makedist discrete uniform distribution

Hi all

I am trying to create a uniform discrete distribution, with 5 values (20% probability each), by using makedist command

However, I am bit confused with the boundaries.

My values I want to be : 20,25,30,35,40

Can you please help?

thanks

Nikolas

Answer 1

[Jeff Miller]

It appears that MATLAB will only give you random integers 1..K for any K you want, so you will have to convert those to your desired values, e.g.:

myrandoms = 15 + randi(5,100,1)*5  % A vector of 100 random numbers from your set

Answer 2

[Image Analyst]

This probably isn't what you want, but the code below will give you those 5 numbers, occurring at random locations in an array. And the probability will definitely be exactly 20% in each bin. Exactly 20% - no variation in the count because we're not using randi(). However the values are not random - they are exactly what you said - you are not going to get a count of 832 for 25, 803 for 30, 799 for 35, etc.. You will get the same count for all of them. You can specify approximately 20% in each bin (this is probably what you want, in which case you're use randi like Jeff did) OR you can specify that you want EXACTLY 20% to be each number (what I did).

% Define the values.
specifiedValues = [20,25,30,35,40]
% Make copies
numCopies = 300; % However many values you want to generate.
distributionValues = repmat(specifiedValues, [1, numCopies]);
% OPTIONAL: Scramble up (randomize) the order.
sortingOrder = randperm(length(distributionValues));
distributionValues = distributionValues(sortingOrder);
% Prove that the distribution is exactly 100% uniform (same count in each bin).
histogram(distributionValues);
grid on;
xticks(specifiedValues);
xlabel('Value');
ylabel('Count');

Answer 3

[Jonathan Sewell]

A function that achieves the random sampling effect you want is randsample().

To get one random sample from your specified population, use this.

randsample([20 25 30 35 40], 1)

To get ten random samples, use this.

randsample([20 25 30 35 40], 10, 1)

I do not know how to get this effect out of a call to makedist(). However, you can create your own distribution object that behaves like objects returned by makedist(). This would involve creating a class. It will need to be a subclass of ProbabilityDistribution, and probably a subclass of UnivariateDistribution or TruncatableDistribution as well. These classes are part of the shared stats library that makedist uses.