Open Jbowman353 opened 5 years ago
@Shane101196 @btdow I think we should try to get some larger test inputs, as the ones we currently have are very small and the test_groebnertools functions complete very quickly, what do you guys think? I'm hoping we can just have a handful of decent test cases to compare f5b and a gpu-based f5b at the very least, and I don't think we should worry about being able to input custom systems., at least for the time being
We can adapt the benchmarking inputs from the GB package fairly easily (https://github.com/ederc/gb/tree/master/benchs) They're already in a readable format, just need to change ^ characters to **
@Shane101196 I'm thinking we should have one entry point file that runs a bunch of tests and outputs to a CSV. My thinking is the format, and also write whether or not the results were the same for every test as a precaution:
Test Name | F5B Runtime | GPU F5B Runtime | Buchberger Runtime |
---|---|---|---|
Kastura N | 5s | 2s | 12s |
@Jbowman353 Okay, I'll try and have a file that can output in this format + check for correctness this weekend. Pretty busy with midterms this week. Just as a side note: F5b produces a reduced groebner basis while buchberger does not with the current code so for checking if the results are the same, we will need to apply the reduced groebner function to buchberger as well, but it should work in theory.
Okay, yeah just ignore buchberger for now then, or if you want to throw it in that's fine, we'll just mention that
I've pushed some stuff for converting gb-package input files to python data structures and using them, not the prettiest but it seems to work okay @Shane101196 @btdow
Going to also assign Brad when I can on this one