def num_dims_vs_time(o_algorithm, settings, o_function, num_dims, plot=False):
algorithm_name = o_algorithm.get_name()
func_name = o_function.func_globals['__name__'].replace('_', ' ').title()
times = []
accuracy = []
optimize_settings(settings)
for test in num_dims:
# create bounds
bounds = []
for i in range(0, test):
bounds.append((-10, 10))
settings['number_of_dimensions'] = test
settings['bounds'] = bounds
algorithm = o_algorithm(settings, o_function)
algorithm.start_timer()
algorithm.run()
algorithm.stop_timer()
times.append(algorithm.get_time())
accuracy.append(algorithm.get_best_x().get_fval())
if plot:
# timing plot
time_fig = plt.figure()
time_ax = time_fig.add_subplot(111)
plt.title("%s Number of Dimensions vs Time on %s" %
(algorithm_name, func_name))
time_ax.set_xlabel("Number of Dimensions")
time_ax.set_ylabel("Time (seconds)")
time_ax.plot(num_dims, times, 'g-')
# accuracy plot
acc_fig = plt.figure()
acc_ax = acc_fig.add_subplot(111)
plt.title("%s Number of Dimensions vs Accuracy on %s" %
(algorithm_name, func_name))
acc_ax.set_xlabel("Number of Dimensions")
acc_ax.set_ylabel("Objective Function Value")
acc_ax.plot(num_dims, accuracy, 'g-')
plt.show()
return (times, accuracy)
def cmp_num_parts_vs_time(o_algorithm1, o_algorithm2, \
settings1, settings2, o_function, num_particles):
func_name = o_function.func_globals['__name__'].replace('_', ' ').title()
times1 = []
times2 = []
accuracy1 = []
accuracy2 = []
optimize_settings(settings1)
optimize_settings(settings2)
for test in num_particles:
t, acc = num_parts_vs_time(o_algorithm1, settings1, o_function, [test],
False)
times1.append(t)
accuracy1.append(acc)
t, acc = num_parts_vs_time(o_algorithm2, settings2, o_function, [test],
False)
times2.append(t)
accuracy2.append(acc)
# timing plot
time_fig = plt.figure()
time_ax = time_fig.add_subplot(111)
time_ax.set_title("GA vs. PSO timing on %s" % func_name)
time_ax.set_xlabel("Number of Particles")
time_ax.set_ylabel("Time (seconds)")
time_ax.plot(num_particles, times1, 'g-')
time_ax.plot(num_particles, times2, 'r-')
time_ax.legend(['GA', 'PSO'])
# accuracy plot
acc_fig = plt.figure()
acc_ax = acc_fig.add_subplot(111)
acc_ax.set_title("GA vs. PSO accuracy on %s" % func_name)
acc_ax.set_xlabel("Number of Particles")
acc_ax.set_ylabel("Objective Function Value")
acc_ax.plot(num_particles, accuracy1, 'g-')
acc_ax.plot(num_particles, accuracy2, 'r-')
acc_ax.legend(['GA', 'PSO'])
plt.ylim(0, 1)
plt.show()
return (times1, times2)
def num_parts_vs_time(o_algorithm,
settings,
o_function,
num_particles,
plot=False):
algorithm_name = o_algorithm.get_name()
func_name = o_function.func_globals['__name__'].replace('_', ' ').title()
times = []
accuracy = []
optimize_settings(settings)
settings['num_iterations'] = 100
for test in num_particles:
settings['population_size'] = test
algorithm = o_algorithm(settings, o_function)
algorithm.start_timer()
algorithm.run()
algorithm.stop_timer()
times.append(algorithm.get_time())
accuracy.append(algorithm.get_best_x().get_fval())
if plot:
plt.title("%s Number of Particles vs Time on %s" %
(algorithm_name, func_name))
plt.xlabel("Number of Particles")
plt.ylabel("Time (Seconds)")
plt.plot(num_particles, times, 'r-')
plt.show()
return (times, accuracy)
def get_pso_two_d_accuracy(o_algorithm, o_settings, o_function, \
x1_start, x1_step, x1_end, \
x2_start, x2_step, x2_end, \
x1_name, x2_name, \
population_size=50, num_tests_per_point=10, plot=True, \
save_histograms=True, response_surface=True, debug=False):
if response_surface:
plot = True
# turn off settings that slow us down
o_settings = optimize_settings(o_settings)
func_name = o_function.func_globals['__name__']
tests = {}
hist_num_bins = 150
o_settings['population_size'] = population_size
