search

NeuralEnsemble / PyNN

A Python package for simulator-independent specification of neuronal network models.
Other
276 stars 126 forks source link

Float comparisons can lead to erroneous exceptions #726

Closed chanokin closed 2 years ago

chanokin commented 3 years ago

When computing the shape of a Grid2D, the number of x units, nx, comes from a floating-point operation. Checking if it's equal to 0 will often return false (e.g. 1e-100 != 0) even if the value should be interpreted as 0. https://github.com/NeuralEnsemble/PyNN/blob/4feecd79a216e41ad65d6399c2cc5fcd9fa0a7a0/pyNN/space.py#L245-L252

I believe the comparison in line 248 should be changed to not math.isclose(n % nx, 0) with the precision of your choosing or just cast nx to an integer before the modulo operation is performed.

apdavison commented 2 years ago

I'm happy to make this change, but I'm struggling to find values for which the current implementation fails (so I can write a regression test). Did you encounter a case where the exception is raised incorrectly?

chanokin commented 2 years ago

I'm happy to make this change, but I'm struggling to find values for which the current implementation fails (so I can write a regression test). Did you encounter a case where the exception is raised incorrectly?

Hi, I think this was an extreme case where I one of my 2D populations ended up with a shape of [1, X] (or [X, 1] can't really remember). I'll try making a minimal script for this and let you know if I got it to behave unexpectedly.