Tag: pseudorandom

Random number generation in game programming

Random number generators are extremely useful in game programming. I have found a lot of uses for randomness in my projects.

What can you do with randomness?

Man, all kinds of stuff. Nearly any value that you want to initialize in a game object is a candidate for possible randomization. Randomness fuzzes up your game, making less deterministic and therefore harder to defeat with simple patterns and more replayable.

Pretty much any time I would normally use a literal or a constant number in my code, anymore I step back and ask myself what range of values might work in that place, and then create a random function that will provide me with a number in that range. The only time I don’t do this is when I really do need a precise value, or when performance is too important to sacrifice the computation time needed for the random function to return its result.

Here are just a few ideas for how you can use random numbers to improve your games:

Unit stats

Nothing makes a video game feel more like a video game than when every enemy you encounter is an exact clone of all the other enemies that look like it. You can use randomness to give your enemies some personality by giving them randomized stats. Instead of fixed values for Attack, Defense, Speed, Damage, etc., use a random range of values to generate stronger and weaker versions of your enemies. It takes a little more time to compute these values on the fly, but modern processors can handle this load easily, unless you’re generating a huge number of units.

Sprite generation

Why spend a lot of time hand-drawing every sprite in your game? Create a generator system that randomly puts pieces together, and create random sprites on the fly. If you’ve played around with an avatar generator such as eightbit.me or the Mii generator on Wii or the XBox Live Arcade avatar generator, imagine that kind of model system, but with a random selector in charge of picking the hair, eyes, etc. You can do this to randomly generate other things, such as buildings, procedurally, as well.


If you’re calling drawing functions, randomizing colors can give your game a lot better visual appeal. If you’re clever in how you pick your random colors, you can come up with color schemes that work nicely, yet are always slightly different each time you play. You can either pre-define a palette of colors and randomly select one, or you can randomly select R, G, B or H, S, V numbers and create a color at runtime. You can experiment with different mathematical tweaks to shape and constrain the randomness.

Map generation

If you can write a good random map generator, you can save yourself from having to hand-design all your maps. GOOD random generation may be very difficult to accomplish, however — especially for more complex games. But even if you can’t guarantee a good random map at runtime, an almost-good random map generator can save you tons of time or spur your creativity by doing most of the work for you, leaving you with something almost good enough, that just needs a little hand-polishing to make shine.

Procedurally generated content in general is a good use for random functions. You can use a random number function to create a deterministic sequence of generated values that is always the same. This is because computer hardware actually does not have a means of creating a truly random number — it fakes it, approximating randomness with a pseudo-random algorithm.

This is used to good effect in one of my favorite Atari 2600 games, Pitfall. A pseudo-random function, using a fixed seed, is used to generate each screen in the game. This achieves a very high information density, since the data that was needed to represent each screen could not be stored on a 4kb ROM, but a generator function that creates that data easily could. This technique is not used very much in modern game development since storage isn’t much of an issue any more, but it is still a very interesting technique and one which merits study.


There are many potential applications of randomness to AI. Whenever your AI needs to make a decision, you potentially can use randomness to make that decision less predictable. Weighted probability is important here, as completely random AI behavior is erratic and seems crazy, while an AI that occasionally does something unexpected will seem tricky or deceptive or clever. Dynamically weighting the probability according to context at runtime will make your AI seem smarter.

GameMaker/GML random functions

These are built in to the Game Maker Language (GML):

  1. random(N): returns a random floating point value between 0 and N, not including N.
  2. random_range(A, B): returns a random a floating point value between A and B, not including B.
  3. irandom(N): returns a random integer value between 0 and N.
  4. irandom_range(A, B): returns a random integer value between A and B.
  5. choose(a,b,c,d,e,f,g,h,i,j,k…): Randomly returns one of the arguments, up to 16 arguments may be passed into a GML function. You can weight the likelihood of one result by repeating it in the arguments list. (e.g., choose(dog, dog, dog, cat) would be 3/4 likely to return dog, 1/4 likely to return cat.)
  6. random_get_seed(): Gets the current seed value for the randomizer.
  7. random_set_seed(): If you need a randomized, but deterministic function, you can set the seed for your random function. (This approach of setting a known seed is how the levels in Pitfall for Atari 2600 were always the same even though they were generated by the Atari’s random number generator.)
  8. randomize(): Sets the seed of the randomizer to a random value.

