assert tlf.z >= tlb.z, (tlf.z,trf.z)
AssertionError: (0.5, 0.5)
As can be seen, I'm suffering from precision problems. How can I rephrase the assert so it does pass for close enough values (how big should the fudge-factor be?) and then fix the rhs should it actually be smaller than the lhs so that it becomes strictly equal?
Try this:
EPSILON = 10 ** -12
assert tlf.z >= tlb.z - EPSILON, (tlf.z,trf.z)
tlf.z = max(tlf.z,tlb.z)
Essentially, you have to define the tolerance you're willing to have for "greater than or equal", and account for it.
What value to pick for EPSILON is a difficult question. It depends on the source of your error, and the number of calculations between that source and the comparison. If there are few calculations, a smaller value for EPSILON is a good bet. I would try the example, and adjust if you still find problems.
assert tlf.z >= tlb.z - EPSILON. Suppose the two values really are equal; the "+ EPSILON" will make the assertion fail.I don't think the answer given by willm1 is correct.
Let me explain.
The following expression:
tlf.z >= tlb.z - EPSILON
is equivalent to:
(tlf.z > tlb.z - EPSILON) or (tlf.z == tlb.z - EPSILON)
If tlf.z > tlb.z is true, even by a difference smaller than EPSILON, then tlf.z > tlb.z - EPSILON will also be true. No matter what the value of EPSILON. Instead, the correct form is:
tlf.z > tlb.z + EPSILON
As for the second expression, tlf.z == tlb.z - EPSILON, it will only be evaluated as true if tlf.z and tlb.z differ exactly by EPSILON, which is not what we want. Instead, we want the difference between them to be less than EPSILON:
abs(tlf.z - tlb.z) <= EPSILON
Concluding, tlf.z >= tlb.z - EPSILON should be written as:
(tlf.z > tlb.z + EPSILON) || (abs(tlf.z - tlb.z) <= EPSILON)
Update:
I was looking at some code, and all of a sudden I noticed that (tlf.z > tlb.z + EPSILON) || (abs(tlf.z - tlb.z) <= EPSILON) is actually equivalent to tlf.z >= tlb.z - EPSILON.
When we're looking for the similarity, abs(tlf.z - tlb.z) <= EPSILON, we want tlb.z to be in the following gray area:
When we're looking for tlf.z > tlb.z + EPSILON:
Hence, we're really looking for:
Which is the same as tlf.z + EPSILON >= tlb.z (equivalent to tlf.z >= tlb.z - EPSILON).
In that case, willm1 was actually right. Sorry :)
math.fabs(tlf.z-tlb.z) < 0.000001
>= I'd just subtract half the fudge factor from the rhs