Seeing JavaScript Math.random Not Be So Random

#javascript #estimation #montecarlosimulations

Read Flow Metrics. Learned about Monte Carlo Simulations. Learned you can use that to make accurate estimations with lots of data points. It’s one thing to hear “Math.random in JavaScript isn’t random”, and another to see it after only 1,000 samples. SOO COOL!

Context: “If you run Math.random enough between 1 & 6, you’re most likely to get 3.5”.

Code: https://ellie-app.com/kVjPdGBbP2ka1

Uses Elm Charts (horribly, sorry, I didn’t spend time styling it)

Article which helped get me some code: https://questsincode.com/posts/monte-carlo-simulation-javascript

Top comments (6)

Matt Ellen • Feb 7 '23

If you want to see the distribution of dice rolls, I think this codepen does an OK job:

Jesse Warden • Feb 7 '23

Rad, thank you! The blog post I linked to has one embedded too, but I like how this one uses Map.

Matt Ellen • Feb 7 '23 • Edited

Thanks, Jesse.

I hope that the frequencies my JS generates can demonstrate that the distribution of random numbers between 1 and 6 is fairly even, rather than being biased. Although the PRNG in use cannot be truly random, it does a good enough job.

I think this is basically what your Monte Carlo simulation tells you, too. It's not saying you'll get 3.5, it's saying the average is 3.5.

Magne • Feb 6 '23 • Edited

Not sure why Math.random eventually settling around the theoretical mean would mean it is not random? Wouldn't it settle around the theoretical mean even if it were random? As per The Law of Large Numbers. Ref: math.stackexchange.com/questions/4...

I'm sure Math.random is not actually fully random, as any algorithm without a bit of outside entropy would be deterministic, but I'm just curious about the inference made in this post.

The Y axis starts at 1 on your graph, not 0. So there are only 5 intervals, and 2.5 interval on each side of the mean. Seeing the mean settle half-way is expected, no?