Best way to split data into random partitions?

I am new to Matlab and still a student. For an assignment, I am stuck on this part.

Create 5 random partitions of the data, splitting each of the classes into 60% training and 40% testing.

I have two classes, Class One and Class Two.

I was reading this https://www.mathworks.com/matlabcentral/answers/377839-split-training-data-and-testing-data

How would I be able to do this?

classOne and classTwo is 10000x2 double histogram

Answer 1

[Cris LaPierre]

I would use the dividerand function in the Deep Learning Toolbox.

For example

[trainInd,valInd,testInd] = dividerand(3000,0.6,0.2,0.2);

Just set the validation percentage to 0 if you don't need it.

[G. Nardi]

Can you tell me what the values mean in your example? 30000, .6 (I am assuming is the training and), .2,.2)

[Cris LaPierre]

I assume you've looked at the documentation page, so I won't duplicate info from there.

This function returns indices. Indices can then be used to select data for training, validation or testing.

From what you mention, let's say ClassOne is 10000x2. I want 60% (6000 rows) for training and 40% (4000 rows) for testing. I'd do this

[trainInd,~,testInd] = dividerand(length(ClassOne),0.6,0,0.4);

Now when you need your training data, you would use

ClassOne(trainInd,:)

and when you need you testing data, you would use

ClassOne(testInd,:)

dividerand creates groups randomly, so you would get different groups each time you called it.

[G. Nardi]

i don't have that toolbox.

[David Goodmanson]

HI Masaki,

Neither do I, but you can use somthing more basic like RandInd = randperm(n), which creates a vector containing a random arrangement of the numbers 1:n. Then you can take the first 60% (or whatever) of RandInd to be TrainInd, etc. and proceed from there.

[G. Nardi]

thanks! i appreciate it.