utils.py 文件源码

def fast_optimize(endog, exog, n_obs=0, n_vars=0, max_iter=10000, tolerance=1e-10):
    """
    A convenience function for the Newton-Raphson method to evaluate a logistic model.
    :param endog: An Nx1 vector of endogenous predictions
    :param exog: An NxK vector of exogenous predictors
    :param n_obs: The number of observations N
    :param n_vars: The number of exogenous predictors K
    :param max_iter: The maximum number of iterations
    :param tolerance: Margin of error for convergence
    :return: The error-minimizing parameters for the model.
    """
    iterations = 0
    oldparams = np.inf
    newparams = np.repeat(0, n_vars)
    while iterations < max_iter and np.any(np.abs(newparams - oldparams) > tolerance):
        oldparams = newparams
        try:
            H = logit_hessian(exog, oldparams, n_obs)
            newparams = oldparams - dot(inv(H), logit_score(endog, exog, oldparams, n_obs))
        except LinAlgError:
            raise LinAlgError
        iterations += 1
    return newparams