trealla
trealla copied to clipboard
trealla-prolog
Missing rounding in float/1 for bigint argument
I get these results (note the 119042423827613008 is wrong):
/* Trealla Prolog 2.2.10 */
?- X is 51^10, Z is float(X).
X = 119042423827613008, Z = 1.19042423827613e+17.
?- Z is float(51^10).
Z = 1.19042423827613e+17.
But one would rather expect that float/1 performs HALF_EVEN rounding. The results should then be (1.1904242382761301E17 instead of 1.19042423827613e+17):
?- X is 51^10, Z is float(X).
X = 119042423827613001, Z = 1.1904242382761301E17.
?- Z is float(51^10).
Z = 1.1904242382761301E17.
That the (^)/2 result is wrong, is rather irrelevant, it is also:
?- Z is float(119042423827613008).
Z = 1.1904242382761301E17.
It's a printing error.
On Wed, Sep 28, 2022 at 7:33 PM Jean-Luc-Picard-2021 < @.***> wrote:
I get these results (note the 119042423827613008 is wrong):
/* Trealla Prolog 2.2.10 */ ?- X is 51^10, Z is float(X). X = 119042423827613008, Z = 1.19042423827613e+17. ?- Z is float(51^10). Z = 1.19042423827613e+17.
But one would rather expect that float/1 performs HALF_EVEN rounding. The results should then be (1.1904242382761301E17 instead of 1.19042423827613e+17):
?- X is 51^10, Z is float(X). X = 119042423827613001, Z = 1.1904242382761301E17. ?- Z is float(51^10). Z = 1.1904242382761301E17.
— Reply to this email directly, view it on GitHub https://github.com/trealla-prolog/trealla/issues/48, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFNKSEX5ECOEJBRKJUEKHODWAQGHRANCNFSM6AAAAAAQXSU4CE . You are receiving this because you are subscribed to this thread.Message ID: @.***>
I mean print rounding error. using 17 digits instead of 18.
On Wed, Sep 28, 2022 at 7:33 PM Jean-Luc-Picard-2021 < @.***> wrote:
I get these results (note the 119042423827613008 is wrong):
/* Trealla Prolog 2.2.10 */ ?- X is 51^10, Z is float(X). X = 119042423827613008, Z = 1.19042423827613e+17. ?- Z is float(51^10). Z = 1.19042423827613e+17.
But one would rather expect that float/1 performs HALF_EVEN rounding. The results should then be (1.1904242382761301E17 instead of 1.19042423827613e+17):
?- X is 51^10, Z is float(X). X = 119042423827613001, Z = 1.1904242382761301E17. ?- Z is float(51^10). Z = 1.1904242382761301E17.
— Reply to this email directly, view it on GitHub https://github.com/trealla-prolog/trealla/issues/48, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFNKSEX5ECOEJBRKJUEKHODWAQGHRANCNFSM6AAAAAAQXSU4CE . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Yes it seems so:
?- Z is float(51^10), T is Z-1.1904242382761301E17.
Z = 1.19042423827613e+17, T = 0.0
Yeah, can fix it by going to 18 digits of precision, but then you get things like 3.00000000000001 or 2.9999999999999997 (Scryer) for what should be integers.
On Wed, Sep 28, 2022 at 9:57 PM Jean-Luc-Picard-2021 < @.***> wrote:
Yes it seems so:
?- Z is float(51^10), T is Z-1.1904242382761301E17. Z = 1.19042423827613e+17, T = 0.0
— Reply to this email directly, view it on GitHub https://github.com/trealla-prolog/trealla/issues/48#issuecomment-1260796849, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFNKSERIF6X6N2I4HVDNKN3WAQXC7ANCNFSM6AAAAAAQXSU4CE . You are receiving this because you commented.Message ID: @.***>
Can you show a Prolog query where this happens? Also its not necessarily a display problem. For example many Math.pow(double,double) implementations cannot
do the problem correctly. They cannot or do not round to the theoretical optimum, they have inherently a higher error than only one Unit in the last place (ULP). So a certain error
is to expect. You see this here:
/* Dogelog Player, JavaScript */
?- X is 51**10.
X = 1.19042423827613E17.
?- X is float(51^10).
X = 119042423827613010.0.
In the above the (**)/2 is Math.pow(double,double) and the
(^)/2 is exact bigint exponentiation. I took the pair (51,10)
from a list of pairs where (**)/2 and (^)/2 often disagree
based on Math.pow(double,double) typically being weak.
