Smalltalk › Squeak › Squeak - Beginners

Float comparsion

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

7 messages Options

Camillo Bruni-2

Float comparsion

Hy all,
I recently tried to compare two Float which apparently should be
(nearly) the same.
I have to following excerpt:

a := (Float halfPi / 2) tan
a = 1.0

The comparison evaluates to false, although it should be 1.0.
How do I compare Floats so I get useful results?

I'm using Squeak 3.8.18beta3U VM and the
Squeak3.10.2-7179-basic.image on Mac OS X 10.5 with an Intel Core 2 Duo.

Thanks,
Camillo
_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners

Bert Freudenberg

Re: Float comparsion

On 27.02.2009, at 13:22, Camillo Bruni wrote:

> Hy all,
> I recently tried to compare two Float which apparently should be
> (nearly) the same.
> I have to following excerpt:
>
> a := (Float halfPi / 2) tan
> a = 1.0
>
> The comparison evaluates to false, although it should be 1.0.
> How do I compare Floats so I get useful results?
>
> I'm using Squeak 3.8.18beta3U VM and the
> Squeak3.10.2-7179-basic.image on Mac OS X 10.5 with an Intel Core 2
> Duo.

Don't they teach you anything in Bern? ;-)

Floats are approximations. You can't meaningfully compare them for
equality. One way is to use

(a - b) abs < epsilon

- Bert -

_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners

Lukas Renggli

Re: Float comparsion

In reply to this post by Camillo Bruni-2

Hi Camillo ;-)

Use the method #closeTo: instead of #=. Comparing floats with #=
almost never works.

Although "(Float halfPi / 2) tan" is printed as 1.0, it is not exactly
1.0 due to rounding errors in the calucaltaion and/or the internal
representation.

(Float halfPi / 2) tan - 1.0 --> -1.110223024625156e-16

Lukas

--
Lukas Renggli
http://www.lukas-renggli.ch
_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners

onierstrasz

Re: Float comparsion

Yes, we looked at that, but closeTo: uses a much larger epsilon than
Float's class variable Epsilon.

I suggested Camillo wrote an extension method similar to closeTo: but
using the existing Epsilon (or an arbitrary one as an additional
parameter).

Is that the right thing to do?

- on

On Feb 27, 2009, at 13:34, Lukas Renggli wrote:

> Hi Camillo ;-)
>
> Use the method #closeTo: instead of #=. Comparing floats with #=
> almost never works.
>
> Although "(Float halfPi / 2) tan" is printed as 1.0, it is not exactly
> 1.0 due to rounding errors in the calucaltaion and/or the internal
> representation.
>
> (Float halfPi / 2) tan - 1.0 --> -1.110223024625156e-16
>
> Lukas
>
> --
> Lukas Renggli
> http://www.lukas-renggli.ch
> _______________________________________________
> Beginners mailing list
> [hidden email]
> http://lists.squeakfoundation.org/mailman/listinfo/beginners

_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners

Lukas Renggli

Re: Float comparsion

> Yes, we looked at that, but closeTo: uses a much larger epsilon than Float's
> class variable Epsilon.

Yeah, that implementation is a bit strange ;-)

As far as I understand the method uses an epsilon value relative to
the bigger absolute value of the two numbers. Depending on the exact
behavior you want, you might need to write your own extension method
indeed.

Number>>closeTo: aNumber epsilon: anEpsilon
^ self between: aNumber - anEpsilon and: aNumber + anEpsilon

Lukas

--
Lukas Renggli
http://www.lukas-renggli.ch
_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners

Bert Freudenberg

Re: Float comparsion

In reply to this post by onierstrasz

The Epsilon class variable is used only when the primitives for some
float functions are not available, in which case they are approximated
to that precision. Using that same epsilon for something else is as
meaningless as the 0.0001 in #closeTo:.

The epsilon really depends on your application.

- Bert -

On 27.02.2009, at 13:41, Oscar Nierstrasz wrote:

>
> Yes, we looked at that, but closeTo: uses a much larger epsilon than
> Float's class variable Epsilon.
>
> I suggested Camillo wrote an extension method similar to closeTo:
> but using the existing Epsilon (or an arbitrary one as an additional
> parameter).
>
> Is that the right thing to do?
>
> - on
>
> On Feb 27, 2009, at 13:34, Lukas Renggli wrote:
>
>> Hi Camillo ;-)
>>
>> Use the method #closeTo: instead of #=. Comparing floats with #=
>> almost never works.
>>
>> Although "(Float halfPi / 2) tan" is printed as 1.0, it is not
>> exactly
>> 1.0 due to rounding errors in the calucaltaion and/or the internal
>> representation.
>>
>> (Float halfPi / 2) tan - 1.0 --> -1.110223024625156e-16
>>
>> Lukas
>>
>> --
>> Lukas Renggli
>> http://www.lukas-renggli.ch
>> _______________________________________________
>> Beginners mailing list
>> [hidden email]
>> http://lists.squeakfoundation.org/mailman/listinfo/beginners
>
> _______________________________________________
> Beginners mailing list
> [hidden email]
> http://lists.squeakfoundation.org/mailman/listinfo/beginners

_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners

Camillo Bruni-2

Re: Float comparsion

hehe thanks for the vivid response ;)

Well then I'll use closeTo, I don't really need a very accurate comparison.

But since I want to create a Renderer the performance will be the biggest

problem. Well thats also why I've chosen Squeak ;).

thanks for the responses

Am 27.02.2009 um 13:51 schrieb Bert Freudenberg:

The Epsilon class variable is used only when the primitives for some float functions are not available, in which case they are approximated to that precision. Using that same epsilon for something else is as meaningless as the 0.0001 in #closeTo:.

The epsilon really depends on your application.

- Bert -

On 27.02.2009, at 13:41, Oscar Nierstrasz wrote:

Yes, we looked at that, but closeTo: uses a much larger epsilon than Float's class variable Epsilon.

I suggested Camillo wrote an extension method similar to closeTo: but using the existing Epsilon (or an arbitrary one as an additional parameter).

Is that the right thing to do?

- on

On Feb 27, 2009, at 13:34, Lukas Renggli wrote:

Hi Camillo ;-)

Use the method #closeTo: instead of #=. Comparing floats with #=
almost never works.

Although "(Float halfPi / 2) tan" is printed as 1.0, it is not exactly
1.0 due to rounding errors in the calucaltaion and/or the internal
representation.

(Float halfPi / 2) tan - 1.0 --> -1.110223024625156e-16

Lukas

--
Lukas Renggli
http://www.lukas-renggli.ch
_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners

_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners

_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners

_______________________________________________
Beginners mailing list
[hidden email]
http://lists.squeakfoundation.org/mailman/listinfo/beginners