[squeak-dev] Re: Is there a better implementation of Floats?
nice
ncellier at ifrance.com
Wed Apr 9 19:23:21 UTC 2008
Bert Freudenberg wrote:
>
> On 09.04.2008, at 07:03, Bernd Elkemann wrote:
>> Hi
>> i am very impressed by squeaks handling of large integers
>> (LargePositiveInteger and resulting from that the Fraction-class).
>> However i was disappointed about the small number of digits of Floats
>> so my first question is: is there a possibility to increase it?
>> the second thing is: i found these examples on the squeak newbee group.
>> 1 - 0.2 -0.05 -0.3 -0.1= 0.35 "true"
>> 1 - 0.2 -0.05 -0.3 -0.1 - 0.10= 0.25 "*****false*****"
>> 1 - 0.2 -0.05 -0.3 -0.1 - 0.10= 0.24999999999999995 "true "
>> and i add the following examples:
>> 0.35 - 0.0 = 0.35 "true"
>> 0.35 - 0.1 = 0.25 "false"
>
> This is an intrinsic property of Floats, not specific to Squeak. It's
> how floating point numbers work in your processor. Google "0.1 floating
> point" if you want to know more.
>
>> Is there a better implementation of Floats?
>
> Depending on your definition of "better" you might consider ScaledDecimals:
>
> 0.35s2 - 0.1s2 = 0.25s2
>
> (in this example with 2 digits of precision)
>
> - Bert -
>
>
Yes, as Bert said, Floating point operations are inherently inexact,
while arithmetic on LargeInteger and Fraction are exact.
I'am personnaly in favour of answering false if the inexact floating
point representation once converted to an exact one is not equal to the
exact one.
(1/3.0) asTrueFraction = (1/3) -> false.
Current implementation is to answer true because the conversion is made
toward inexact:
(1/3.0) = (1/3) asFloat -> true.
Some are more extremists and argue such equality betwen exact and
inexact should ALWAYS return false.
ScaledDecimal as suggested by Bert are just Fraction with a different
method for printing (using an approximate decimal notation, beware, not
rounded in Squeak but truncated to a fixed number of decimal places
after the decimal point).
If you really want Floating point operations (that is inexact) with more
digits, I initiated the ArbitraryPrecisionFloat package at
http://www.squeaksource.com/ArbitraryPrecisionFl/ . You can download it
using Monticello or SqueakMap (better use Monticello, the SqueakMap will
point on an old version).
There are also other interesting work, MathMorph did have an
AlgebraicNumber if I remember well. However, you have either to take an
older squeak version, or wait for MathMorph to be ported in newer
versions to try it.
You now have enough refs to proceed by yourself...
Cheers
Nicolas
More information about the Squeak-dev
mailing list
|