link to shining panda
... would be nice to have.
aside from helping to ensure code correctness (check results), it would also help to avoid simple NameErrors/AttributeErrors that can easily creep in when code has no complete test coverage - then those errors are often found by users in production.
related to this is the question whether doctests scale to such good coverage or whether separate tests would be better.
for separate tests, I'ld recommend using pytest, which is a nice framework (and you have to type less than with the usual unittest stuff).