Comparing translation techniques

[dorchard]

We want to evaluate possible translation techniques

1. Gen+AI (+ adding in Peter Isaac's code into the prompt)
2. smop

We are going to use cpdFmax2pCp3.m as a test case; @j-emberton is making some tests so it is ready for that.

[dorchard]

@isaacaka to pick up writing the tests

[tztsai]

The smop translator is being implemented in PR #17. This PR also parametrizes the matlab_engine fixture (and renames it to ustar_cp) so that it can generate either the matlab engine instance, or the module containing the translated python code.

[dorchard]

Thanks. We now have a candidate set of tests for cpdFmax2pCp3.m here: (#51).

TODO:

• @tztsai to run these tests on the SMOP-translated cpdFmax2pCp3.py and let us know the results.
• @j-emberton make a branch with gen-AI translated cpdFmax2pCp and similarly to the above, run the tests and let us know the results.

In both cases the structure/scaffolding of the tests will need tweaking slightly to get it to address the Python version of the code rather than the matlab; I think do what works for now so that we can get the info we need to do the comparison of the two.

[tztsai]

I have translated the code to python and currently it can pass 16 out of 28 tests.

[dorchard]

In the translation phase, we've used a mixture of techniques. Sometimes the smop generated code works (sometimes it does not), other times some digging is needed with the help of an LLM. Sometimes we've been able to use Peter's code. So the mixed mode seems to be a good approach.