', 'Female')
        Passenger sex.
    Age : str ('Adult', 'Child')
        Passenger age category.
    Survived : str ('Yes', 'No')
        Whether the passenger survived.

    Returns
    -------
    DataFrame
        2201 rows, 4 columns. All columns are categorical strings.

    References
    ----------
    R Core Team (R `datasets::Titanic`): contingency table of survival on
    the Titanic. Public domain.

    Examples
    --------
    >>> from pycatdap.datasets import load_titanic
    >>> df = load_titanic()
    >>> df.shape
    (2201, 4)
    >>> int(df["Survived"].value_counts()["Yes"])
    711
    """
    path = _data_path("titanic.csv")
    return pd.read_csv(path)

load_iris

load_iris() -> pd.DataFrame

Load Fisher's iris dataset (150 observations, 5 variables).

Classic dataset with three iris species and four continuous morphological measurements. Useful for demonstrating CATDAP-02 with continuous variable pooling.

Variables

Sepal.Length : float Sepal length in cm. Sepal.Width : float Sepal width in cm. Petal.Length : float Petal length in cm. Petal.Width : float Petal width in cm. Species : str ('setosa', 'versicolor', 'virginica') Iris species (response variable).

Returns:

Type	Description
DataFrame	150 rows, 5 columns. Continuous variables for the 4 features and a categorical Species column.

References

Fisher, R. A. (1936). The use of multiple measurements in taxonomic problems. Public domain.

Examples:

>>> from pycatdap.datasets import load_iris
>>> df = load_iris()
>>> df.shape
(150, 5)
>>> sorted(df["Species"].unique().tolist())
['setosa', 'versicolor', 'virginica']

Source code in src/pycatdap/datasets.py

def load_iris() -> pd.DataFrame:
    """Load Fisher's iris dataset (150 observations, 5 variables).

    Classic dataset with three iris species and four continuous
    morphological measurements. Useful for demonstrating CATDAP-02 with
    continuous variable pooling.

    Variables
    ---------
    Sepal.Length : float
        Sepal length in cm.
    Sepal.Width : float
        Sepal width in cm.
    Petal.Length : float
        Petal length in cm.
    Petal.Width : float
        Petal width in cm.
    Species : str ('setosa', 'versicolor', 'virginica')
        Iris species (response variable).

    Returns
    -------
    DataFrame
        150 rows, 5 columns. Continuous variables for the 4 features
        and a categorical Species column.

    References
    ----------
    Fisher, R. A. (1936). The use of multiple measurements in taxonomic
    problems. Public domain.

    Examples
    --------
    >>> from pycatdap.datasets import load_iris
    >>> df = load_iris()
    >>> df.shape
    (150, 5)
    >>> sorted(df["Species"].unique().tolist())
    ['setosa', 'versicolor', 'virginica']
    """
    path = _data_path("iris.csv")
    return pd.read_csv(path)

load_german_credit

load_german_credit() -> pd.DataFrame

Load the Statlog German Credit dataset (1000 observations, 21 variables).

Binary classification benchmark used across the fairness / ML error analysis literature. Each row represents a credit applicant with mixed categorical and continuous attributes; the class column is the label ("good" / "bad" credit risk).

Variables

checking_status, savings_status, employment, ... : str Categorical attributes describing the applicant's banking, employment, housing, and demographic profile. duration, credit_amount, age, installment_commitment, ... : int Continuous attributes. class : str ('good', 'bad') Binary credit risk label.

Returns:

Type	Description
DataFrame	1000 rows, 21 columns. 17 categorical columns and 4 integer columns including the binary class label.

References

Hofmann, H. (1994). Statlog (German Credit Data) Data Set. UCI Machine Learning Repository. https://archive.ics.uci.edu/ml/datasets/Statlog+%28German+Credit+Data%29 Bundled here via OpenML id credit-g v1. Public domain.

