You test them the same way as you test short functions:
- Provide a set with specific input data and compare the results with known outputs
- Check if the output is very sensitive to a tiny variation of the inputs. If so, the function is an expensive random number generator.
- Call the function with irregulare values and check, if it throws error messages
- Check for NaN, +/-Inf, empty input, wrong file names, cell arrays instead of matrices etc.
Because it might be really hard to obtain some valid output data without using the original function, a validation of code usually requires to split the code to smaller functions, which can be tested independently. If the code is e.g. an integrator, you could compare the results with other integrators.
The base of unit-testing is to split the functions into units, whose output can be checked with controllable methods.