Training the Classifiers - Design and Analysis of Learning Classifier Systems - page 91

Weight and noise estimation error for sinusoid over [0, pi/2]

0.05 5e-05

NLMS weight
RLS weight
NLMS noise
RLSLMS noise
RLS noise

0.04 4e-05

0.03 3e-05

0.02 2e-05

0.01 1e-05

0 0

0 10 20 30 40 50

Observations

Weight and noise estimation error for sinusoid over [pi/2, pi]

0.4 0.005

0.35

0.004

0.3

0.25

0.003

0.2

0.002

0.15

0.1

NLMS weight
RLS weight
NLMS noise
RLSLMS noise
RLS noise

0.001

0.05

0

0

0 50 100 150 200 250 300

Observations

Fig. 5.2. The graphs show the MSE of the weight vector estimate (on the left scale)
and squared noise variance estimate error (on the right scale) of different classifiers
when approximating a sinusoid. The classifiers are presented with input x n =(1 ,i n ) T
and output y n =sin( i n ). In the upper graph, the sinusoid was sampled from the range
i n ∈ [0 ,π/ 2], and in the lower graph the samples are taken from the range i n ∈ [ π/ 2 ,π ].
The MSE of the weight vector estimate for the RLSLMS classifier is not show, as it is
equivalent to the MSE of the RLS classifier.