Truth Table Data Structure
I am throwing some ideas around my head, but I want some feedback from the GameDev community.
How would you go about implementing a 'truth table' in C++? i.e. what data structure would you use?
Basically if I have some graph (nodes and directed edges), each node in the graph will have an associated truth table. The number of entries in the node's table is a function of the # of parents of the node (entries = 2^numParents).
For example:
0 parents = 1 entry = t
1 parents = 2 entries = t / f
2 parents = 4 entries = tt / tf / ft / ff
.. etc.
Given the node, you will of course know the values of it's parents. These parent values are used to index into the current node's truth table. It gets tricky, though, because the # of indices used to index into the truth table equals the number of parents of that node. Because of this uniqueness to the problem, you can't use a simple associative array.
This is essentially a Bayesian Network. It's so common I'm wondering how most applications go about implementing this.
Thanks~
In this case you could use a single dimensional array of size 2^N.
You should then be able to use the predecessors' true/false as bits of the index into the array:
This should also account for the "no predecessors" case where your index is always 0 (the first index).
Is that sort of what you wanted? The major downside of this approach is that it will use an astronomical amount of memory if you get a lot of predecessors.
If you have a large number of predecessors but only a couple entries in the truth table are true (or only a couple are false) then you would be better off with a set of bit-patterns that represent the "rare" state and assume everything else is the "default" state.
T *truthTable = new T[1 << numPredecessors];
You should then be able to use the predecessors' true/false as bits of the index into the array:
int index = 0;for (int i=0; i<numPredecessors; ++i) index |= (predecessors->IsTrue() ? 1 : 0) << i;
This should also account for the "no predecessors" case where your index is always 0 (the first index).
Is that sort of what you wanted? The major downside of this approach is that it will use an astronomical amount of memory if you get a lot of predecessors.
If you have a large number of predecessors but only a couple entries in the truth table are true (or only a couple are false) then you would be better off with a set of bit-patterns that represent the "rare" state and assume everything else is the "default" state.
... I started trying to write this and it wasn't much fun. But probably something with boost::multi_array<bool, N> and type erasure techniques (so that you don't have to know the number of dimensions ahead of time).
Edit: Of course, that's assuming you demand a pretty, iterative-subscripting interface. If you're willing to pack bits together, then just indexing into 2^N bit field is just fine. (Consider using a std::vector<std::bitset<32> > for example, and doing some shifting to figure out a vector index and a position within the bitset.)
Edit: Of course, that's assuming you demand a pretty, iterative-subscripting interface. If you're willing to pack bits together, then just indexing into 2^N bit field is just fine. (Consider using a std::vector<std::bitset<32> > for example, and doing some shifting to figure out a vector index and a position within the bitset.)
Thanks for the help!
The one thing I wasn't clear on, I guess, was that each entry in the truth table holds some float value (not a bool). So I guess I should have just called it a Bayesian Network instead of a truth table :) There is no way around the fact that a node with a lot of parents will use up a lot of memory storing these table entries.
Based on the above posts, I have come up with a solution, using an array indexing scheme. I can allocate the correct # of required entries for the table, and then use an indexing algorithm to index into the array.
Thanks again.
Quote:Original post by Nypyren
Is that sort of what you wanted? The major downside of this approach is that it will use an astronomical amount of memory if you get a lot of predecessors.
If you have a large number of predecessors but only a couple entries in the truth table are true (or only a couple are false) then you would be better off with a set of bit-patterns that represent the "rare" state and assume everything else is the "default" state.
The one thing I wasn't clear on, I guess, was that each entry in the truth table holds some float value (not a bool). So I guess I should have just called it a Bayesian Network instead of a truth table :) There is no way around the fact that a node with a lot of parents will use up a lot of memory storing these table entries.
Based on the above posts, I have come up with a solution, using an array indexing scheme. I can allocate the correct # of required entries for the table, and then use an indexing algorithm to index into the array.
Thanks again.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement