There's a random-real generator in
"math/private/utils/flonum-tests.rkt". The generator Typed Racket uses
is in "tests/typed-racket/random-real.rkt".
I haven't settled on a general-purpose real number generator. For
flonums I usually write something like this:
#lang racket
(require math/base math/flonum)
(define max-ord (flonum->ordinal +max.0))
(define (random-flonum)
(define r (random))
(cond [(< r 0.025) +nan.0]
[(< r 0.050) +inf.0]
[(< r 0.075) -inf.0]
[(< r 0.100) -0.0]
[(< r 0.125) +0.0]
[else (ordinal->flonum (random-integer (- max-ord)
(+ max-ord 1)))]))
Sometimes I have cases for -max.0, -min.0, +min.0 and +max.0 to test
near-underflow and near-overflow conditions, or a (- (random) 0.5) or
similar case to test more typical inputs.
Rationals are a trickier because they're unbounded in both size and
precision. Probabilistically, they're a weird domain: a uniform
probability distribution over any given interval of rationals doesn't
exist. You can get around this by using a fixed precision, but what
would you fix it at?
Neil ⊥
On 05/13/2014 11:46 AM, Robby Findler wrote:
Okay, then I'll go with this:
(define/contract (angle->proper-range α)
(-> real? (between/c 0 360))
(let loop ([θ (- α (* 360 (floor (/ α 360))))])
(cond [(negative? θ) (+ θ 360)]
[(>= θ 360) (- θ 360)]
[else θ])))
Can you point me to your random real number generator, btw?
Robby
On Tue, May 13, 2014 at 9:06 AM, Neil Toronto <neil.toro...@gmail.com> wrote:
I can't get it to take more than on iteration, either. It's there in case I
missed something. :)
Neil ⊥
On 05/13/2014 05:59 AM, Robby Findler wrote:
Thanks, Neil!
Why is the loop needed? I can't seem to get it to take more than one
iteration.
Robby
On Mon, May 12, 2014 at 11:24 PM, Neil Toronto <neil.toro...@gmail.com>
wrote:
He went with exact rationals. Here's another option, which preserves
inexactness:
(define (angle->proper-range α)
(let loop ([θ (- α (* 360 (floor (/ α 360))))])
(cond [(negative? θ) (loop (+ θ 360))]
[(>= θ 360) (loop (- θ 360))]
[else θ])))
Its accuracy drops off outside of about [-1e16,1e16].
The fact that this is hard to get right might be good motivation for an
`flmodulo` function.
Neil ⊥
On 05/12/2014 09:49 PM, Sean Kanaley wrote:
Interesting, my code has the same bug then. I called it modulo/real,
used for things like displaying the space ship's rotation to the user or
wrapping x coordinates to stay in the world. Apparently it's going to
fail at some point with vector ref out of range. What was your fix? I
was thinking to just clamp explicitly like mod/real = (max 0 (min
the-mod-minus-1 (old-modulo/real x)))
On Mon, May 12, 2014 at 11:12 PM, Robby Findler
<ro...@eecs.northwestern.edu <mailto:ro...@eecs.northwestern.edu>>
wrote:
Right. Probably there is a better fix, but the essential problem,
as I
understand it, is that there are more floating points between 0 and
1
than between any two other integers and the code made the
assumption
that that didn't happen....
The basic desire is to turn a real number into a number in [0,360)
such that the result represents the same number in degrees but is
normalized somehow.
Robby
On Mon, May 12, 2014 at 10:06 PM, Danny Yoo <d...@hashcollision.org
<mailto:d...@hashcollision.org>> wrote:
> Wow. Floating point really is nasty. I see how it might have
happened now.
>
> ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
>> -0.0000000000000001
> -1e-16
>> (+ 360 -1e-16)
> 360.0
>>
> ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
____________________
Racket Users list:
http://lists.racket-lang.org/users
____________________
Racket Users list:
http://lists.racket-lang.org/users
____________________
Racket Users list:
http://lists.racket-lang.org/users
