RuntimeWarning：在log（stats.boxcox）中遇到除以零

Question

当我试图运行RuntimeWarning: divide by zero encountered in log时，我得到stats.boxcox(list)错误。

该列表不包含零值。我已经提到了其他一些帖子，说我们可以使用np.errstate()忽略错误。

但是，就我而言，我不认为这是正确的，因为stats.boxcox()的价值回报毫无意义（所有不变，见下文）。

array([ 0.11472901,  0.11472901,  0.11472901,  0.11472901,  0.11472901,
    0.11472901,  0.11472901,  0.11472901,  0.11472901,  0.11472901,
    0.11472901,  0.11472901,  0.11472901,  0.11472901,  0.11472901,
    0.11472901,  0.11472901,  ... ... ]

错误的详细信息如下：

/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/scipy/stats/morestats.py:591: RuntimeWarning: divide by zero encountered in log
llf -= N / 2.0 * np.log(np.sum((y - y_mean)**2. / N, axis=0))
/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/scipy/optimize/optimize.py:2063: RuntimeWarning: invalid value encountered in double_scalars
w = xb - ((xb - xc) * tmp2 - (xb - xa) * tmp1) / denomenter code here

Answer 1

我跟着这个Github线程的建议并为我工作：

https://github.com/scipy/scipy/issues/6873

建议是使用stats.boxcox_normmax(array)来获得理想的lambda，然后进行boxcox转换。

boxcox_lambda = boxcox_normmax(array, brack=(-1.9, 2.0),  method='mle')
transformed_array = boxcox(array, boxcox_lambda)