Examples:

>>> from pycatdap.datasets import load_german_credit
>>> df = load_german_credit()
>>> df.shape
(1000, 21)
>>> df["class"].value_counts().to_dict()
{'good': 700, 'bad': 300}

Source code in src/pycatdap/datasets.py

def load_german_credit() -> pd.DataFrame:
    """Load the Statlog German Credit dataset (1000 observations, 21 variables).

    Binary classification benchmark used across the fairness / ML
    error analysis literature. Each row represents a credit applicant
    with mixed categorical and continuous attributes; the ``class``
    column is the label (``"good"`` / ``"bad"`` credit risk).

    Variables
    ---------
    checking_status, savings_status, employment, ... : str
        Categorical attributes describing the applicant's banking,
        employment, housing, and demographic profile.
    duration, credit_amount, age, installment_commitment, ... : int
        Continuous attributes.
    class : str ('good', 'bad')
        Binary credit risk label.

    Returns
    -------
    DataFrame
        1000 rows, 21 columns. 17 categorical columns and 4 integer
        columns including the binary ``class`` label.

    References
    ----------
    Hofmann, H. (1994). Statlog (German Credit Data) Data Set.
    UCI Machine Learning Repository.
    https://archive.ics.uci.edu/ml/datasets/Statlog+%28German+Credit+Data%29
    Bundled here via OpenML id ``credit-g`` v1. Public domain.

    Examples
    --------
    >>> from pycatdap.datasets import load_german_credit
    >>> df = load_german_credit()
    >>> df.shape
    (1000, 21)
    >>> df["class"].value_counts().to_dict()
    {'good': 700, 'bad': 300}
    """
    path = _data_path("german_credit.csv")
    return pd.read_csv(path)

load_heart_disease

load_heart_disease() -> pd.DataFrame

Load the Cleveland Heart Disease dataset (303 observations, 14 variables).

Binary classification benchmark from the UCI Machine Learning Repository (the processed Cleveland subset). Used widely in fairness and interpretability research.

Variables

age, trestbps, chol, thalach, oldpeak : float Continuous clinical measurements. sex, cp, fbs, restecg, exang, slope, ca, thal : float Encoded categorical clinical attributes. target : float (0, 1) Binary indicator of heart disease presence.

Returns:

Type	Description
DataFrame	303 rows, 14 columns. All numeric columns; target is the binary label.

References

Detrano, R., et al. (1989). Heart Disease Data Set. UCI Machine Learning Repository. https://archive.ics.uci.edu/ml/datasets/heart+disease Bundled here via OpenML id heart-disease v1. CC BY 4.0.

Examples:

>>> from pycatdap.datasets import load_heart_disease
>>> df = load_heart_disease()
>>> df.shape
(303, 14)

Source code in src/pycatdap/datasets.py

def load_heart_disease() -> pd.DataFrame:
    """Load the Cleveland Heart Disease dataset (303 observations, 14 variables).

    Binary classification benchmark from the UCI Machine Learning
    Repository (the processed Cleveland subset). Used widely in
    fairness and interpretability research.

    Variables
    ---------
    age, trestbps, chol, thalach, oldpeak : float
        Continuous clinical measurements.
    sex, cp, fbs, restecg, exang, slope, ca, thal : float
        Encoded categorical clinical attributes.
    target : float (0, 1)
        Binary indicator of heart disease presence.

    Returns
    -------
    DataFrame
        303 rows, 14 columns. All numeric columns; ``target`` is the
        binary label.

    References
    ----------
    Detrano, R., et al. (1989). Heart Disease Data Set.
    UCI Machine Learning Repository.
    https://archive.ics.uci.edu/ml/datasets/heart+disease
    Bundled here via OpenML id ``heart-disease`` v1. CC BY 4.0.

