How does confusion matrix table work on 60 class or more

i wont do onfusion matrix for 60 class or more like this picture To find out TP,TN,FP,FN

[Walter Roberson]

confusionchart perhaps

[Abdullah]

It works on a few class

i have 60 class or more

[Walter Roberson]

confusionchart can handle those sizes.

confusionchart(randi(20,60,60))

[Abdullah]

Is it possible to make it look like this or no?

[Walter Roberson]

https://www.mathworks.com/help/deeplearning/ref/plotconfusion.html#mw_6e69e9bc-83d4-415a-b1af-471779286b0d perhaps

[Abdullah]

THANK YOU IT'S THE SAME NOT WORK

[Walter Roberson]

The example at that link shows 10 classes being plotted in the format you asked for. What happens when you do your 60 classes?

[Abdullah]

[Walter Roberson]

@Abdullah

You are using confusionchart() rather than the plotconfusion() that I posted the more recent link to.

[Abdullah]

[Walter Roberson]

That looks to me as if it is working. Just enlarge your plot area.

You want to be able to have outputs such as '46.2%' . You can't really read anything less than 8 point font, which is going to need 26 pixels wide for that character string (and more space would be preferred.) That would take at least

26*60

ans = 1560

1560 pixels wide, plus some more for the ticks and the axes labels; those class labels would take at least 32 pixels wide. So you are probably looking at around 1650 or more pixels wide to be readable.

[Abdullah]

its 60 classes Each of them has 20 elements

[Walter Roberson]

Yes?

Your sample plot has a '51.6%' in the middle of it. After having reduced down to 8 pixel font instead of the default 10 pixel font, the string '51.6%' takes 26 pixels width. You have 60 classes, so you need to allow space for (60 copies of 26 pixels) = 1560 pixels wide for the percentages drawn in the boxes. You have to allow 32 pixels for the '200170' class label. You need to allow 8 pixels for the sideways 'Output Class' label. That is a minimum of 1600 pixels width you need in order to be able to draw that chart -- probably more. If your window is not at least that wide then you will get overlap of characters.

[Abdullah]

It's not enough for thes big dataset

In addition to casia v4 which contains 1000 class

[Walter Roberson]

Okay, suppose that you are dealing with a chart with 1000 classes. And suppose that you have an Apple Retina display, 2880 x 1800 pixels. Then at best you can allocate each class 2.88 pixels of width. You wish to display 3 numeric characters plus a symbol, (such as 51.6%) so you can afford to allocate 2.88/4 = 0.72 pixels per character at best.

Are you able to read characters that are 0.72 pixels wide?

What is your plan to be able to display 1000 classes in a single chart? How many pixels per box were you thinking of allocating? Do you have a display that is 1000 times that number wide?

[Abdullah]

I just want to have a table that contains TP,TN,FP,FN like this picture

[Walter Roberson]

@Abdullah

Yes, and you want to have that for 1000 classes. That table is 2 classes plus a marginal value, and is (2+1) boxes wide by (2+1) boxes high. For 1000 classes you would have (1000+1) boxes wide by (1000+1) boxes high. Each box needs at least 28 pixels wide if an 8 point font is used, and the class label takes 32 pixels, so you need a display that is at least

(1000+1)*28 + 32

ans = 28060

over 28 thousand pixels wide.

Unless, that is, you want MATLAB to generate

1000 * (1000-1)/2

ans = 499500

close to half a million different 3 x 3 graphs, each comparing exactly two classes against each other ? Like one for class 17 vs 83, one for 17 vs 84, one for 83 vs 717, and so on ?? You might be able to fit 3 of those comfortably across, so it would only take about

200 * (499500/3)  %pixels high times graphs required divided by graphs wide

ans = 33300000

33 million or so pixels high.

[Abdullah]

This is impossible.

thank you very much sir

[Walter Roberson]

@Abdullah

MATLAB can generate the graph, and you can see the details if you zoom in enough, probably about 28:1

[Abdullah]

yes, you're right