Raise an error if max [in random(max)] is too large

[sorawee]

The result is wrong when calling random with large max. Below are the edge cases.

> random(9000000000000001)
83467816695816903274201855090766.0
> random(9000000000000000)
7878686929825859

Perhaps the easiest way to prevent this is to raise an error if max is too large.

This happens with both Chrome and Firefox on Mac

[sorawee]

We get the ridiculous result because the conversion from type 'number' (from Math.random()) to type 'BigInteger' is incorrect. We can prevent this by converting the result from Math.random() to Float before multiplying it with max (change return makeNumber(jsnums.floor(jsnums.multiply(Math.random(), max))); to return makeNumber(jsnums.floor(jsnums.multiply(jsnums.makeFloat(Math.random()), max)));)

There is still a problem when max grows larger (such as 9000000000000000001) because when BigInt is converted to fixnum, it will lose its precision, so the result will have a bunch of 0 at the end.

[jpolitz]

Fixed by 3db6cac by just using JS numbers in this case.

[sorawee]

@jpolitz : Now it becomes:

> random(10000000000000000000000000)
995534509420394700000000

This is not what we expect, isn't it?

[jpolitz]

It does satisfy being between 0 and 10000000000000000000000000, which is
what random() does. The precision is unfortunately bad because it's using
floating points now, so there are a lot of trailing zeroes.

In the long term, I'd like this to give more fine-grained results, but I'm
feel like the immediate problem is fixed until we have a more robust
numeric library to use with random().

Or are you pointing out a different problem that I'm not understanding? I
might just be missing something here.

On Mon Nov 03 2014 at 8:29:50 PM sorawee notifications@github.com wrote:

@jpolitz https://github.com/jpolitz : Now it becomes:

random(10000000000000000000000000)
995534509420394700000000

This is not what we expect, isn't it?

—
Reply to this email directly or view it on GitHub
#349 (comment).

[sorawee]

Yes, you understand what I tried to say correctly. In Racket, if max is too big, it will result in an error. I therefore thought that raising an error is a nice and easiest solution, but implementing bignum RNG is fine as well.

[sorawee]

That this issue was closed is only a workaround. The problem lying in this issue is not fixed yet because jsnums is supposed to work in any case (I just encountered this again while working with d3).

[jpolitz]

@ds26gte is there an easy way to do random numbers with bignums? Maybe not; I guess the underlying issue is that Math.random() simply doesn't give enough bits back. To actually get a good distribution of random numbers over, say, inputs like 9000000000000000000000000000000000000000000000000, we might have to do multiple calls to it.