Procedural generation strategies

Procedural generation strategies
--------------------------------

Before deciding what algorithm to use, you need to remember that there probably isn't a perfect algorithm. There rarely is. So whatever you choose it will have advantages and disadvantages. You have to think in terms of trade-offs.

Typical trade-offs are:
Run-time;
Memory usage;
Quality of solution;
Variety of solutions;
Work time (To design+write your algorithm).

--------------------------------
[1] Generate a random solution.

- Predictable fast run-time.
- The solution could be good or bad. Expect bad.

--------------------------------
[2] Generate random solutions over and over, until you find one "good enough" that passes some tolerance.

- It will find a good solution, providing the search terminates.
- Unpredictable run-time. May not terminate at all if the tolerance is set too high.

--------------------------------
[3] Generate a fixed number of random solutions, and pick the best of those.

- Predictable run-time. Could be fast or slow, depending on the number chosen.
- Not guaranteed to find a good solution every time. It might even be a terrible solution some times.

--------------------------------
[4] Generate a random solution, then iteratively modify it to get better solutions.

- Should be fast.
- May find a "local maximum", whereby your modifications won't improve it, because the much better solutions are too different from what you ended up with.

--------------------------------
[5] Solve the problem on paper, and hand-pick solutions you like.

- Very fast run-time, as the computer is just picking a random thing from a list.
- Potentially good solutions, because you've hand-picked them.
- Lacks variation, as the solutions are limited to the few you've chosen.
- Depending on the nature of the problem, it could use a lot of memory to store all these picked solution. It might be a lot of work.


[...]