`random_color()` and `random_bright_color()` should have a parameter `random_seed`

[kolibril13]

As it is always good to reproduce randomness, these two functions should have a parameter seed.
https://docs.manim.community/en/stable/_modules/manim/utils/color.html#random_color
Then it should be possible to just init that seed.
https://towardsdatascience.com/stop-using-numpy-random-seed-581a9972805f
EDIT: Second link does not apply here.

[behackl]

The link you have posted regarding the numpy random seed doesn't really apply for the random_color function, it doesn't use numpy's random module.

[kolibril13]

True, sorry!
I got confused, because there is also a choise function in numpy.
Here is the correct one:
https://docs.python.org/3/library/random.html#bookkeeping-functions

[RaghavPrabhakar66]

Hi, Can I take this issue?

[kolibril13]

Hi, Can I take this issue?

there was already a pull request one month ago here: #1676
but it seems to be stuck.
@RaghavPrabhakar66 : Feel free to review this pr, or, in case that you have another idea, you can also make a new pr

[RaghavPrabhakar66]

@kolibril13 I reviewed PR #1676, it works but I don't think we need if-else code statement, we can do it directly also.
Like rng = random.Random(random_seed).
Some Experiments:

>>> from manim.utils.color import random_color, random_bright_color
>>> random_color()
'#FFEA94'
>>> random_color()
'#00FF00'
>>> random_color(42)
'#ECABC1'
>>> random_color(random_seed=42)
'#ECABC1'
>>> random_color()
'#FFF1B6'
>>> random_bright_color()
<Color #ddd>
>>> random_bright_color()
<Color #fbd0d1>
>>> random_bright_color(42)
<Color #f5d5e0>
>>> random_bright_color(42)
<Color #f5d5e0>

So should I create a new pr?

[behackl]

So should I create a new pr?

Hey, I have reviewed that PR already a while ago, but the fundamental issue noted in #1676 (comment) has not been discussed or resolved yet. Removing the if-else will also not change that.

[RaghavPrabhakar66]

I don't think that this behavior is expected, but I also don't really have a good / simple solution for this. I think that essentially, we'd have to check whether a RNG with the given seed has already been initialized, and then use the existing one – or create one.

I don't know if I understood it correctly. What is the expected behavior? Is it if random.Random() is already defined with random_seed then no need to create RNG Object and If not then, we create RNG Object and set it's seed to random_seed.

I looked into it and I don't think its possible with python random module. Random does give us function such getstate() and setstate() which can be used to get info on its current state. getstate() function returns a list of size 624 and I dont think we can extract seed from that.

>>> import random
>>> random.getstate()[1][0]
2147483648
>>> random.seed(42)
>>> random.getstate()[1][0]
2147483648

Though this can be done in Numpy random module as np.random.get_state()[1][0] returns random_seed.

>>> import numpy as np
>>> np.random.get_state()[1][0]
2147483648
>>> np.random.seed(42)
>>> np.random.get_state()[1][0]
42
>>> np.random.randint(0, 100)
51
>>> np.random.randint(0, 100)
92
>>> np.random.randint(0, 100)
14
>>> np.random.get_state()[1][0]
723970371

But it also changes after some tries, so we can't determine the seed. I found something on stackoverflow related to it:

In summary on np.random.seed(s):
- get_state()[1][0] immediately after setting: retrieves s that exactly recreates the state
- get_state()[1][0] after generating numbers: may or may not retrieve s, but it will not recreate the current state (at get_state())
- get_state()[1][0] after generating many numbers: will not retrieve s. This is because pos exhausted its representation.
- get_state() at any point: will exactly recreate that point.
Lastly, behavior may also differ due to get_state()[3:] (and of course [0]).

So, I don't think we can retrieve current Random seed as the seed is used to update the internal state of the random number generator is not directly stored anywhere.

References:

• https://stackoverflow.com/questions/12697299/is-there-a-way-to-read-back-the-random-seed-in-python
• https://stackoverflow.com/questions/32172054/how-can-i-retrieve-the-current-seed-of-numpys-random-number-generator

[behackl]

Before we get into implementation details, we have to agree on what kind of behavior we should actually expect. Personally, I think that a situation like

>>> random_bright_color(42)
<Color #f5d5e0>
>>> random_bright_color(42)
<Color #f5d5e0>

is not desireable, because setting the seed then is equivalent to hard-coding one specific color. I would expect that repeatedly calling random_color(random_seed=42) continues to give random colors, but in a reproducible way.

You are right that the current seed of an initialized RNG cannot be retrieved -- but as it is mentioned in the SO answers you've posted, we can still store the seed of the RNG that has been initialized last somewhere. (Maybe in Manim's configuration, maybe "just" in the module; not sure about that.) Then, when random_color is called again, the passed seed can be compared to the stored one, and either the already initialized RNG can be used, or a new one can be initialized.

[RaghavPrabhakar66]

Yeah, we can store seed in config file and compare it if users provide random_seed.

I would expect that repeatedly calling random_color(random_seed=42) continues to give random colors, but in a reproducible way.

I was thinking maybe we could use getstate() and setstate() functions. Like we could return the state of seed at that with getstate() and later on set it back to its previous state using setstate().

In the meanwhile, I can try working on storing it in config system.

[hickmott99]

Has anyone closed this issue? Can I tackle it? @RaghavPrabhakar66 @behackl