How to fix ZeroDivisionError: float division by zero when calculating Gaussian probability

Question

我正在尝试计算高斯概率，但是当我运行通过我的数据时，我得到 ZeroDivisionError: float division by zero

代码如下：

def GaussianProbability(x, mean, std):
    exponent = math.exp(-(math.pow(x-mean,2)/(2*math.pow(std,2))))
    return (1 / (math.sqrt(2*math.pi) * std)) * exponent

以及完整的错误日志：

ZeroDivisionError                         Traceback (most recent call last)
<ipython-input-22-b3fd5204c6fa> in <module>
----> 1 precisions_PDZ1,MAP,recalls_PDZ1,fp_rates_PDZ1 = CV_results(PDZ1_graphs,PDZ1_targets,'undirected')

<ipython-input-14-044940745269> in CV_results(list_graphs, list_targets, network_type, multi)
      7     for part in tqdm(range(5)):
      8 
----> 9         probabilities_0,predictions = prediction(list_graphs[part], list_targets[part], network_type, multi)
     10 
     11         if recalls == [] and fp_rates == []:

<ipython-input-12-beeed13223ea> in prediction(graph, actual_edges, network_type, multi)
     11 
     12     for i in tqdm(range(len(X_test))):
---> 13         prob = ClassProbabilities(summaries, prob_class, X_test[i])
     14         if (prob[0]+prob[1])==0:
     15             # both probabilities are too low

<ipython-input-11-286b09a753e9> in ClassProbabilities(summaries, prob_class, x)
      8         for i in range(len(classSummaries)):  # for every attribute
      9             mean, std = classSummaries[i]
---> 10             probabilities[classValue] *= GaussianProbability(x[i], mean, std)
     11 
     12     return probabilities

<ipython-input-10-4a13cc5a69b2> in GaussianProbability(x, mean, std)
      1 def GaussianProbability(x, mean, std):
----> 2     exponent = math.exp(-(math.pow(x-mean,2)/(2*math.pow(std,2))))
      3     return (1 / (math.sqrt(2*math.pi) * std)) * exponent

ZeroDivisionError: float division by zero

Answer 1

当你的 standerdv 等于零时，你的函数需要有一个 else 语句

您可以查看此参考资料，它可能会有所帮助 What is the normal distribution when standard deviation is zero?

Answer 2

唯一可能出现这种错误的地方是(2*math.pow(std,2)) == 0。这意味着对于某些数据，当所有样本都具有完全相同的值时，您的 std == 0 就会发生这种情况。

你需要处理这种情况！

除非你想自己实现（不推荐）你可以使用scipy package:

import scipy
scipy.stats.norm(0, 1).pdf(0)

访问scipy docs

您还可以使用内置统计数据包 python 提供（更简单、更快速的计算）。看看here

How to fix ZeroDivisionError: float division by zero when calculating Gaussian probability

How to fix ZeroDivisionError: float division by zero when calculating Gaussian probability

python

normal-distribution

probability

gaussian