拟合自定义分布 scipy.stats 会溢出

Fitting custom distribution scipy.stats gives overflow

我正在尝试将广义误差分布拟合到我拥有的一些数据中。分布形式为

我尝试了以下实现

import numpy as np
import scipy.stats as st
from scipy.special import gamma

class ged(st.rv_continuous):

    def _pdf(self, x, mu, sigma, kappa):
        
        term1 = gamma(3*kappa)/gamma(kappa)
        
        exponent = (term1 * ((x - mu)/sigma)**2)**(1/(2*kappa))
        
        term2 = np.exp(-exponent)
        
        term3 = 2*sigma*gamma(kappa+1)
        
        fx = term1**0.5 * term2/term3

        return fx

ged_inst = ged(name='ged')
data = np.random.normal(size=1000)
ged_inst.fit(data, 0, 0.01, 1)

然而这给出了

OverflowError: (34, 'Numerical result out of range')

如何正确实施此分布?我试图适应真实数据(不是问题中生成的玩具正常数据)

如评论中所述,要使此功能正常工作,我需要覆盖默认的 _argcheck 函数。以下作品:

class ged(st.rv_continuous):

    def _pdf(self, x, mu, sigma, kappa):
        
        term1 = gamma(3*kappa)/gamma(kappa)
        
        exponent = (term1 * ((x - mu)/sigma)**2)**(1/(2*kappa))
        
        term2 = np.exp(-exponent)
        
        term3 = 2*sigma*gamma(kappa+1)
        
        fx = term1**0.5 * term2/term3

        return fx
    
    def _argcheck(self, mu, sigma, kappa):
        
        s = sigma > 0
        k = kappa < 1
        
        return s and k