Initial checkin for a module for tolerant comparison of reals #353

arjenmarkus · 2021-03-20T11:35:09Z

The module is based on the idea to allow for a small margin when comparing reals. Such a margin can solve the problem that the finite precision of floating-point calculations make the results inaccurate. Two essentially equal real numbers may therefore appear to be different. This is in response to issue #327

The module is based on the idea to allow for a small margin when comparing reals. Such a margin can solve the problem that the finite precision of floating-point calculations make the results inaccurate. Two essentially equal real numbers may therefore appear to be different

src/CMakeLists.bak

This backup file (from an edit action) was not supposed to go into the repository

arjenmarkus · 2021-03-20T11:58:06Z

Certainly not - github desktop is all too keen on submitting all files residing in the checkout directory. Op za 20 mrt. 2021 om 12:42 schreef Sebastian Ehlert < ***@***.***>:

…

***@***.**** commented on this pull request. ------------------------------ In src/CMakeLists.bak <#353 (comment)>: > @@ -0,0 +1,73 @@ +### Pre-process: .fpp -> .f90 via Fypp src/CMakeLists.bak is not supposed to be committed here, is it? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#353 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAN6YRZNNYMKTSPCFL45KS3TESCZJANCNFSM4ZQLBXTQ> .

milancurcic · 2021-03-22T19:46:35Z

Thanks @arjenmarkus, I think these will be useful.

Is "tolerant" a commonly and uniquely used term? I wonder if most people would understand what it is just based on that word.

What do you think about stdlib_comparisons? But then it misses the point of the comparisons being tolerant. And stdlib_tolerant_comparisons is unwieldy.

I guess my main gripe with the word "tolerant" is that it's an adjective and not a noun, i.e. it doesn't say what is in the module, but what something is like in the module.

arjenmarkus · 2021-03-23T08:15:36Z

The original source code mentions "fuzzy Fortran operators", that may be more apt, but then I fear people would expect fuzzy sets and the like. I am not too happy with the term tolerant either, so I am open to suggestions. Op ma 22 mrt. 2021 om 20:46 schreef Milan Curcic ***@***.***>:

…

Thanks @arjenmarkus <https://github.com/arjenmarkus>, I think these will be useful. Is "tolerant" a commonly and uniquely used term? I wonder if most people would understand what it is just based on that word. What do you think about stdlib_comparisons? But then it misses the point of the comparisons being tolerant. And stdlib_tolerant_comparisons is unwieldy. I guess my main gripe with the word "tolerant" is that it's an adjective and not a noun, i.e. it doesn't say what is in the module, but what something is like in the module. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#353 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAN6YR7YUGSMHHJXY77MEXDTE6NCXANCNFSM4ZQLBXTQ> .

milancurcic · 2021-03-23T22:35:02Z

Currently stdlib_comparison or stdlib_comparisons is my favorite. We can request feedback on the call tomorrow.

arjenmarkus · 2021-03-24T08:13:03Z

The plural has my preference. As I will not be able to join today, I will simply await the results. Op di 23 mrt. 2021 om 23:35 schreef Milan Curcic ***@***.***>:

…

Currently stdlib_comparison or stdlib_comparisons is my favorite. We can request feedback on the call tomorrow. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#353 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAN6YR7REP2OB5A4FINV4V3TFEJSJANCNFSM4ZQLBXTQ> .

jvdp1

Thank ypou @arjenmarkus for this PR. Here are some comments.
I agree with @milancurcic: stdlib_comparison or stdlib_comparisons would be better thatn stdlib_tolerant. However, they don't cover tfloor,,... What do you think about stdlib_operators?

General comments:

Some hyperlinks mst still be added in the code, but this could be done at the end of the PR
fypp could be used to generated the code for the different kinds of real.

jvdp1 · 2021-07-25T18:45:18Z

doc/specs/stdlib_tolerant.md

+
+Provide tolerant versions of the `floor`, `ceil` and `round` functions that take a small interval into account.
+While the actual interval can be set, the advised default is three times epsilon(). Note that the implementation
+is actually much more involved  than would seem necessary. It is the result of extensive research.


Suggested change

is actually much more involved than would seem necessary. It is the result of extensive research.

is actually much more involved than would seem necessary. It is the result of extensive research.

jvdp1 · 2021-07-25T18:45:42Z

doc/specs/stdlib_tolerant.md

+#### Description
+
+Provide tolerant versions of the `floor`, `ceil` and `round` functions that take a small interval into account.
+While the actual interval can be set, the advised default is three times epsilon(). Note that the implementation


Suggested change

While the actual interval can be set, the advised default is three times epsilon(). Note that the implementation

While the actual interval can be set, the advised default is three times `epsilon()`. Note that the implementation

jvdp1 · 2021-07-25T18:46:22Z

doc/specs/stdlib_tolerant.md

+
+`x`: The real number that is to be truncated or rounded
+
+`ct`: Tolearance for comparison (optional, defaults to 3*epsilon)


Suggested change

`ct`: Tolearance for comparison (optional, defaults to 3*epsilon)

`ct`: Tolerance for comparison. It is an `intent(in)` and `optional` argument. The default is 3*`epsilon().

