Use FastRounding for speed #25

dpsanders · 2017-05-01T18:56:50Z

Use FastRounding.jl as a drop-in replacement for changing the rounding mode. This gives a 3-4x speed-up for basic arithmetic operations.

codecov-io · 2017-05-02T00:12:08Z

Codecov Report

Merging #25 into master will decrease coverage by 0.65%.
The diff coverage is 68%.

@@            Coverage Diff             @@
##           master      #25      +/-   ##
==========================================
- Coverage   92.29%   91.64%   -0.66%     
==========================================
  Files          21       21              
  Lines         935      945      +10     
==========================================
+ Hits          863      866       +3     
- Misses         72       79       +7

Impacted Files	Coverage Δ
src/IntervalArithmetic.jl	`100% <ø> (ø)`	⬆️
src/multidim/intervalbox.jl	`81.25% <ø> (ø)`	⬆️
src/intervals/arithmetic.jl	`98.76% <100%> (ø)`	⬆️
src/intervals/rounding.jl	`75% <65.21%> (-1%)`	⬇️
src/intervals/conversion.jl	`56.52% <0%> (-13.05%)`	⬇️
src/decorations/intervals.jl	`86.84% <0%> (-2.64%)`	⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b54e103...9ab304b. Read the comment docs.

coveralls · 2017-05-20T13:45:30Z

Coverage decreased (-0.2%) to 92.159% when pulling 6e7963b on fast_rounding into 64f72c7 on master.

coveralls · 2017-05-28T13:07:12Z

Coverage decreased (-0.2%) to 92.159% when pulling c5a5e4f on fast_rounding into 0d7f112 on master.

lbenet

Thanks! This is really nice. I suggest only to include docstrings, explain the restrictions of :errorfree and rename the rounding modes.

lbenet · 2017-06-01T22:55:56Z

src/intervals/arithmetic.jl


 mid{T}(a::Interval{Rational{T}}) = (1//2) * (a.lo + a.hi)

+function simple_mid(a::Interval)


Do we really need simple_mid as a new function? I think its sole use is inside that method for mid.

In any case, it is now good timing to add docstrings and tests.

I think you're right that we no longer need simple_mid.

lbenet · 2017-06-01T22:56:53Z

src/intervals/rounding.jl

@@ -9,6 +9,7 @@ interval rounding types, e.g.
 +(IntervalRounding{:none}, a, b, RoundDown)

 The current allowed rounding types are
+- :errorfree # use errorfree arithmetic via FastRounding.jl


I suggest the following changes for naming the rounding modes:

:errorfree should be named :fast
-:fast should be named :wider (or :wide?)

It should also be mentioned somewhere (the docstrings and perhaps a warning too) that :errorfree works only with Float64 and Float32.

I shall still refer to :errorfree instead of (the new) :fast to avoid confusions.

Agree with :fast and :wide, thanks for the suggestions.

Following the standard, how about :errorfree -> :fast; :fast -> :accurate; :correct-> :tight?

The problem is that this implies that :fast is not :tight, which is incorrect. (Errorfree arithmetic does give tight bounds.)

lbenet · 2017-06-01T23:01:42Z

src/intervals/rounding.jl

@@ -157,23 +199,31 @@ doc"""
    setrounding(Interval, rounding_type::Symbol)

 Set the rounding type used for all interval calculations on Julia v0.6 and above.
-Valid `rounding_type`s are `:correct`, `:fast` and `:none`.
+Valid `rounding_type`s are `:correct`, `:fast` and `:none`, `:errorfree`.


I think this should read as "Valid rounding_types are :correct, :fast, :none, and :errorfree".

Good point.

lbenet · 2017-06-01T23:06:52Z

I forgot to say that this should probably be rebased against current master; the tests related to cancelminus and cancelplus should all pass.

lbenet · 2017-06-04T17:36:07Z

I just read in the standard (cf page 9, definition of tightness) about what we are discussing as rounding modes. They use the nomenclature "tightest", "accurate" and "valid", which correspond respectively to :correct, :fast and :wide. I mention this in case we want to change the nomenclature to what appears in the standard.

