Improve ulps plot: #417
Improve ulps plot: #417
Conversation
Use numeric_limits for precision comparison, plot out of bounds points at the clip boundary so we can still see that they're there.
|
@NAThompson : A quick question:
Other than that this fixes the case where PreciseReal is a multiprecision type. |
| // Got to think on what to do about it. | ||
| static_assert(sizeof(PreciseReal) >= sizeof(CoarseReal), "PreciseReal must have larger size than CoarseReal"); | ||
| // Use digits10 for this comparison in case the two types have differeing radixes: | ||
| static_assert(std::numeric_limits<PreciseReal>::digits10 >= std::numeric_limits<CoarseReal>::digits10, "PreciseReal must have higher precision that CoarseReal"); |
There was a problem hiding this comment.
Yes, I've noticed this problem before for MPFR types; since the sizeof operator returns the size of a pointer. Thanks!
| continue; | ||
| CoarseReal x = x_scale(plot.coarse_abscissas_[j]); | ||
| PreciseReal y = y_scale(static_cast<PreciseReal>(ulp[j] < 0 ? -plot.clip_ : plot.clip_)); | ||
| fs << "<circle cx='" << x << "' cy='" << y << "' r='1' fill='" << color << "'/>\n"; |
There was a problem hiding this comment.
Should we maybe color the clipped values differently? Suppose red for all clips?
There was a problem hiding this comment.
One further point: If the condition number goes infinite somewhere, which is very common, the top of the plot is going to be a bloodbath, even if there's really nothing wrong with the implementation.
I'm not sure what to do about this, since it seems like something no one wants an option for!
There was a problem hiding this comment.
+1 on red for clipped values, and yes the top of the plot may be a bloodbath, but my feeling is it's better to know than not about this stuff.
Yeah, but it could be a natural value to return at both precisions; e.g. sin(1/x) at x = 0. In this case there is no problem with the ulps plot, you just don't plot that value. So the only problem is when only one of the two precisions is a nan. In this case, maybe put it at the clip? But if there is no clip specified, then I would probably do nothing; or a message to |
|
OK yes, I see that we need to support NaN's when the function goes out of domain or whatever. It's tough to do the right thing in all cases here! |
Add more docs.
actually a helper function which returns Then I wouldn't make that one lgamma mistake in boostorg/multiprecision#264 :-)) |
|
@jzmaddock : I like these diffs; should I squash merge them? |
Yes please do, thanks! |
Use numeric_limits for precision comparison, plot out of bounds points at the clip boundary so we can still see that they're there.