import numpy as np


def aic_statistic(dist, data, axis):
    """
    AIC-based goodness-of-fit statistic.

    AIC = 2k - 2ln(L)

    Lower AIC indicates better fit, but since goodness_of_fit()
    treats larger statistic values as worse fit, AIC works directly.
    """

    k = len(dist.args)
    log_likelihood = np.sum(dist.logpdf(data), axis=axis)
    aic = 2 * k - 2 * log_likelihood

    return aic

def bic_statistic(dist, data, axis):
    """
    BIC-based goodness-of-fit statistic.

    BIC = ln(n)k - 2ln(L)s

    Lower BIC indicates better fit, but since goodness_of_fit()
    treats larger statistic values as worse fit, BIC works directly.
    """

    n = data.size if axis is None else data.shape[axis]
    k = len(dist.args)
    log_likelihood = np.sum(dist.logpdf(data), axis=axis)
    bic = np.log(n) * k - 2 * log_likelihood

    return bic