… on 0.5 and 0.6

…ressed

lbenet · 2017-06-07T17:42:55Z

What is faster, :errorfree (now :fast) or :fast (now :accurate)?

I actually thought that the bounds provided by :errorfree were tighter than using prevfloat and nextfloat, and almost as good as :correct (now :tight).

I suggest to have the docstring in decreasing order of tightness.

dpsanders · 2017-06-07T17:45:28Z

What was :errorfree (now :fast) is exactly as accurate as :tight (what was :correct).
In fact, :tight (using changes of rounding mode) is now obsolete -- except to check that :fast is correct!

dpsanders · 2017-06-07T17:46:06Z

So maybe we can replace :errorfree / :fast with :tight, and :correct / :tight with :slow ;)

lbenet · 2017-06-07T17:51:17Z

Maybe you are right; give me a bit of time, since I thought i had some counterexample.

lbenet · 2017-06-07T18:49:20Z

So maybe we can replace :errorfree / :fast with :tight, and :correct / :tight with :slow ;)

I just run all ITF1788 suite tests and they pass with :errorfree / :fast. So, I agree: it should be renamed as :tight, and :correct / :tight with :slow, and the other that involves prevfloat/nextfloat should be :fast.

Sorry for being so skeptic...

coveralls · 2017-06-07T20:27:53Z

Coverage decreased (-0.2%) to 92.063% when pulling 28fa625 on fast_rounding into b54e103 on master.

dpsanders · 2017-06-07T22:51:05Z

I was quite sceptical actually that this would ever be possible. But thanks to @JeffreySarnoff and some hard work fixing difficult corner cases, it seems that it is finally finished!

dpsanders · 2017-06-07T22:52:15Z

So far, I have :tight, :accurate (the one just using prevfloat and nextfloat, for which you suggested :fast), :slow and :none. Is that OK with you?

JeffreySarnoff · 2017-06-07T22:53:10Z

!

…one)

lbenet · 2017-06-07T23:04:49Z

That's ok with me!

Once travis shows the green lights, I'll merge this.

coveralls · 2017-06-08T00:45:06Z

Coverage decreased (-0.7%) to 91.64% when pulling 9ab304b on fast_rounding into b54e103 on master.

coveralls · 2017-06-08T00:45:06Z

Coverage decreased (-0.7%) to 91.64% when pulling 9ab304b on fast_rounding into b54e103 on master.

coveralls · 2017-06-08T00:45:06Z

Coverage decreased (-0.7%) to 91.64% when pulling 9ab304b on fast_rounding into b54e103 on master.

dpsanders · 2017-06-08T00:57:44Z

Everything is passing!

lbenet · 2017-06-08T01:34:05Z

This is really nice! Merging...

dpsanders · 2017-06-08T04:18:01Z

Thanks!

dpsanders force-pushed the fast_rounding branch from 471bb6c to 5d65442 Compare May 1, 2017 23:19

dpsanders force-pushed the fast_rounding branch from 5d65442 to d409065 Compare May 2, 2017 03:26

dpsanders mentioned this pull request May 4, 2017

Test FastRounding to see if it is faster than changing the rounding mode JuliaIntervals/ValidatedNumerics.jl#150

Closed

dpsanders force-pushed the fast_rounding branch from d409065 to 6e7963b Compare May 20, 2017 12:52

dpsanders force-pushed the fast_rounding branch from 6e7963b to c5a5e4f Compare May 28, 2017 12:27

lbenet mentioned this pull request May 30, 2017

Constructing intervals #36

Closed

lbenet approved these changes Jun 1, 2017

View reviewed changes

dpsanders force-pushed the fast_rounding branch 2 times, most recently from 05b5e71 to 0228710 Compare June 5, 2017 09:12

