Open elmisback opened 4 days ago
(Artem's paraphrased response below)
For Linear benchmarks, their inputs can be reduced to [0,1e20] and it may work
Link to Example: https://github.com/AYadrov/FPGuard/blob/master/Benchmarks/nonline-interpolation/SAT/interpol.txt Photo of Example:
Or, probably, a better idea is to go through FPTaylor's examples, like this one:
You can rewrite it to FPCore and run in Herbie
Can check out other examples on their website: https://monadius.github.io/FPTaylorJS/#/
I think that you can neglect Bounds and focus more on the absolute error, since it is more important. FPTaylor, for some reasons, has infinite bounds even for simple programs
Use FPTaylor paper example!
Please put Artem's response here