对于某些 class 解而非其他解，熵输出为 NaN

Question

我是运行 R 中的潜在 class 分析并使用熵函数。我想了解为什么在输出中，它会为较低的 nclasses 生成结果，然后为较高的 Nclasses 生成 NaN。

我是软件初学者！

此处供参考，是输出和代码：

> entropy<-function (p) sum(-p*log(p))
> error_prior <- entropy(France_2class$P) # Class proportions
> error_post <- mean(apply(France_2class$posterior, 1, entropy))
> R2_entropy <- (error_prior - error_post) / error_prior
> R2_entropy
[1] 0.8121263
> 
> entropy<-function (p) sum(-p*log(p))
> error_prior <- entropy(France_3class$P) # Class proportions
> error_post <- mean(apply(France_3class$posterior, 1, entropy))
> R2_entropy <- (error_prior - error_post) / error_prior
> R2_entropy
[1] 0.8139903
> 
> entropy<-function (p) sum(-p*log(p))
> error_prior <- entropy(France_4class$P) # Class proportions
> error_post <- mean(apply(France_4class$posterior, 1, entropy))
> R2_entropy <- (error_prior - error_post) / error_prior
> R2_entropy
[1] NaN
> 
> entropy<-function (p) sum(-p*log(p))
> error_prior <- entropy(France_5class$P) # Class proportions
> error_post <- mean(apply(France_5class$posterior, 1, entropy))
> R2_entropy <- (error_prior - error_post) / error_prior
> R2_entropy
[1] NaN
> 
> entropy<-function (p) sum(-p*log(p))
> error_prior <- entropy(France_6class$P) # Class proportions
> error_post <- mean(apply(France_6class$posterior, 1, entropy))
> R2_entropy <- (error_prior - error_post) / error_prior
> R2_entropy
[1] NaN

有人可以帮忙吗？谢谢

Answer 1

我猜问题出在entropy的定义上。更准确地说，如果0包含在p中，那么您将获得NaN，例如

> entropy(p1)
[1] 1.279854

> entropy(p2)
[1] NaN

> entropy(p3)
[1] 0.5004024

要修复它，您可以将 na.omit 添加到函数 entropy 中，如下所示

entropy<-function(p) sum(na.omit(-p*log(p)))

然后你可以看到

> entropy(p1)
[1] 1.279854

> entropy(p2)
[1] 0.5004024

> entropy(p3)
[1] 0.5004024

数据

p1 <- c(0.1,0.2,0.3,0.4)
p2 <- c(0,0.2,0.8)
p3 <- c(0.2,0.8)

对于某些 class 解而非其他解，熵输出为 NaN

Entropy output is NaN for some class solutions and not others

r

entropy