GA-LMS standard with multivector entries (Python 2.7)

The following Python 2.7 script calls the GA-LMS (standard) binary to perform a system identification task.

In this example, the multivectors samples belong to the Geometric Algebra of $\mathbb{R}^3$, namely $\mathcal{G}(\mathbb{R}^3)$. Thus, each regressor and weight vector entry has 8 coefficients, i.e., for each entry of the weight vector, 8 coefficients have to be estimated. This constrasts with the usal LMS which only estimates real/complex entries. For further information, please refer to the GA documentation at www.openga.org.

At the end, the learning curves Mean-Square Error (MSE) and/or Excess Mean-Square Error (EMSE) are plotted.

Start by importing the necessary Python modules:

In [14]:
# Script to call the GAAFs binaries.
# Case: GA-LMS standard with multivector entries
#
# Wilder Lopes - wil@openga.org
# Dec 2015

import sys, string, os
import matplotlib.pyplot as plt
import numpy as np
from math import log

The user is able to set the following AF parameters:

Number of filter taps (system order): M

Realizations: L

Time iterations: N

AF step size: mu

Measurement Noise variance: sigma2v

In [15]:
# Simulation parameters
M       = 5 # System order
L       = 100 # Realizations
N       = 1000 # Time iterations
mu      = 0.005 # AF Step size
sigma2v = 1e-3 # Variance of measurement noise

The binary is called below using the previously set parameters. The GA-LMS runs and returns .txt files with the results: _galms.out and _theory.out, where represents "MSE" or "EMSE". _galms.out files store the ensemble-average learning curves (EMSE), while *_theory.out files store the theoretical steady-state value for MSE and EMSE.

In [16]:
# Calling binary
arguments = " " + str(M) + " " + str(L) + " " + str(N) + " " + str(mu)+ \
        " " + str(sigma2v)
os.system("../GAAFs_Standard/GA-LMS/build/GA-LMS" + arguments)
Out[16]:
0

Show final estimate for weight array:

In [17]:
w_galms = open('w_galms.out', 'r')
print 'Final weight array in GAALET notation:' + '\n'
for line in w_galms:
    print line

Final weight array in GAALET notation:

[ 0.549881 -0.000330202 1.00007 2.00002 0.709914 1.2999 4.49985 3.00041 ] { 0 1 2 3 4 5 6 7 }

[ 0.550185 1.05796e-05 0.999707 1.99969 0.71029 1.29997 4.50005 3.00003 ] { 0 1 2 3 4 5 6 7 }

[ 0.550135 -0.000194987 1.00016 2.00029 0.710378 1.29994 4.50008 3 ] { 0 1 2 3 4 5 6 7 }

[ 0.549849 -0.000100298 1.00004 2.0001 0.710213 1.29985 4.50025 3.00012 ] { 0 1 2 3 4 5 6 7 }

[ 0.550053 -7.70494e-05 0.999982 2.00001 0.710416 1.30031 4.50018 2.99979 ] { 0 1 2 3 4 5 6 7 }

Load file MSE_galms.out and MSE_theory.out to plot MSE learning curve and theoretical curve:

In [19]:
f1 = open('MSE_galms.out', 'r')
data_label1 = ['MSE_galms']
data1_list = []
for line in f1:
    data1_list.append(line.rstrip('\n'))

f2 = open('MSE_theory.out', 'r')
data_label2 = ['MSE_theory']
data2_list = []
for line in f2:
    for i in range(len(data1_list)):
        data2_list.append(line.rstrip('\n'))

data1 = [float(j) for j in data1_list] # Converts to float
data2 = [float(j) for j in data2_list] # Converts to float
data1_dB = [10*log(x,10) for x in data1]
data2_dB = [10*log(x,10) for x in data2]

plt.title('MSE curves')
plt.ylabel('MSE (dB)')
plt.xlabel('Iterations')
plt.plot(data1_dB, label = 'MSE_galms', color = 'blue')
plt.plot(data2_dB, label = 'MSE_theory', color = 'magenta')
plt.legend()
plt.show()

Load file EMSE_galms.out and EMSE_theory.out to plot EMSE learning curve and theoretical curve:

In [20]:
f1 = open('EMSE_galms.out', 'r')
data_label1 = ['EMSE_galms']
data1_list = []
for line in f1:
    data1_list.append(line.rstrip('\n'))

f2 = open('EMSE_theory.out', 'r')
data_label2 = ['EMSE_theory']
data2_list = []
for line in f2:
    for i in range(len(data1_list)):
        data2_list.append(line.rstrip('\n'))

data1 = [float(j) for j in data1_list] # Converts to float
data2 = [float(j) for j in data2_list] # Converts to float
data1_dB = [10*log(x,10) for x in data1]
data2_dB = [10*log(x,10) for x in data2]

plt.title('EMSE curves')
plt.ylabel('EMSE (dB)')
plt.xlabel('Iterations')
plt.plot(data1_dB, label = 'EMSE_galms', color = 'r')
plt.plot(data2_dB, label = 'EMSE_theory', color = 'magenta')
plt.legend()
plt.show()