member index bug

Replies to This Discussion

Permalink Reply by David Rutten on January 24, 2016 at 3:00pm

Just because two numbers look the same, doesn't mean they are the same. Numbers are formatted to a limited number of decimal places (otherwise they are always really hard to read). If you format them without loss of data, you can see they are not the same points after all:

Permalink Reply by Rabbit on January 24, 2016 at 3:10pm

Hi David,

You're right, they're not the same number, but what is strange is that they should be.

Those numbers are the coordinates of points,which should be the same point.

There's 2 curves, the region intersection is intersecting:

curve A with itself and with curve B (two resulting area centers),

and it's also intersecting

curve B with itself and with curve A (other two resulting area centers).

how can A-B be different than B-A?

maybe I'm missing something in my logical way of putting it... it's driving me crazy haha

Permalink Reply by David Rutten on January 24, 2016 at 4:41pm

Floating point arithmetic is not exact. Every time you perform some calculation you will get some errors in the least significant digits, and these errors aren't necessarily symmetric. The more calculations you perform, the worse the errors get. Especially subtraction is prone to a particularly severe type of error called catastrophic cancellation.

You don't have to understand the details of this, you just need to remember that when comparing floating point numbers, you almost never want to rely on exact equality. You must always pick the right tolerances for the problem at hand and treat values as equal if they fall within the same tolerance.

Permalink Reply by Andrew Heumann on January 24, 2016 at 8:57pm

I feel like this "problem" comes up enough that it almost warrants a voronoi-style user warning message on first use.