jvdp1 · 2021-07-25T18:47:44Z

doc/specs/stdlib_tolerant.md

+
+#### Arguments
+
+`x`: The real number that is to be truncated or rounded


Suggested change

`x`: The real number that is to be truncated or rounded

`x`: The real number that is to be truncated or rounded. It is an `intent(in)` argument.

jvdp1 · 2021-07-25T18:49:08Z

src/stdlib_tolerant.f90

+    integer, parameter, private       :: sp = kind(1.0)
+    integer, parameter, private       :: dp = kind(1.0d0)


Would it be possible to use the ones defined in stdlib_kinds?

jvdp1 · 2021-07-25T18:51:34Z

doc/specs/stdlib_tolerant.md

+
+#### Arguments
+
+`x`, `y`: Two reals (of the same kind) to be compared


Are they scalar? Or are arrays also possible (and then the function should be elemental?

jvdp1 · 2021-07-25T18:52:59Z

src/stdlib_tolerant.f90

+logical function teq_sp( x, y ) result(cmp)
+    real(kind=sp), intent(in) :: x, y
+


All the functions for the different kinds could be generated with fypp.

jvdp1 · 2021-07-25T18:55:36Z

src/tests/tolerant/test_tolerant.f90

+
+    implicit none
+
+    integer, parameter :: dp = kind(1.0d0)


maybe using the ones defined by stdlib_kinds?

jvdp1 · 2021-07-25T18:56:44Z

src/tests/tolerant/test_tolerant.f90

+    real(kind=dp)      :: x, y, z
+    real(kind=dp)      :: yfloor, yceil
+    integer            :: i
+
+    real(kind=dp)      :: eps3 = 3.0_dp * epsilon(eps3)


fypp could be used to generate the same tests for the other kinds.

awvwgk · 2021-12-19T14:09:14Z

With #488 this functionality was implemented, therefore I'm going to close this PR.

Feel free to reopen this PR if you think #488 doesn't address all changes included in this patch.

awvwgk reviewed Mar 20, 2021

View reviewed changes

src/CMakeLists.bak Outdated Show resolved Hide resolved

arjenmarkus added 2 commits March 20, 2021 12:52

Merge branch 'master' into tolerant

3cc6f62

Delete CMakeLists.bak

e837d9e

This backup file (from an edit action) was not supposed to go into the repository

awvwgk added reviewers needed This patch requires extra eyes API Discussion on a specific API for a proposal (exploratory) labels Apr 17, 2021

jvdp1 reviewed Jul 25, 2021

View reviewed changes

jvdp1 added the waiting for OP This patch requires action from the OP label Jul 25, 2021

milancurcic mentioned this pull request Aug 19, 2021

[stdlib_math] add is_close routines. #488

Merged

6 tasks

awvwgk added topic: mathematics linear algebra, sparse matrices, special functions, FFT, random numbers, statistics, ... and removed reviewers needed This patch requires extra eyes labels Sep 18, 2021

awvwgk closed this Dec 19, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Initial checkin for a module for tolerant comparison of reals #353

Initial checkin for a module for tolerant comparison of reals #353

arjenmarkus commented Mar 20, 2021

arjenmarkus commented Mar 20, 2021 via email

milancurcic commented Mar 22, 2021

arjenmarkus commented Mar 23, 2021 via email

milancurcic commented Mar 23, 2021

arjenmarkus commented Mar 24, 2021 via email

jvdp1 left a comment

jvdp1 Jul 25, 2021

jvdp1 Jul 25, 2021

jvdp1 Jul 25, 2021

jvdp1 Jul 25, 2021

jvdp1 Jul 25, 2021

jvdp1 Jul 25, 2021

jvdp1 Jul 25, 2021

jvdp1 Jul 25, 2021

jvdp1 Jul 25, 2021

awvwgk commented Dec 19, 2021

	is actually much more involved than would seem necessary. It is the result of extensive research.
	is actually much more involved than would seem necessary. It is the result of extensive research.

	While the actual interval can be set, the advised default is three times epsilon(). Note that the implementation
	While the actual interval can be set, the advised default is three times `epsilon()`. Note that the implementation


		`x`: The real number that is to be truncated or rounded

		`ct`: Tolearance for comparison (optional, defaults to 3*epsilon)

	`ct`: Tolearance for comparison (optional, defaults to 3*epsilon)
	`ct`: Tolerance for comparison. It is an `intent(in)` and `optional` argument. The default is 3*`epsilon().


		#### Arguments

		`x`: The real number that is to be truncated or rounded

		integer, parameter, private :: sp = kind(1.0)
		integer, parameter, private :: dp = kind(1.0d0)


		#### Arguments

		`x`, `y`: Two reals (of the same kind) to be compared

		logical function teq_sp( x, y ) result(cmp)
		real(kind=sp), intent(in) :: x, y

Initial checkin for a module for tolerant comparison of reals #353

Initial checkin for a module for tolerant comparison of reals #353

Conversation

arjenmarkus commented Mar 20, 2021

arjenmarkus commented Mar 20, 2021 via email

milancurcic commented Mar 22, 2021

arjenmarkus commented Mar 23, 2021 via email

milancurcic commented Mar 23, 2021

arjenmarkus commented Mar 24, 2021 via email

jvdp1 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

awvwgk commented Dec 19, 2021