Module ktrain.tabular.data

Functions

def tabular_from_csv(train_csv, label_columns=[], date_columns=[], val_csv=None, val_pct=0.1, index_col=None, is_regression=False, max_card=20, random_state=None)

Loads tabular data from CSV file

Expand source code

def tabular_from_csv(
    train_csv,
    label_columns=[],
    date_columns=[],
    val_csv=None,
    val_pct=0.1,
    index_col=None,
    is_regression=False,
    max_card=20,
    random_state=None,
):
    """
    ```
    Loads tabular data from CSV file
    ```
    """

    # read in dataset
    train_df = pd.read_csv(train_csv, index_col=index_col)
    val_df = None
    if val_csv is not None:
        val_df = pd.read_csv(val_csv, index_col=index_col)
    return tabular_from_df(
        train_df,
        label_columns=label_columns,
        date_columns=date_columns,
        val_df=val_df,
        val_pct=val_pct,
        is_regression=is_regression,
        max_card=max_card,
        random_state=random_state,
    )

def tabular_from_df(train_df, label_columns=[], date_columns=[], val_df=None, val_pct=0.1, is_regression=False, max_card=20, random_state=None, verbose=1)

Expand source code

def tabular_from_df(
    train_df,
    label_columns=[],
    date_columns=[],
    val_df=None,
    val_pct=0.1,
    is_regression=False,
    max_card=20,
    random_state=None,
    verbose=1,
):
    train_df = train_df.copy()

    # strip space from string columns and check supplied val_df
    train_type_dict = pp.clean_df(train_df, val_df=val_df, return_types=True)

    # check label_columns
    if label_columns is None or (
        isinstance(label_columns, (list, np.ndarray)) and len(label_columns) == 0
    ):
        raise ValueError("label_columns is required")
    if isinstance(label_columns, (list, np.ndarray)) and len(label_columns) == 1:
        label_columns = label_columns[0]

    # define original predictor_columns
    predictor_columns = [
        col for col in train_df.columns.values if col not in label_columns
    ]

    # create validation set
    if val_df is None:
        if val_pct:
            df = train_df.copy()
            prop = 1 - val_pct
            if random_state is not None:
                np.random.seed(42)
            msk = np.random.rand(len(df)) < prop
            train_df = df[msk]
            val_df = df[~msk]
    else:
        val_df = val_df.copy()

    procs = [pp.FillMissing, pp.Categorify, pp.Normalize]
    preproc = pp.TabularPreprocessor(
        predictor_columns,
        label_columns,
        date_columns=date_columns,
        is_regression=is_regression,
        procs=procs,
        max_card=max_card,
    )
    trn = preproc.preprocess_train(train_df, verbose=verbose)
    if verbose:
        integer_cats = []
        for col in preproc.cat_names:
            if train_type_dict.get(col, None) == "integer":
                integer_cats.append(col)
        if integer_cats:
            print(
                f"\nThe following integer column(s) are being treated as categorical variables:\n{integer_cats}\n"
                + "To treat any of these column(s) as numerical, cast the column to float in DataFrame or CSV\n and re-run tabular_from* function.\n"
            )
    val = None if val_df is None else preproc.preprocess_test(val_df, verbose=verbose)
    return (trn, val, preproc)