Naive Bayes Multiclass error : singular matrix

[sakuranbokis]

Hi

I've started learning how to use PlantCV and I got stuck at Naive Bayes Multiclass classification.

I am now following the example - classifying disease area of leaves with four classes

and when I execute the code it gives me error message : LinAlgError : singular matrix

(Tested both on Win7 and Ubuntu 16)

and if I have only two classes (like plant and background) it works well.

What should I do? Thanks for reading.

[nfahlgren]

Hi @sakuranbokis, the error is occurring during kernel density estimation. I can think of two potential causes. 1) In at least one class there are not enough sample points. There’s no firm number because there’s a relationship to how diverse the class colors are, but I think ~100 works as a minimum most of the time. 2) One of the classes doesn’t have enough variation. Sometimes the background class is very clean (e.g. all black or white) and the kernel density algorithm fails because the distribution of colors in the class is too uniform. Do either of these sound possible for your case?

[lydiakienbaum]

As @nfahlgren has explained, the class colors need to be more diverse. I could solve the issue by removing the duplicates from the script.