Algorithm Selection: From Meta-Learning to Hyper-Heuristics - Intelligent Systems - page 95

Where H i,j,l indicates the number of documents consistent with the query l , D i,j,l indicates the
length of the route to obtain the documents, i represented the current node and j is the node
chosen, and Z x is a measure of current performance. In this work the visibility states are: m 1
= (α > 1)&( TTL < D )&( TTL != 1), m 2 = (α > 1)&( TTL < D )&( TTL = 1), m 3 = ( H = 0) ||(( α >
1)&( TTL ≥ D ))|| (( α ≤ 1)&( TTL = 1)) and m 4 = ( α ≤ 1)&( TTL > 1). All the visibility states are
calculated to identify which heuristic will be applied to TTL.
5.2.2 Experimentation
The experimental environment used during experiments, and the results obtained are
presented in this section. Software: Microsoft Windows 7 Home Premium; Java programming
language, Java Platform, JDK 1.6; and integrated development, Eclipse 3.4. Hardware:
Computer equipment with processor Intel (R) Core (TM) i5 CPU M430 2.27 GHz and RAM
memory of 4 GB. Instances: It has 90 different SQRP instances; each of them consists of three
files that represent the topology, queries and repositories. The description of the features can
be found in (Cruz et al. 2008).
The average performance was studied by computing three performance measures of each
100 queries: Average hops , defined as the average amount of links traveled by a Forward
Ant until its death that is, reaching either the maximum amount of results required or
running out of TTL. Average hits , defined as the average number of resources found by
each Forward Ant until its death, and Average efficiency , defined as the average of
resources found per traversed edge (hits/hops). The initial Configuration of HH_AdaNAS
is shown in Table 2. The parameter values were based on values suggested of the literature
as (Dorigo & Stützle, 2004; Michlmayr, 2007; Aguirre, 2008 and Rivera, 2009).
In this section we show experimentally that HH_AdaNAS algorithm outperforms the
AdaNAS algorithm. Also HH_AdaNAS outperforms NAS (Aguirre, 2008), SemAnt
(Michlmayr, 2007) and random walk algorithms (Cruz et al., 2008), this was reported in
(Gómez et al., 2010), so HH_AdaNAS algorithm is positioned as the best of them.
Table 2. Shows the assignment of values for each HH_AdaNAS parameter.
In this experiment, we compare the HH_AdaNAS and AdaNAS algorithms. The
performance achieved is measured by the rate of found documents and the experiments
were conducted under equal conditions, so each algorithm was run 30 times per instance
and used the same configuration parameters for the two algorithms, which is described
in Table 2.