An Experimental Evaluation for Asynchronous Concurrent Systems

Download Report

Transcript An Experimental Evaluation for Asynchronous Concurrent Systems

Heuristics for Minimum Brauer Chain Problem
Fatih Gelgi
Melih Onus
Outline

Problem Definition
 Binary Method
 Factor Method
 Heuristics
 Experimental Results
 Conclusions
Brauer Chain


A Brauer chain for a positive integer n is a sequence
of integers 1 = a0, a1, a2, …, ar = n such that ai = ai-1 +
ak for some 0 ≤ k < i and 1 ≤ i ≤ n
Example:
1
1, 1+1=2
1, 2, 2+2=4
1, 2, 4, 4+2=6
1, 2, 4, 6, 6+6=12
1, 2, 4, 6, 12, 12+2=14
is a Brauer chain for 14 with length 5
Binary Method




Write the number n in binary form
Replace each 1 with DA and each 0 with D
Remove the leading DA
Begin from 1, follow the sequence from left to right
– For each D, double the current number
– For each A, add 1 to the current number

Example:
– Binary representation of 19 is 10011
– Sequence is DDDADA
– 1, 1+1=2, 2+2=4, 4+4=8, 8+1=9, 9+9=18, 18+1=19
Factor Method





Let n = p*q where n, p, qZ+. First calculate Brauer
chain for p and q
Let 1 = a1, a2, …, ak = p is Brauer Chain for p
Let 1 = b1, b2, …, bk = q is Brauer Chain for q
1 = a1, a2, …, ak = p = p*b1, p*b2, …, p*bm = p*q is a
Brauer chain for n=p*q
Example:
– <1, 2 ,4, 5> <1, 2, 3>
– <1, 2 ,4, 5, 10, 15>
Heuristics

Binary Heuristic
 Factorization Heuristic
 Dynamic Heuristic:
– It uses the previous solutions to obtain the best solution for
current n value
– We store only one solution for each number
– The dynamic formula is, l(n) = min{l(k)+1} where k<n and the
solution sequence of k must contain n-k

2-3-5 Heuristic
– This algorithm always begins with numbers 1, 2, 3, 5, ….
– For the next element, it selects as maximum as possible


2-3-6 Heuristic
2-4-8 Heuristic
Experiments



We did 3 types of experiments
– We calculated all n values up to about 4500
– We calculated only one randomly chosen n value
within each interval of 200 up to 10000
– We calculated all n values up to 20000 (using
factorization and dynamic heuristics)
To calculate optimum value, we used exhaustive
search with branch and bound technique
Although our pruning conditions make the search
quite faster, we were able to calculate the optimum
values up to 4500 since the running time is O(n!)
Optimum and 1.5 optimum values
for n up to 4500
The performance of heuristics:
Experiment II


To obtain optimum values for larger n’s and increase
the quality of our experiments, for each interval with
size 200 we chose a random n value
All the solutions of heuristics are clearly smaller than
1.5 optimum
Experiment II
Approximation Ratios




All the heuristics obviously has 1.5 approximation ratios
Better than 2-4-8, 2-3-6, 2-3-5 and binary, factorization
has 1.25 approximation value
Dynamic has an incredible approximation ratio with 1.1
Empirically, factorization is even smaller than 1.5 lg n
and dynamic is smaller than 1.4 lg n
Conclusion






Several heuristics for approximating minimum Brauer chain
problem is discussed
The optimum function is not monotone we couldn’t prove a
theoretical approximation ratio better than 2
Experimental results show that there is empirically an
approximation with 1.1 which is incredible for the problem
For approximation results, we also achieved 1.4 lg n length for
any number n where the trivial lower bound is lg n
Providing a better lower bound, the approximation factor can be
decreased
With a good lower bound, approximation factor can be proved
theoretically