PsiMarginal.py 文件源码

def pf(parameters, psyfun='cGauss'):
    """Generate conditional probabilities from psychometric function.

    Arguments
    ---------
        parameters: ndarray (float64) containing parameters as columns
            mu   : threshold

            sigma    : slope

            gamma   : guessing rate (optional), default is 0.2

            lambda  : lapse rate (optional), default is 0.04

            x       : stimulus intensity

        psyfun  : type of psychometric function.
                'cGauss' cumulative Gaussian

                'Gumbel' Gumbel, aka log Weibull

    Returns
    -------
    1D-array of conditional probabilities p(response | mu,sigma,gamma,lambda,x)
    """

    # Unpack parameters
    if np.size(parameters, 1) == 5:
        [mu, sigma, gamma, llambda, x] = np.transpose(parameters)
    elif np.size(parameters, 1) == 4:
        [mu, sigma, llambda, x] = np.transpose(parameters)
        gamma = llambda
    elif np.size(parameters, 1) == 3:
        [mu, sigma, x] = np.transpose(parameters)
        gamma = 0.2
        llambda = 0.04
    else:  # insufficient number of parameters will give a flat line
        psyfun = None
        gamma = 0.2
        llambda = 0.04
    # Psychometric function
    ones = np.ones(np.shape(mu))
    if psyfun == 'cGauss':
        # F(x; mu, sigma) = Normcdf(mu, sigma) = 1/2 * erfc(-sigma * (x-mu) /sqrt(2))
        z = np.divide(np.subtract(x, mu), sigma)
        p = 0.5 * erfc(-z / np.sqrt(2))
    elif psyfun == 'Gumbel':
        # F(x; mu, sigma) = 1 - exp(-10^(sigma(x-mu)))
        p = ones - np.exp(-np.power((np.multiply(ones, 10.0)), (np.multiply(sigma, (np.subtract(x, mu))))))
    elif psyfun == 'Weibull':
        # F(x; mu, sigma)
        p = 1 - np.exp(-(np.divide(x, mu)) ** sigma)
    else:
        # flat line if no psychometric function is specified
        p = np.ones(np.shape(mu))
    y = gamma + np.multiply((ones - gamma - llambda), p)
    return y