    Examples
    --------
    >>> from pycatdap.datasets import load_heart_disease
    >>> df = load_heart_disease()
    >>> df.shape
    (303, 14)
    """
    path = _data_path("heart_disease.csv")
    return pd.read_csv(path)

load_penguins

load_penguins() -> pd.DataFrame

Load the Palmer Penguins dataset (344 observations, 7 variables).

Three-class classification dataset based on observations of Adelie, Chinstrap, and Gentoo penguins from the Palmer Station Long Term Ecological Research program. Popular replacement for iris in classification tutorials.

Variables

species : str ('Adelie', 'Chinstrap', 'Gentoo') Three-class species label. island : str ('Torgersen', 'Biscoe', 'Dream') Island on which the penguin was observed. culmen_length_mm, culmen_depth_mm, flipper_length_mm, body_mass_g : float Morphological measurements (may contain NaN). sex : str ('MALE', 'FEMALE') Reported sex (may contain NaN).

Returns:

Type	Description
DataFrame	344 rows, 7 columns. 3 string columns and 4 float columns.

References

Horst, A. M., Hill, A. P., & Gorman, K. B. (2020). palmerpenguins. https://allisonhorst.github.io/palmerpenguins/ Bundled here via OpenML id penguins v1. CC0 1.0 Universal.

Examples:

>>> from pycatdap.datasets import load_penguins
>>> df = load_penguins()
>>> df.shape
(344, 7)
>>> sorted(df["species"].unique().tolist())
['Adelie', 'Chinstrap', 'Gentoo']

Source code in src/pycatdap/datasets.py

def load_penguins() -> pd.DataFrame:
    """Load the Palmer Penguins dataset (344 observations, 7 variables).

    Three-class classification dataset based on observations of
    Adelie, Chinstrap, and Gentoo penguins from the Palmer Station
    Long Term Ecological Research program. Popular replacement for
    iris in classification tutorials.

    Variables
    ---------
    species : str ('Adelie', 'Chinstrap', 'Gentoo')
        Three-class species label.
    island : str ('Torgersen', 'Biscoe', 'Dream')
        Island on which the penguin was observed.
    culmen_length_mm, culmen_depth_mm, flipper_length_mm, body_mass_g : float
        Morphological measurements (may contain NaN).
    sex : str ('MALE', 'FEMALE')
        Reported sex (may contain NaN).

    Returns
    -------
    DataFrame
        344 rows, 7 columns. 3 string columns and 4 float columns.

    References
    ----------
    Horst, A. M., Hill, A. P., & Gorman, K. B. (2020). palmerpenguins.
    https://allisonhorst.github.io/palmerpenguins/
    Bundled here via OpenML id ``penguins`` v1. CC0 1.0 Universal.

    Examples
    --------
    >>> from pycatdap.datasets import load_penguins
    >>> df = load_penguins()
    >>> df.shape
    (344, 7)
    >>> sorted(df["species"].unique().tolist())
    ['Adelie', 'Chinstrap', 'Gentoo']
    """
    path = _data_path("penguins.csv")
    return pd.read_csv(path)

fetch_california_housing

fetch_california_housing() -> pd.DataFrame

Fetch the California Housing regression dataset (H-0011 D4).

Thin wrapper over :func:sklearn.datasets.fetch_california_housing with as_frame=True. The target column MedHouseVal is concatenated into the returned DataFrame so it can be passed straight to :func:pycatdap.error_analysis as the regression response.

Returns:

Type	Description
DataFrame	20,640 rows × 9 columns (8 features + MedHouseVal target).

Raises:

Type	Description
ImportError	When scikit-learn is not installed. Install via pip install pycatdap[data].

Notes

• First call downloads the dataset to the sklearn cache (~/scikit_learn_data/ by default; override with SCIKIT_LEARN_DATA env var).
• Subsequent calls are offline.

Examples:

>>> import pycatdap
>>> df = pycatdap.datasets.fetch_california_housing()
>>> df.shape
(20640, 9)

Source code in src/pycatdap/datasets.py

def fetch_california_housing() -> pd.DataFrame:
    """Fetch the California Housing regression dataset (H-0011 D4).

