Using seed breaks .integer

I’m using 1.0.18. Reproduce with this: new Chance(Math.random()).integer({min:1,max:3}) This returns a value of 2 100% of the time.

[somejeff]

Hey @gerardolima, here’s my feedback as to why Math.random() as a seed isn’t optimal, but also technically supported.

TL;DR: The built-in MersenneTwister takes the seed, unsigned right shifted, and goes from there. Anything between 0 and 1 becomes 0.

---

The built-in MersenneTwister uses the seed to populate an array of pseudorandom numbers. It’s expecting a seed between 0 and 10_000_000_000_000

Let’s pick a number from 1 to 10, using seeds 0 to 4.

new Chance(0).integer({min:1,max:10})

Seed	First few elements of the array	First call to mt.random() to choose an int	Result
0	[0, 1, 1812433255, ...]	0.548813502304256	6
1	[1, 1812433254, ...]	0.4170219984371215	5
2	[2, 3624866507, ...]	0.43599490262567997	5
3	[3, 1142332464, ...]	0.5507979043759406	6
4	[4, 2954765717, ...]	0.9670298383571208	10

Initializing the MT with such close integers and only providing a field of 1-10 for integer() to choose (once), returns 6,5,5,10.

---

Narrow the field to {min:1,max:3} is like shrinking the dart board down to only 3 options.

new Chance(0).integer({min:1,max:3})

Seed	First few elements of the array	First call to mt.random() to choose an int	Result
0	[0, 1, 1812433255, ...]	0.548813502304256	2
1	[1, 1812433254, ...]	0.4170219984371215	2
2	[2, 3624866507, ...]	0.43599490262567997	2
3	[3, 1142332464, ...]	0.5507979043759406	2
4	[4, 2954765717, ...]	0.9670298383571208	3

---

Let’s see what happens if you pass a seed between 0 and 1, essentially populating the array with zero as the first entry and always returning the same pseudorandom value.

new Chance(0.x).integer({min:1,max:10})

Seed	First few elements of the array	First call to mt.random() to choose an int	Result
0	[0, 1, 1812433255, ...]	0.548813502304256	6
0.1	[0, 1, 1812433255, ...]	0.548813502304256	6
0.2	[0, 1, 1812433255, ...]	0.548813502304256	6
0.3	[0, 1, 1812433255, ...]	0.548813502304256	6
0.4	[0, 1, 1812433255, ...]	0.548813502304256	6
0.5	[0, 1, 1812433255, ...]	0.548813502304256	6
0.6	[0, 1, 1812433255, ...]	0.548813502304256	6
0.7	[0, 1, 1812433255, ...]	0.548813502304256	6
0.8	[0, 1, 1812433255, ...]	0.548813502304256	6
0.9	[0, 1, 1812433255, ...]	0.548813502304256	6
0.99999999	[0, 1, 1812433255, ...]	0.548813502304256	6
1	[1, 1812433254, ...]	0.4170219984371215	5

The same would apply to {min:1,max:3} and always return 2.

---

Could the MersenneTwister be improved to support real numbers between 0 and 1 instead of ints?
Sure, but any change to the twister is a breaking change for all those expecting patterns from their fixed seeds.

Could the constructor validate and reject real numbers? Also a breaking change for those passing a real number with a larger field, such as: new Chance(Math.random()*10_000_000_000_000);

String seeds? If a String is passed, it’s converted into an integer using each letter of the string. 😐

---

Recommendation?

Take advantage of the full 32 bits for a seed and stay away from 0 -> 1 and the other limit too:

new Chance(Math.random() * Infinity);  // https://xkcd.com/221/

[gerardolima]

hey again, @somejeff, I understand your explanation and I agree that keeping a stable interface is precious for any library.

My comment was about improving the ergonomics of the API by avoiding what looks to be a rather silly, but also one easy mistake to make. As a suggestion, Chance could trigger an error when seed % 1 !== 0 or maybe being even more explicit, when 0 < seed < 1, which is the range of Math.random(). If throwing an error should sound too much, maybe log a warning on console?

That said, I understand that this might break the behaviour in some cases – and maybe unfold some unintended usage of Chance(Math.random()) – so it’s a balance and it’s obviously your call. Thank you for your good work!