Feature #565
FloatApprox for TwinFloat values?
Description
Does it make sense to have a fn like FloatApprox for twin-float values?
Currently the only way to get a "real value" out of a twin-float is to apply IsRational (or IsInteger); this is very limited.
Related issues
History
#1
Updated by John Abbott almost 10 years ago
Updated by - Status changed from New to In Progress
- Assignee set to John Abbott
- % Done changed from 0 to 20
- Estimated time changed from 40.00 h to 20.00 h
After further thought it seemed to make more sense to output the twin-float value as a MantExp2 structure, which better allows the actual accuracy to be represented. It is then simple to convert the MantExp2 structure into a rational if so desired.
I have a first prototype impl; it worked on a couple of trivial tests. Needs a lot more work though.
#2
Updated by John Abbott almost 8 years ago
- Related to Slug #907: ApproxSolve very slow on this example added
#3
Updated by John Abbott almost 8 years ago
My argument that creating a MantExp2 structure makes evident the precision is correct, but there are several occasions when the precision is of secondary importance (after all RingTwinFloat guarantees that it is better than a certain bound), and it would be helpful to have a function which produces directly a BigRat.
I am undecided exactly what the fn should be called. I observe that there is a fn called SimplestBinaryRatBetween, so the name should probably contain BinaryRat.
#4
Updated by John Abbott almost 8 years ago
How about the name BinaryRatApprox?
Mmm, well, that is exactly what FloatApprox does in other cases, but it's called FloatApprox.