22.01.2015 Views

Refined Buneman Trees

Refined Buneman Trees

Refined Buneman Trees

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Chapter 15<br />

Complexity<br />

This chapter deals with the time and space complexity of the RBT-algorithm.<br />

Before the advent of [BFÖ+ 03], a non-trivial algorithm for computing the RBT<br />

did exist ([BB99]), but with a running time of O(n 5 ) and space consumption<br />

O(n 4 ). The goals of [BFÖ+ 03] was to reduce these factors and make RBTs<br />

computationally competitive to methods based on neighbor joining and on plain<br />

<strong>Buneman</strong> trees, and in turn one of the goals of this thesis is to implement this<br />

algorithm and demonstrate how RBTs can be used in practice. Therefore we<br />

need to demonstrate that the implementation does indeed run in time O(n 3 )<br />

and space O(n 2 ).<br />

15.1 Running time<br />

To analyse the running time of the algorithm the author has written the test<br />

program which is parameterized by starting input size, ending input size and<br />

number of repetitions per input size. The test program measures the running<br />

time of the refined <strong>Buneman</strong> treee algorithm found by measuring the system<br />

time before and after each computation of the algorithm, excluding the initialization<br />

of a random input matrix. Input size and running time is reported for<br />

each repetition.<br />

The timing is not optimal; the running time does not reflect the exact running<br />

time of the algorithm, since the algorithm does not run exclusively on the<br />

test PC. However, the author made sure that the test PC was largely unused<br />

during testing, meaning that no users and only a few system processes were<br />

running during the time trials. Therefore the timing data might show a slightly<br />

skewed picture of the running time, but very importantly, the skew is evenly<br />

distributed and thus does not affect the test goal, with is to determine the<br />

asymptotic running time behaviour of the algorithm.<br />

The test was run with the following parameters:<br />

• Input start size 4<br />

81

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!