Question 1)
I have a large sparse vector of doubles in c++, I need to efficiently parse out the indices of the non zero elements from the vector. I can obviously loop over the length and do it, any better way to do it?
Asked
Active
Viewed 1,953 times
-2
-
1[What have you tried?](http://whathaveyoutried.com) – Andy Prowl Feb 17 '13 at 17:45
-
If you have a second question, make a second post. – Benjamin Lindley Feb 17 '13 at 17:45
-
I mention it right, its trivial to loop over it and solve. is there a better way ? – ganesh reddy Feb 17 '13 at 17:52
-
There's no standard sparse vector in C++. It is unclear what you mean. Do you have some data structure (which, and what is meant by "parsing out")? A data file (in which format, and what kind of data structure you want to create?) Something else? – n. m. could be an AI Feb 17 '13 at 18:02
-
sorry sparse means that the vector say is 200 long, but mostly is 0's – ganesh reddy Feb 17 '13 at 18:04
3 Answers
3
Unless you have some special knowledge of the makeup of the vector of doubles, (for example, it's sorted), a loop over its entirety is the most efficient you're gonna get.
Of course, a change in structure as suggested by eladidan is probably something you should consider.
Benjamin Lindley
- 101,917
- 9
- 204
- 274
3
I have a large sparse vector of doubles in c++, I need to efficiently parse out the indices of the non zero elements from the vector. I can obviously loop over the length and do it, any better way to do it?
If the vector is truly sparse (n = o(N) where n is the number of non-zero elements and N is the size of the vector), then representing it in an std::map<int,double> or std::unordered_map<int,double> is probably best. With std::mapway you get to find an element in O(log(n)) . With std::unordered_map a find operation takes amortized time of O(1). In both cases, the number of non-zero elements is simply the size of the container. Both approaches also take O(n) space instead of O(N).
eladidan
- 2,634
- 2
- 26
- 39
0
If you cannot change the representation of your data, you have to examine each element to filter out those that are almost equal to zero. However, this task is embarrassingly parallel, so maybe you can partition the workload to a bunch of threads to at least improve the runtime (although not the complexity).
