No: tests don't change, especially iteratively. If you're editing and re-running your tests in a loop (rather than editing and re-running the code in a loop against the tests), then your tests aren't doing anything related to "testing", because you're changing their pass/fail criteria with each edit. To do so is just to abuse a test suite+test runner as a REPL.
On the other hand, once you know what you're doing, the expectation can be codified into a test (dead code) to assert that the live code maintains an equivalent property with no regressions. A "finished" IPython notebook can indeed be replaced with a regression test to ensure the behavior remains the way you previously determined it to be via experiment. (Though note that this is subtly different from functional testing: you're not asserting any preconceived notion of how an API "should" respond; you're just asserting that the API seems to hold to a certain contract—the one you discovered from your experimentation—and that it's a regression if it then goes against any of the parts of the discovered contract that you had come to rely upon.)
On the other hand, once you know what you're doing, the expectation can be codified into a test (dead code) to assert that the live code maintains an equivalent property with no regressions. A "finished" IPython notebook can indeed be replaced with a regression test to ensure the behavior remains the way you previously determined it to be via experiment. (Though note that this is subtly different from functional testing: you're not asserting any preconceived notion of how an API "should" respond; you're just asserting that the API seems to hold to a certain contract—the one you discovered from your experimentation—and that it's a regression if it then goes against any of the parts of the discovered contract that you had come to rely upon.)