    Thin wrapper over :func:`sklearn.datasets.fetch_california_housing`
    with ``as_frame=True``. The target column ``MedHouseVal`` is
    concatenated into the returned DataFrame so it can be passed
    straight to :func:`pycatdap.error_analysis` as the regression
    response.

    Returns
    -------
    DataFrame
        20,640 rows × 9 columns (8 features + ``MedHouseVal`` target).

    Raises
    ------
    ImportError
        When ``scikit-learn`` is not installed. Install via
        ``pip install pycatdap[data]``.

    Notes
    -----
    - First call downloads the dataset to the sklearn cache
      (``~/scikit_learn_data/`` by default; override with
      ``SCIKIT_LEARN_DATA`` env var).
    - Subsequent calls are offline.

    Examples
    --------
    >>> import pycatdap                                # doctest: +SKIP
    >>> df = pycatdap.datasets.fetch_california_housing()  # doctest: +SKIP
    >>> df.shape                                       # doctest: +SKIP
    (20640, 9)
    """
    try:
        from sklearn.datasets import fetch_california_housing as _fetch
    except ImportError as exc:
        msg = (
            "scikit-learn is required for fetch_california_housing. "
            "Install with: pip install 'pycatdap[data]'"
        )
        raise ImportError(msg) from exc

    bundle = _fetch(as_frame=True)
    df: pd.DataFrame = bundle.frame
    return df

fetch_adult_income

fetch_adult_income() -> pd.DataFrame

Fetch the Adult Income (Census) classification dataset (H-0011 D4).

Wrapper over :func:sklearn.datasets.fetch_openml with name="adult", version=2, as_frame=True. Binary target class (<=50K / >50K) is concatenated into the DataFrame.

Returns:

Type	Description
DataFrame	~48,842 rows × 15 columns (14 features + class target).

Raises:

Type	Description
ImportError	When scikit-learn is not installed. Install via pip install pycatdap[data].

Notes

• First call downloads from OpenML to the sklearn cache.
• Useful for demonstrating fairness-aware error analysis (the dataset has known protected-attribute associations on sex / race).

Examples:

>>> import pycatdap
>>> df = pycatdap.datasets.fetch_adult_income()
>>> "class" in df.columns
True

Source code in src/pycatdap/datasets.py

def fetch_adult_income() -> pd.DataFrame:
    """Fetch the Adult Income (Census) classification dataset (H-0011 D4).

    Wrapper over :func:`sklearn.datasets.fetch_openml` with
    ``name="adult"``, ``version=2``, ``as_frame=True``. Binary target
    ``class`` (``<=50K`` / ``>50K``) is concatenated into the
    DataFrame.

    Returns
    -------
    DataFrame
        ~48,842 rows × 15 columns (14 features + ``class`` target).

    Raises
    ------
    ImportError
        When ``scikit-learn`` is not installed. Install via
        ``pip install pycatdap[data]``.

    Notes
    -----
    - First call downloads from OpenML to the sklearn cache.
    - Useful for demonstrating fairness-aware error analysis (the
      dataset has known protected-attribute associations on
      ``sex`` / ``race``).

    Examples
    --------
    >>> import pycatdap                              # doctest: +SKIP
    >>> df = pycatdap.datasets.fetch_adult_income()  # doctest: +SKIP
    >>> "class" in df.columns                        # doctest: +SKIP
    True
    """
    try:
        from sklearn.datasets import fetch_openml
    except ImportError as exc:
        msg = (
            "scikit-learn is required for fetch_adult_income. "
            "Install with: pip install 'pycatdap[data]'"
        )
        raise ImportError(msg) from exc

    bundle = fetch_openml(name="adult", version=2, as_frame=True)
    df: pd.DataFrame = bundle.frame
    return df

fetch_compas

fetch_compas() -> pd.DataFrame

Fetch the COMPAS Two-Year Recidivism dataset (H-0011 D4).

