Testing n_columns=0

[HenrikMettler]

So, weirdly n_columns = 0 works in the minimal example and in a small test in population it gives the expected sympy expression. However a lower level test fails when calling directly CartesianGraph(genome).to_sympy(). Any ideas why?
Also imo we should rather deprecate the n_columns=0 or n_rows=0 functionality, since they don't make much sense anyway.
Closes #283

[jakobj]

great, thanks for working on this! 🚀

i've added a few inline comments that should be addressed.

regarding your question: the low-level error was generated because the genome was not initialized. in also, i think we should not explicitly forbid n_columns=0 if our library does the right thing, why not?

[HenrikMettler]

Ok, i am a bit lost. Somehow the test doesn't pass on the python 3.7 testsuite, however I can't reproduce the error locally. Any ideas why?

[mschmidt87]

I also tried to reproduce the error locally without success. I also tried to google the error and found this link which might (or might not) be helpful: https://stackoverflow.com/questions/66640404/basic-loop-with-sympy-gives-me-this-error-message-symbol-object-is-not-subsc
Another, a bit brute-force option, is to do:

• remove python 3.7 temporarily to see whether 3.6 and 3.8 throw the same error in the CI
• add some debug output to the test and rerun the CI.
  Also, I did create a new conda env with python 3.7 and installed the library, so I basically followed the steps in the CI, but I didn't confirm that the configuration is really exactly identical with the CI environment. So, you could/should also do that with your local environment.

[jakobj]

have you tried rebasing on the most recent master? one of the recent PRs has removed the necessity to do sympy_expr[0] if only one expression is encoded in the genome, as it is automatically unpacked. i think this might be where the error comes from.

[HenrikMettler]

ah yes, that was it! my bad, thanks for suggesting

[jakobj]

looking good! 👍

and merging