Hi Nils! On 6 Sep., 19:44, Nils Bruin <nbr...@sfu.ca> wrote: > If your tests depend on the choice of random numbers in a non- > deterministic algorithm, they are going to be very fragile.
Sure. But the funtionality "determine a Sylow subgroup", "find an isomorphism between two finite groups" etc. is essential to the computations. This is currently only available (vie GAP) by randomized algorithms. So, I don't think I can easily change it, and have to somehow cope with the instabilities (what I currently try to do). > One possible way to handle this is by rewriting your tests in such a > way that you test the properties that matter (and should not depend on > some underlying implementation), ... ... yes, such as testing against the Hilbert-Poincaré series and other invariants. Sure. Probably I should use those kind of tests more often. > However, when you constructed > the example, you probably found your element with certain properties > by doing some computations and tests in the representation that sage > gave you. You could make those computations part of your test. This is often done: There are many "high-level" tests that test against theorems. E.g., the program computes a homogeneous system of parameters, and the test then demonstrates (as an example for the user) how to conclude from properties of the parameters that the ring presentation is complete. In these cases I often also show how the parameters look like, but it would not harm to mark this part of the test as "random". More difficult are the "easy" tests: Testing __mul__, __cmp__ etc, because here one has no theorems to test against. > Resetting random seeds is useful for reproducing errors. Less so for > reproducing test results. Good point. Cheers, Simon -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