Wrapper over :func:sklearn.datasets.fetch_openml with name="compas-two-years" (ProPublica re-release on OpenML). Target two_year_recid predicts whether a defendant was rearrested within two years.

.. warning:: COMPAS is a fairness-critical dataset originally analysed by ProPublica to expose racial bias in pre-trial risk scoring. It is included here as a demonstration dataset for fairness analysis, NOT as a model-evaluation benchmark for recidivism-prediction systems intended for real-world use. See ProPublica's original methodology and discussion: https://www.propublica.org/article/how-we-analyzed-the-compas-recidivism-algorithm.

Returns:

Type	Description
DataFrame	~5,278 rows × 14 columns (features + target).

Raises:

Type	Description
ImportError	When scikit-learn is not installed. Install via pip install pycatdap[data].

Examples:

>>> import pycatdap
>>> df = pycatdap.datasets.fetch_compas()

Source code in src/pycatdap/datasets.py

def fetch_compas() -> pd.DataFrame:
    """Fetch the COMPAS Two-Year Recidivism dataset (H-0011 D4).

    Wrapper over :func:`sklearn.datasets.fetch_openml` with
    ``name="compas-two-years"`` (ProPublica re-release on OpenML).
    Target ``two_year_recid`` predicts whether a defendant was rearrested
    within two years.

    .. warning::
        COMPAS is a **fairness-critical** dataset originally analysed
        by ProPublica to expose racial bias in pre-trial risk scoring.
        It is included here as a *demonstration dataset for fairness
        analysis*, NOT as a model-evaluation benchmark for
        recidivism-prediction systems intended for real-world use.
        See ProPublica's original methodology and discussion:
        <https://www.propublica.org/article/how-we-analyzed-the-compas-recidivism-algorithm>.

    Returns
    -------
    DataFrame
        ~5,278 rows × 14 columns (features + target).

    Raises
    ------
    ImportError
        When ``scikit-learn`` is not installed. Install via
        ``pip install pycatdap[data]``.

    Examples
    --------
    >>> import pycatdap                          # doctest: +SKIP
    >>> df = pycatdap.datasets.fetch_compas()    # doctest: +SKIP
    """
    try:
        from sklearn.datasets import fetch_openml
    except ImportError as exc:
        msg = (
            "scikit-learn is required for fetch_compas. "
            "Install with: pip install 'pycatdap[data]'"
        )
        raise ImportError(msg) from exc

    bundle = fetch_openml(name="compas-two-years", version=4, as_frame=True)
    df: pd.DataFrame = bundle.frame
    return df

fetch_wine_quality

fetch_wine_quality() -> pd.DataFrame

Fetch the UCI Wine Quality dataset (red + white combined, H-0017 D5).

Wrapper over :func:sklearn.datasets.fetch_openml that downloads the wine-quality-red (1,599 rows) and wine-quality-white (4,898 rows) datasets, tags each with a color column, and concatenates them into a single frame (6,497 rows). The target column is normalised to quality regardless of the OpenML attribute name.

Returns:

Type	Description
DataFrame	6,497 rows × 13 columns (11 physicochemical features + quality target + color).

Raises:

Type	Description
ImportError	When scikit-learn is not installed. Install via pip install pycatdap[data].

Notes

• First call downloads from OpenML to the sklearn cache.
• color (red / white) is a useful explanatory variable for AIC-based EDA and a natural grouping for slice discovery.

Examples:

>>> import pycatdap
>>> df = pycatdap.datasets.fetch_wine_quality()
>>> df.shape
(6497, 13)

Source code in src/pycatdap/datasets.py

def fetch_wine_quality() -> pd.DataFrame:
    """Fetch the UCI Wine Quality dataset (red + white combined, H-0017 D5).