dpsanders added 10 commits June 7, 2017 12:32

Try using a VERSION test to define IntervalBox with correct subtyping…

4f6b1f3

… on 0.5 and 0.6

Fix docstring

d486a56

Add error-free arithmetic and make it the default rounding type

da7342b

Resolve ambiguities for fast rounding on Julia 0.5

e01d28e

Rewrite definition of mid to fix test

e6c2f44

Return to error-free rounding

4b9875c

Add standard midpoint method if no alpha given

dcf58c0

Remove simple_mid function

51403f1

Improve description in rounding.jl

92d1ae6

Replace types with instances in rounding code

07f3211

dpsanders force-pushed the fast_rounding branch from 6eed03f to 07f3211 Compare June 7, 2017 17:32

dpsanders added 2 commits June 7, 2017 12:34

Move rounding tests back to interval_tests now that warnings are supp…

7feb9d3

…ressed

Rename :errorfree->:fast; :fast->:accurate; :correct->:tight

28fa625

Remove references to Suppressor

1c10fd6

Change rounding type names once more to (:tight, :accurate, :slow, :n…

9ab304b

…one)

lbenet merged commit 2ecefe5 into master Jun 8, 2017

dpsanders deleted the fast_rounding branch August 2, 2017 19:40

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use FastRounding for speed #25

Use FastRounding for speed #25

dpsanders commented May 1, 2017 •

edited

Loading

codecov-io commented May 2, 2017 •

edited

Loading

coveralls commented May 20, 2017 •

edited

Loading

coveralls commented May 28, 2017 •

edited

Loading

lbenet left a comment

lbenet Jun 1, 2017

dpsanders Jun 2, 2017

lbenet Jun 1, 2017

lbenet Jun 1, 2017

lbenet Jun 1, 2017

dpsanders Jun 2, 2017

dpsanders Jun 7, 2017

dpsanders Jun 7, 2017

lbenet Jun 1, 2017

dpsanders Jun 2, 2017

lbenet commented Jun 1, 2017

lbenet commented Jun 4, 2017

lbenet commented Jun 7, 2017

dpsanders commented Jun 7, 2017

dpsanders commented Jun 7, 2017

lbenet commented Jun 7, 2017

lbenet commented Jun 7, 2017

coveralls commented Jun 7, 2017 •

edited

Loading

dpsanders commented Jun 7, 2017

dpsanders commented Jun 7, 2017

JeffreySarnoff commented Jun 7, 2017

lbenet commented Jun 7, 2017

coveralls commented Jun 8, 2017

coveralls commented Jun 8, 2017

coveralls commented Jun 8, 2017 •

edited

Loading

dpsanders commented Jun 8, 2017

lbenet commented Jun 8, 2017

dpsanders commented Jun 8, 2017


		mid{T}(a::Interval{Rational{T}}) = (1//2) * (a.lo + a.hi)

		function simple_mid(a::Interval)

Use FastRounding for speed #25

Use FastRounding for speed #25

Conversation

dpsanders commented May 1, 2017 • edited Loading

codecov-io commented May 2, 2017 • edited Loading

Codecov Report

coveralls commented May 20, 2017 • edited Loading

coveralls commented May 28, 2017 • edited Loading

lbenet left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lbenet commented Jun 1, 2017

lbenet commented Jun 4, 2017

lbenet commented Jun 7, 2017

dpsanders commented Jun 7, 2017

dpsanders commented Jun 7, 2017

lbenet commented Jun 7, 2017

lbenet commented Jun 7, 2017

coveralls commented Jun 7, 2017 • edited Loading

dpsanders commented Jun 7, 2017

dpsanders commented Jun 7, 2017

JeffreySarnoff commented Jun 7, 2017

lbenet commented Jun 7, 2017

coveralls commented Jun 8, 2017

coveralls commented Jun 8, 2017

coveralls commented Jun 8, 2017 • edited Loading

dpsanders commented Jun 8, 2017

lbenet commented Jun 8, 2017

dpsanders commented Jun 8, 2017

dpsanders commented May 1, 2017 •

edited

Loading

codecov-io commented May 2, 2017 •

edited

Loading

coveralls commented May 20, 2017 •

edited

Loading

coveralls commented May 28, 2017 •

edited

Loading

coveralls commented Jun 7, 2017 •

edited

Loading

coveralls commented Jun 8, 2017 •

edited

Loading