Suggestion: add option for collide force to be deterministic

[benstevens48]

I'd like my force layout to produce the same results given the same inputs. However, since I'm using the collide force, this is not the case because the jiggle function uses Math.random. Can I suggest having the option to pass a seed number to the collide force constructor, which can be used as a seed for a pseudo-random number generator instead of Math.random? Or possibly it could be an option which applies to the whole simulation, and then any forces that use a random number can use the same seeded pseudo-random number generator, thus always producing the same layout given the same inputs.

[alexburner]

I'd love this too, we have this cute little monster in our webpack config right now:

{
    test: /jiggle\.js$/,
    include: /node_modules\/d3-force-3d/,
    use: [
        {
            loader: 'patch.loader.js',
            options: {
                search: 'return (Math.random() - 0.5) * 1e-6;',
                replace: 'return 0.1;',
            }
        },
    ],
},

[Fil]

We could add a seeded & deterministic prng (by default, no need for it to be optional?). See #160

[Fil]

Please test and confirm that #160 would work for the use cases you have in mind?

[benstevens48]

Probably. A couple of issues with that implementation of the random number generator though.

1. It's not centered around 0.
2. If a seed is provided, the result will likely be very small unless the seed is very large.

I would suggest this modification. The Math.abs is just in case the provided seed is negative.

return ((s = (a * Math.abs(seed || s) + c) % m) / m - 0.5) * 1e-6;

[Fil]

ah yes! thanks for catching this.

[mbostock]

do we need to expose jiggle(seed) so people could set the seed? (I don't think so)

Yes, I think we do; otherwise the only way to reset the state of a force simulation is to reload the page. I’m thinking that the force simulation should have an assignable random source (similar to d3-random’s random.source) allowing the caller to pass in the desired random generator. If we like, we could have that default to a linear congruential generator with a fixed seed, but I’d scope that to the simulation instance, so that by creating a new simulation you can effectively reset the state of the prng. I’ll put up a PR.