These are scripts you may import into your project:

  1. gauss(median, deviation): Returns a random value with a gaussian (“normal”) distribution around a median value. From GMLScripts.com.

All programming languages have similar random functions or classes built into them. Whatever tool you happen to be using, it pays to learn about how it can produce random numbers, and how you can use them to do useful and interesting things.

If you have any favorite ways of using random numbers in your programs, post a comment below and share it!

Keeping randomness under control

One of the mistakes most game developers will make when using random functions is to use too wide a range of random values, or failing to control the range of values returned by the random function.

Know your math

Randomness feels most random when the probability distribution is flat. However, this often does not make for the most interesting gameplay mechanics. It’s often more useful to have a weighted function that has a greater probability of returning a value in one part of the range than in another. Understanding probability math is key to getting your randomized functions under control. The other key is to develop your intuition to know what range of values will work best for a given situation.

If you’ve ever played tabletop role playing games, then you know about dice. Dice are good analog randomizers, and can help us understand probability and randomness in a computer program. In classic Dungeons & Dragons, character ability scores are randomly determined by adding the values of three six-sided dice. This results in a bell curve, meaning that the results of a 3d6 roll are distributed in such a way that an “average” score between 9-12 is far more likely than an extreme score of 3 or 18. So one way to directly simulate this type of dice rolling in a computer program would be :

N = 0;
repeat(3){N+=irandom_range(1,6)}; // generates a value between 3-18, distributed around 10.5

In computer programming, there are more efficient ways to achieve a bell curve distribution than this. Calling random() multiple times, and writing loops will make your code slow, so if there’s ways to avoid doing that, it’s a good idea. The gauss() function from gmlscrips.com creates a “normal” distribution around the agrument passed into it, and is fast and efficient.

round(gauss(10.5, 3.5)); //simulates rolling 3d6, approximately

Note that this will not return exactly the same distribution of values as a true 3d6 roll will. But this is because 3d6 is actually an approximation of a gaussian distribution — the gauss() distribution is more accurate to a “standard normal” statistical distribution. If you compared graphs of the bell curves of 3d6 vs. the gauss() function, the gauss curve would be smoother, and would include values outside the 3-18 range (2 and 19 would show up a tiny percent of the time).

There are other types of distibutions that you might want to achieve with your randomized functions, for some purpose. Knowing your math is important here. Learn the graphs of common functions, and understand the relationship between the shape of the graph and the probability distribution of a randomized function modified by each function. For example, random(x), ln(random(x)), and random(x)^2 have very different looking distributions. Knowing this, you can tailor an equation to fit your needs.

Once you get comfortable with the math, it’s actually fun. Play around with a graphic calculator and see what different graphs you can come up with. Each time you discover an interesting or useful shape, make note and file it away for a time when it might be useful.


Because adding randomness to your functions make the game non-deterministic, it can make things more difficult to test. Certain conditions become hard to duplicate, because you don’t directly control them, and this can make repeatable testing of your game seem impossible.

There are approaches you can take to ensure that your code works, still. First, when you are building up your functions, ensure that the non-random parts of the code work well before you introduce randomness. If necessary, temporarily remove the randomization and replace it with a literal value, a constant, or a variable. Once you are have tested thoroughly and are sure the code is working correctly with a range of controlled values, you may safely replace the controlled values with random values that are constrained to the ranges you tested.

In some cases, you may need to go back and re-test code. It would be a pain to have to find/replace every random() call in your code. Not only would it be time consuming, it would increase the opportunity for errors to creep in. A better approach may be to comment out the equivalent non-random code next to the random code, and leave it in your code file. That way you just have to comment out the randomized function and un-comment the deterministic version.

Even this is time consuming and error prone, however. You may want to create randomized and non-random versions of your functions, and introduce a configuration variable that you can toggle to enable/disable randomness. Then you can pepper your code with things like:

if settings.random {randomized_function()} else {deterministic_function()};

All this extra branching in your code can get ugly and unmangeabeable too, so try to limit this to keep it to a minimum.

You could also use polymorphism to create sibling classes, one which inherits randomized functions, and one which doesn’t, and just spawn the appropriate class instances according to the game configuration at runtime.

The great thing about being able to turn randomness on or off at runtime is that it allows you to very quickly see the difference, reducing the lag time between test runs. This could even turn into a feature of the game, rather than a debug exercise.

With proper care taken during development, randomized functions can be just as reliable as deterministic functions. It just takes a little extra forethought and planning.