Check distribution of arcade.math.rand_* functions

pythonarcade / arcade

Easy to use Python library for creating 2D arcade games.

http://arcade.academy

Other

1.71k stars 330 forks source link

Check distribution of arcade.math.rand_* functions #2428

Open pushfoo opened 1 month ago

pushfoo commented 1 month ago

Enhancement request:

TL;DR: Check for more random call centering bias like that of #2425

What needs verification

In #2426, a call to math.sqrt to wrap the random call seems to fix an issue with center-clustering bias in rand_in_circle. It seems reasonable to check for & fix any similar bias in:

[x] arcade.math.rand_on_circle
[x] arcade.math.rand_in_rect
[x] arcade.math.rand_on_line
[x] arcade.math.rand_angle_360_deg
[x] arcade.math.rand_angle_spread_deg
[x] arcade.math.rand_spread_deg
[x] arcade.math.rand_magnitude
[x] ~~Any upstream equivalents in the pyglet codebase?~~ no use of random() seems to exist on dev branch outside tests

What would it help with?

Evenness of randomized points.

Best approach?

I'm open to suggestions on this. These seem like hot code paths so I'm uncertain of OOP-y abstractions / indirection layers to customize which distribution function we're mapping into. Thoughts welcome.

pushfoo commented 1 month ago

@Rapen765 Are you up for taking a look at this?

Rapen765 commented 1 month ago

Yes, I am currently checking the functions and I am sure that those are correct: rand_on_line - the output is a line so a linear random function is good rand_on_circle - the output is also a line but not straight, but still the function is good rand_angle_360_deg - this is just a random value from 0 to 360 rand_magnitude - this is just like rand_on_line but instead of two points there is an angle and two magnitudes rand_angle_spread_deg - this is just like rand_angle_360_deg but instead of the fixed angle you can change it rand_vec_spread_deg - this is the rand_angle_spread_deg but converted to vector the rand_in_rect function is also correct.

Rapen765 commented 1 month ago

The only problem is the function rand_on_line documentation

def rand_on_circle(center: Point2, radius: float) -> Point2: """ Generate a point on a circle.

.. note: by passing a random value in for float,
         you can achieve what rand_in_circle() does

Args:
    center (Point2): The center of the circle
    radius (float): The radius of the circle
"""
angle = 2 * math.pi * random.random()
return (radius * math.cos(angle) + center[0], radius * math.sin(angle) + center[1])

the note is now incorrect because you would have to pass a sqrt(random()) and not random(). Also not in for float but for radius.

pushfoo commented 1 month ago

While we wait for feedback from Dragon on the tuple vs Vec2 perf issues, would you be up for adding this doc change to #2430?