Leave One Out Cross Validation Hyperparameter Tuning

Hi,

I have a dataset of motion data from different subjects which is used to do activity recognition.

To make u understand something like this

SUBJECTS DATA1 DATA2 DATA3 ........ ACTIVITY

1 32133 322332 212121 WALKING

1 54322 424242 342757 LYING

2 32133 322332 212121 WALKING

2 52322 424652 321217 RUNNING

Now my prediction is the activity column and i know there are specific function that can separate my dataset in training test and tuning hyperparameters ....

However while i'm building the classifier model i want to do a specific cross-validation in which every fold is made by all data from all subjects except one for testing. So i was thinking to use the built-in function of MATLAB and bayesopt to tune the hyperparameter of the model.

I have a data set of around 16600 observation and if i use the 'LeaveOut' option in cvpartion it gave me back the same number of folds.

Does anyone can give some advice on this ?

Every help or documentation is really appreciate.

P.S.

Observations from all subjects are not always the same number