Math/Random and its tests implementation by sariug · Pull Request #432 · mosra/magnum

sariug · 2020-04-06T15:20:42Z

Hello hello,

Did some simple implementation; would be happy to develop it further with the suggestions.
Prepared for one engine ("mt19937")

mosra

I really appreciate that you did a test as well 👍

Since this is your first contribution, I'm not commenting on coding style -- let's get this working first :)

src/Magnum/Math/Test/RandomTest.cpp

mosra · 2020-04-08T15:03:45Z

src/Magnum/Math/Random.h

+public:
+    static std::mt19937 &generator()
+    {
+        static std::mt19937 g{seeds};


static is problematic because of thread safety issues, global constructors/destructors etc. What do you think about this?

Math::RandomGenerator g; // the user is required to instantiate this first auto a = Math::randomSignedScalar(g); auto b = Math::randomUnitVector2(g);

The RandomGenerator thus wouldn't be in Implementation anymore, but will be exposed to the user, even though its only use would be to get passed to the random*() functions.

I thought about it, and tried to keep minimalistic(however you are right for the thread safety), to keep it as a 1-step function.

if you have no other proposal and think that this would be the best, then why not :)

Not sure if best, but I didn't come up with anything better yet.

I think this should be also robust in case I decide to implement my own random generator later.

Implemented some stuff. Wondering how you would find these. I feel quiteee doubtful.(Especially would love to hide the "generate" function;

mosra · 2020-04-08T15:17:05Z

src/Magnum/Math/Random.h

+Vector2<T> randomUnitVector2()
+{
+    auto a = Implementation::RandomGenerator::generate(0.0f, 2 * Math::Constants<T>::pi());
+    return {std::cos(a), std::sin(a)};


I guess this is where things become hard 😅

I actually have no idea here -- will the distribution be still uniform if sin/cos is used? I guess it will? Ideally this would be without the extra overhead of trig functions, but I don't have any idea if that's doable ... in this thread on SO they use a Gaussian distribution as an input, but .. ¯\_(ツ)_/¯

If you have better references than me, mention them here please :)

For this case sin and cos shall not be get affected ? (sin (+ + - - )[50%], cos (+ - - + )[50%] )
So this shall be as accurate as the generator itself ?

Can you also comment about what I read here ? It looks accurate according to this. The main discussion seems like "getting 3 random numbers and normalizing" vs "2 numbers(theta and height) and calculating". The latter seems more accurate, which is sin/cos.

Yes that's a great reference, thanks -- a link to it should go in the documentation. I didn't absorb it fully yet, but yeah it sounds like a good proof.

getting 3 random numbers and normalizing

this is what you do for quaternions tho .. can the two-/three-dimensional case be extended for those as well?

Quaternion is link here.

Added both to the code :)

src/Magnum/Math/Random.h

mosra · 2020-04-08T15:27:20Z

src/Magnum/Math/Random.h

+bool randomBool()
+{
+    return static_cast<bool>(randomUnsignedScalar<Int>());
+}


Hmm, do we actually need this? :) The randomUnsignedScalar() for integers isn't really useful either. Maybe instead add a static_assert(IsFloatingPoint<T>::value, ""); to the RandomGenerator and remove everything related to integers? Those can be always added back later. (The trait is in Magnum/Math/TypeTraits.h.)

Well, like i said in the beginning, a initiation for a new magnum functionality :) Can grow later with demand :)

Deleting/Fixing. :)

I first deleted, and then put it back haha :D
From our initial discussion:

Float randomUnsignedScalar(); // range [0, 1] Float randomSignedScalar(); // range [-1, 1] Vector2 randomPointOnACircle(); // always length = 1 Vector3 randomPointOnASphere(); // also Quaternion randomRotation(); // always normalized

Maybe keeping it with integers to not to cast anything if we roll a dice?(randomUnsignedScalar<int>(1,6))or to see if that girl is single?(randomUnsignedScalar<int>(0,1)[bool to cast, yes. maybe type traits for numerics?])

mosra · 2020-04-08T15:30:42Z

src/Magnum/Math/Test/RandomTest.cpp

+    {
+        CORRADE_COMPARE(Math::Random::randomRotation().length(), 1.0f);
+    }
+}


It would be great to have some sort of distribution verification in the tests, to ensure we're not accidentally skewing the distribution to something non-uniform -- how good are your statistics skills? :)

Found this on SO, it suggests using a Chi Squared test. I never did such a thing myself tho :D

We have easter incoming ! :)

Well, I put a chi square test. In testing, I couldnt see a "tolerant" testing scheme(Chi square let 1 fail; like 99% ).

src/Magnum/Math/Test/RandomTest.cpp

mosra · 2020-04-16T11:21:06Z

Okay, great job! Everything seems to compile and pass tests except a few cases where

Value error_count is not less than 2, actual is 2 but expected < 2

I'm not sure how bad this is tho, seems to be depending on how good the standard implementation is? :)

I'll need some time to write the docs etc., but I think this can go in 🎉

sariug · 2020-04-16T18:32:14Z

Great news !

Value error_count is not less than 2, actual is 2 but expected < 2

I'm not sure how bad this is tho, seems to be depending on how good the standard implementation is? :)

So this happens in chisquare I guess, which the randomness in belong to std engine..
Let me know if i can support anyhow !

mosra · 2020-04-17T11:52:51Z

TODOs for me (adding a comment so I don't forget):

move <random> to a cpp file because it's heavy -- https://artificial-mind.net/projects/compile-health/
investigate what's up with std::seed_seq -- https://twitter.com/horenmar_ctu/status/1251044181558124544

sariug · 2020-08-23T11:13:28Z

Hey hey. Currently I was checking out Perlin noise(I've seen your "SmootherStep" :D ). Would you also like that to be implemented as an extension of this ?

Some noise here:

sariug added 2 commits April 5, 2020 22:04

Initial Random.h

46a026e

also some tests

8f73873

sariug changed the title ~~Math/Random amnd its tests implementation~~ Math/Random and its tests implementation Apr 6, 2020

Get rid of iosteam.

9de2987

mosra added this to the 2020.0a milestone Apr 8, 2020

mosra reviewed Apr 8, 2020

View reviewed changes

src/Magnum/Math/Test/RandomTest.cpp Outdated Show resolved Hide resolved

mosra reviewed Apr 8, 2020

View reviewed changes

src/Magnum/Math/Test/RandomTest.cpp Outdated Show resolved Hide resolved

sariug added 8 commits April 8, 2020 22:49

add repeated tests, use better macros!

282f3d5

Better quaternion, get rid of bool, put the links

9a0dffe

Merge branch 'master' of github.com:sariug/magnum

b201ab3

Mosra was right.

9337219

ChiSquare is implemented. seems pretty neat !

3196dac

according to travis not all the compilers like initializer lists.

7bcc433

public functions

c5e5804

no accumulate..MSVC dwants a header.

ac52472

mosra modified the milestones: 2020.0a, 2020.0b Jun 26, 2020

mosra mentioned this pull request Aug 22, 2020

Initial work on ssao mosra/magnum-examples#91

Open

Comments

Conversation

sariug commented Apr 6, 2020

Uh oh!

mosra left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sariug Apr 8, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

mosra commented Apr 16, 2020

Uh oh!

sariug commented Apr 16, 2020

Uh oh!

mosra commented Apr 17, 2020

Uh oh!

sariug commented Aug 23, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

sariug Apr 8, 2020 •

edited

Loading