    Wrapper over :func:`sklearn.datasets.fetch_openml` that downloads the
    ``wine-quality-red`` (1,599 rows) and ``wine-quality-white`` (4,898
    rows) datasets, tags each with a ``color`` column, and concatenates
    them into a single frame (6,497 rows). The target column is
    normalised to ``quality`` regardless of the OpenML attribute name.

    Returns
    -------
    DataFrame
        6,497 rows × 13 columns (11 physicochemical features +
        ``quality`` target + ``color``).

    Raises
    ------
    ImportError
        When ``scikit-learn`` is not installed. Install via
        ``pip install pycatdap[data]``.

    Notes
    -----
    - First call downloads from OpenML to the sklearn cache.
    - ``color`` (``red`` / ``white``) is a useful explanatory variable
      for AIC-based EDA and a natural grouping for slice discovery.

    Examples
    --------
    >>> import pycatdap                              # doctest: +SKIP
    >>> df = pycatdap.datasets.fetch_wine_quality()  # doctest: +SKIP
    >>> df.shape                                     # doctest: +SKIP
    (6497, 13)
    """
    try:
        from sklearn.datasets import fetch_openml
    except ImportError as exc:
        msg = (
            "scikit-learn is required for fetch_wine_quality. "
            "Install with: pip install 'pycatdap[data]'"
        )
        raise ImportError(msg) from exc

    frames = []
    for name, color in (("wine-quality-red", "red"), ("wine-quality-white", "white")):
        bundle = fetch_openml(name=name, version=1, as_frame=True)
        frame: pd.DataFrame = bundle.data.copy()
        frame["quality"] = bundle.target.to_numpy()
        frame["color"] = color
        frames.append(frame)
    return pd.concat(frames, ignore_index=True)

fetch_bank_marketing

fetch_bank_marketing() -> pd.DataFrame

Fetch the UCI Bank Marketing dataset (H-0017 D5).

Wrapper over :func:sklearn.datasets.fetch_openml with name="bank-marketing". The OpenML release exposes the 16 features as generic V1..V16; this loader renames them to the UCI names (age, job, marital, ...) and names the target y. If a future OpenML version already ships interpretable names, the rename is skipped.

Returns:

Type	Description
DataFrame	45,211 rows × 17 columns (16 features + y target).

Raises:

Type	Description
ImportError	When scikit-learn is not installed. Install via pip install pycatdap[data].

Notes

• First call downloads from OpenML to the sklearn cache.
• Classification target y (term-deposit subscription) makes this a good slice-discovery showcase.

Examples:

>>> import pycatdap
>>> df = pycatdap.datasets.fetch_bank_marketing()
>>> "age" in df.columns
True

Source code in src/pycatdap/datasets.py

def fetch_bank_marketing() -> pd.DataFrame:
    """Fetch the UCI Bank Marketing dataset (H-0017 D5).

    Wrapper over :func:`sklearn.datasets.fetch_openml` with
    ``name="bank-marketing"``. The OpenML release exposes the 16 features
    as generic ``V1``..``V16``; this loader renames them to the UCI names
    (``age``, ``job``, ``marital``, ...) and names the target ``y``. If a
    future OpenML version already ships interpretable names, the rename
    is skipped.

    Returns
    -------
    DataFrame
        45,211 rows × 17 columns (16 features + ``y`` target).

    Raises
    ------
    ImportError
        When ``scikit-learn`` is not installed. Install via
        ``pip install pycatdap[data]``.

    Notes
    -----
    - First call downloads from OpenML to the sklearn cache.
    - Classification target ``y`` (term-deposit subscription) makes this
      a good slice-discovery showcase.

