MATLAB Grader. How to check vectroization in students solution?

Hi,

I would like to chck in MATLAB Grader if the students have used vectorization, eg. they write in their solution

x = (1:5).^(5:-1:1)

instead of

x = [1^5 2^4 3^3 4^2 5^1]

Is it possible?

[Martin Rudolph]

Hello,

Referring to:

https://www.mathworks.com/matlabcentral/answers/486153-writing-the-text-from-a-matlab-grader-student-solution-to-a-variable

you could try something like that (with sprintf values from the reference solution could also be used):

SolutionFile = fileread('solution.m');
if ~contains(SolutionFile,'(1:5).^(5:-1:1)')
    error('You did not vectorized the code as intended, please refer to the following example...')
end

or something less specific to allow alternative solutions:

if ~contains(regexprep(SolutionFile,'\s',''),'.^')
    error('You did not used the elementwise power operator')
end

or

MaxNumber = 4;
NumberOfPowerOperators = numel(strfind(SolutionFile,'^'));
if NumberOfPowerOperators>MaxNumber
    error('You used the power operator to often, it seems you did not vectorized your code properly')
end

Using splitlines and a loop you could also check each line for the amount of used operators. However, even if you write a complex test you cannot consider all possible solutions by the students. But a few lines of code may help you to identify strange solutions for a manual review.

Answer 1

[Cris LaPierre]

There is no built-in way to check that they used element-wise operators (.^).

You can prohibit the use of the keyword for, to avoid their solving it with a for loop.

My suggestion would be to assign the number to use randomly. That way, they can't hardcode their solution.

% provide the following as locked line in the student template
n = randi([4,10]);

% Student provides their solution underneath
x=(1:n).^(n:-1:1);

Because the reference and learner solutions share the same random number seed, the reference solution would just be

n = randi([4,10]);
x=(1:n).^(n:-1:1);

[Cris LaPierre]

Because this is graded automatically, the other thing you could do is make n so high that it would be painful to hardcode the answer. For example,

x=(1:1000).^(1000:-1:1)

The computer will still have no trouble grading it, but the likelihood someone is going to write it all out by hand decreases significantly.

[Walter Roberson]

Though that particular expression would generate a bunch of inf !

[Cris LaPierre]

True, but even those can still be graded by MATLAB Grader.

[BLP]

Thanks for all suggestions. I though abou using high n but the suggestion from Cris with n generated radomly is in even a better idea which I may use in ather problems e.g.

x = randi(10,[1 5])
y = randi(10,[1 5])

z = x.^y

I need urgently to convert my course in programming in MTALB into an online one due to coronavirus epidemy. Your suggestions showed me that I should be more creative, and do not simple copy the problems.

[Cris LaPierre]

Let me recommend our Teaching with MATLAB self-paced course. I'd suggest quickly browsing all the chapters, as they all focus on our resources for teaching online. Chapter 6 in particular covers MATLAB Grader.

Also, Grader comes with some pre-built problems. The Getting Started with MATLAB Grader collection was specifically designed to help you get started. The last page of the Teaching with MATLAB course contains a table that helps identify problems to use as a template for common requests (working with files, random numbers, etc). For example, look at the Coordinate transformations Navigating a robot problem to see how you can use random numbers to create variables for your students to use.

[Cris LaPierre]

Grader is essentially a variable checker. It is not a good solution for interactive/gui exercises. It should work well for your laboratories provided the code does not cause grader to time out (<1 minute to run and assess code).