When printing with DBL_DECIMAL_DIG-1 (17 i think) digits of precision
?- N9 is 370370367037037036703703703670 / 123456789012345678901234567890. N9 = 3.0. ?- X is 51^10, Z is float(X). X = 119042423827613001, Z = 1.19042423827613e+17.
when printing with DBL_DECIMAL_DIG (18 i think) digits of precision
?- N9 is 370370367037037036703703703670 / 123456789012345678901234567890. N9 = 3.0000000000000004. ?- X is 51^10, Z is float(X). X = 119042423827613001, Z = 1.1904242382761301e+17.
$ scryer-prolog -f ?- N9 is 370370367037037036703703703670 / 123456789012345678901234567890. N9 = 2.9999999999999996. ?- X is 51^10, Z is float(X). X = 119042423827613001, Z = 1.19042423827613e17.
On Wed, Sep 28, 2022 at 11:22 PM Jean-Luc-Picard-2021 < @.***> wrote:
Can you make show a query where this happens? I don't know how to reproduce 3.00000000000001 or 2.9999999999999997 which should be 3. After all 3 is an exact float.
— Reply to this email directly, view it on GitHub https://github.com/trealla-prolog/trealla/issues/48#issuecomment-1260907889, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFNKSESP3H75CVQUQHAAGA3WARBARANCNFSM6AAAAAAQXSU4CE . You are receiving this because you commented.Message ID: @.***>
$ swipl Welcome to SWI-Prolog (threaded, 64 bits, version 8.4.3) ?- N9 is 370370367037037036703703703670 / 123456789012345678901234567890. N9 = 3. ?- X is 51^10, Z is float(X). X = 119042423827613001, Z = 1.1904242382761301e+17.
On Wed, Sep 28, 2022 at 11:22 PM Jean-Luc-Picard-2021 < @.***> wrote:
Can you make show a query where this happens? I don't know how to reproduce 3.00000000000001 or 2.9999999999999997 which should be 3. After all 3 is an exact float.
— Reply to this email directly, view it on GitHub https://github.com/trealla-prolog/trealla/issues/48#issuecomment-1260907889, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFNKSESP3H75CVQUQHAAGA3WARBARANCNFSM6AAAAAAQXSU4CE . You are receiving this because you commented.Message ID: @.***>
I have opened a separate ticket for printing. Because this can get extremly confusing, if printing is not precise enough. The ticket is here:
not precise enough float printing https://github.com/trealla-prolog/trealla/issues/49
And i suppose X is 1/10 with 18 digits you get
$ tpl ?- X is 1/10. X = 0.10000000000000001.
(as does GNUProlog).
Versus a better looking 0.1 (SWI, Scryer, BProlog).
On Wed, Sep 28, 2022 at 11:22 PM Jean-Luc-Picard-2021 < @.***> wrote:
Can you make show a query where this happens? I don't know how to reproduce 3.00000000000001 or 2.9999999999999997 which should be 3. After all 3 is an exact float.
— Reply to this email directly, view it on GitHub https://github.com/trealla-prolog/trealla/issues/48#issuecomment-1260907889, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFNKSESP3H75CVQUQHAAGA3WARBARANCNFSM6AAAAAAQXSU4CE . You are receiving this because you commented.Message ID: @.***>
printing needs to satisfy. If this is not satisfied, its very hard to debug other errors, it will "mask" other errors, like erros or variations in arithmetic such
as (/)/2, (**)/2, (^)/2 and float/2:
read(write(X))) == X
in the 3.0 example its clear. In SWI-Prolog the result is different. Try subtracting 3 from the division? The 1/10 example, I don't know, maybe better discuss it here:
not precise enough float printing https://github.com/trealla-prolog/trealla/issues/49
Here you see SWI-Prolog computes another number, the issue is currently discussed in SWI-Prolog forum. They use mqr_get_d instead mqr_get_d_nearest or
somesuch is their problem. Also their a varying approachs to
realize (/)/2, either with divmod(A,B,...) or simply
(float)A/(float)B. Their system might also differ in this respect:
/* Trella Prolog */
?- N9 is 370370367037037036703703703670 / 123456789012345678901234567890 - 3.
N9 = 4.440892098500626e-16.
/* SWI-Prolog */
?- N9 is 370370367037037036703703703670 / 123456789012345678901234567890 - 3.
N9 = 0.
The (^)/2 error is gone, I now find:
Trealla Prolog (c) Infradig 2020-2022, v2.7.15
?- X is 51^10.
X = 119042423827613001
And the rest is a printing artefact, as already observed:
?- X is 51^10, Z is float(X), Z =:= 1.19042423827613008E17.
X = 119042423827613001, Z = 1.19042423827613e+17.
Closing this ticket.