    Examples
    --------
    >>> import pycatdap                                # doctest: +SKIP
    >>> df = pycatdap.datasets.fetch_bank_marketing()  # doctest: +SKIP
    >>> "age" in df.columns                            # doctest: +SKIP
    True
    """
    try:
        from sklearn.datasets import fetch_openml
    except ImportError as exc:
        msg = (
            "scikit-learn is required for fetch_bank_marketing. "
            "Install with: pip install 'pycatdap[data]'"
        )
        raise ImportError(msg) from exc

    bundle = fetch_openml(name="bank-marketing", version=1, as_frame=True)
    frame: pd.DataFrame = bundle.data.copy()
    # Rename the generic OpenML V1..V16 columns positionally to UCI names.
    # A future OpenML version that already ships interpretable names is left
    # untouched (silent no-op). If the names are still generic but the count
    # no longer matches 16, we cannot rename safely — warn instead of
    # silently returning ``V1..V16`` (#25 wants interpretable columns).
    columns_are_generic = all(
        str(c).startswith("V") and str(c)[1:].isdigit() for c in frame.columns
    )
    if columns_are_generic and len(frame.columns) == len(_BANK_MARKETING_COLUMNS):
        frame.columns = pd.Index(_BANK_MARKETING_COLUMNS)
    elif columns_are_generic:
        warnings.warn(
            "fetch_bank_marketing: OpenML returned generic column names "
            f"({list(frame.columns)[:3]}...) but the count "
            f"({len(frame.columns)}) does not match the expected "
            f"{len(_BANK_MARKETING_COLUMNS)} UCI features; returning generic "
            "names unchanged.",
            stacklevel=2,
        )
    frame["y"] = bundle.target.to_numpy()
    return frame

fetch_mushroom

fetch_mushroom() -> pd.DataFrame

Fetch the UCI Mushroom dataset (H-0017 D5).

Wrapper over :func:sklearn.datasets.fetch_openml with name="mushroom". All 22 features and the class target (e = edible / p = poisonous) are categorical, making this the canonical fully-categorical CATDAP demonstration dataset.

Returns:

Type	Description
DataFrame	8,124 rows × 23 columns (22 categorical features + class target).

Raises:

Type	Description
ImportError	When scikit-learn is not installed. Install via pip install pycatdap[data].

Notes

• First call downloads from OpenML to the sklearn cache.
• Every column is categorical, so catdap2 can run with the default (all-categorical) pooling.

Examples:

>>> import pycatdap
>>> df = pycatdap.datasets.fetch_mushroom()
>>> "class" in df.columns
True

Source code in src/pycatdap/datasets.py

def fetch_mushroom() -> pd.DataFrame:
    """Fetch the UCI Mushroom dataset (H-0017 D5).

    Wrapper over :func:`sklearn.datasets.fetch_openml` with
    ``name="mushroom"``. All 22 features and the ``class`` target
    (``e`` = edible / ``p`` = poisonous) are categorical, making this the
    canonical fully-categorical CATDAP demonstration dataset.

    Returns
    -------
    DataFrame
        8,124 rows × 23 columns (22 categorical features + ``class``
        target).

    Raises
    ------
    ImportError
        When ``scikit-learn`` is not installed. Install via
        ``pip install pycatdap[data]``.

    Notes
    -----
    - First call downloads from OpenML to the sklearn cache.
    - Every column is categorical, so ``catdap2`` can run with the
      default (all-categorical) pooling.

    Examples
    --------
    >>> import pycatdap                          # doctest: +SKIP
    >>> df = pycatdap.datasets.fetch_mushroom()  # doctest: +SKIP
    >>> "class" in df.columns                    # doctest: +SKIP
    True
    """
    try:
        from sklearn.datasets import fetch_openml
    except ImportError as exc:
        msg = (
            "scikit-learn is required for fetch_mushroom. "
            "Install with: pip install 'pycatdap[data]'"
        )
        raise ImportError(msg) from exc

    bundle = fetch_openml(name="mushroom", version=1, as_frame=True)
    df: pd.DataFrame = bundle.frame.copy()
    return df