num_tests_per = num_tests_per_point
x1_srt = x1_start
x1_e = x1_end
x1_stp = x1_step
x2_srt = x2_start
x2_e = x2_end
x2_stp = x2_step
num_tests = int(( (int(100*x1_e)-int(100*x1_srt))/int(100*x1_stp) + 1 )* \
( (int(100*x2_e)-int(100*x2_srt))/int(100*x2_stp) + 1 ))
X = np.ones(shape=(num_tests, 6))
y = np.zeros(shape=(num_tests, 1))
n = 0
if plot:
fig = plt.figure()
ax1 = fig.add_subplot(111, projection='3d')
if o_function != griewank_function.objective_function:
ax1.set_zlim(0, 1)
for i in np.arange(x1_srt, x1_e + x1_stp, x1_stp):
for j in np.arange(x2_srt, x2_e + x2_stp, x2_stp):
# set settings for this test
o_settings[x1_name] = i
o_settings[x2_name] = j
# initial variables
values = []
euclid_distance = []
test_name = 'cp' + '(' + str(o_settings['cp']) + ')' + ',' + \
'cg' + '(' + str(o_settings['cg']) + ')'
print("Running test %s on %s" % (test_name, func_name))
# create histogram plot if true
if save_histograms:
hist_fig = plt.figure()
hist_ax = hist_fig.add_subplot(111)
# run optimization algorithm
for k in range(0, num_tests_per):
algorithm = o_algorithm(o_settings, o_function)
algorithm.run()
# save enf values
values.append(algorithm.get_best_x().get_fval())
# euclidean distance
squares = 0
for pos in algorithm.get_best_x().pos:
if o_function == rosenbrock_function.objective_function:
squares += (pos - 1)**2
elif o_function == easom_function.objective_function:
squares += (pos - np.pi)**2
else:
squares += pos**2
euclid_distance.append(np.sqrt(squares))
# save histogram if true
if save_histograms:
hist_ax.hist(euclid_distance,
hist_num_bins,
range=(0, 9),
normed=True)
hist_fig.savefig(test_name + '.png')
plt.close(hist_fig)
# find average and save data
#avg = sum(values)/len(values)
#avg = median(values)
#avg = sum(euclid_distance)/len(euclid_distance)
avg = np.median(euclid_distance)
tests[test_name] = avg
if plot:
ax1.scatter(i, j, avg)
X[n][1] = i
X[n][2] = j
X[n][3] = i * i
X[n][4] = i * j
X[n][5] = j * j
y[n] = avg
# increment test number
n += 1
# fname = gen_filename(x1_name, x2_name, func_name)
# write_xy_data(X, y, fname)
if debug:
print("\n*** DATA ***")
print("X")
print(X)
print("\ny")
print(y)
print("\ntests")
print(tests)
if response_surface:
# get regression coefficients
b = regression_utils.get_regression_coef(X, y)
pltx = np.arange(x1_start, x1_end + x1_step, x1_step)
plty = np.arange(x2_start, x2_end + x2_step, x2_step)
pltX, pltY = np.meshgrid(pltx, plty)
F = b[0] + b[1] * pltX + b[2] * pltY + b[3] * pltX * pltX + b[
4] * pltX * pltY + b[5] * pltY * pltY
ax1.plot_wireframe(pltX, pltY, F)
if plot:
print("\nPlotting ...")
x1_name = x1_name[0].upper() + x1_name[1:]
x2_name = x2_name[0].upper() + x2_name[1:]
ax1.set_xlabel(x1_name)
ax1.set_ylabel(x2_name)
ax1.set_zlabel('Average Euclidean Distance from Global Minimum')
plt.show()
return (X, y)
def get_3d_accuracy(o_algorithm, o_settings, o_function, \
x1_info, x2_info, x3_info, \
population_size=50, num_tests_per_point=10, \
save_histograms=True, debug=False):
# turn off settings that slow us down
o_settings = optimize_settings(o_settings)
func_name = o_function.func_globals['__name__']
tests = {}
hist_num_bins = 150
o_settings['population_size'] = population_size
num_tests_per = num_tests_per_point
x1_name = x1_info[NAME]
x2_name = x2_info[NAME]
x3_name = x3_info[NAME]
x1_s = x1_info[START]
x1_e = x1_info[END]
x1_stp = x1_info[STEP]
x2_s = x2_info[START]
x2_e = x2_info[END]
x2_stp = x2_info[STEP]
x3_s = x3_info[START]
x3_e = x3_info[END]
x3_stp = x3_info[STEP]
num_tests = int(( (int(100*x1_e)-int(100*x1_s))/int(100*x1_stp) + 1 )* \
( (int(100*x2_e)-int(100*x2_s))/int(100*x2_stp) + 1 )* \
( (int(100*x3_e)-int(100*x3_s))/int(100*x3_stp) + 1 ) )
X = np.ones(shape=(num_tests, 10))
y = np.zeros(shape=(num_tests, 1))
n = 0
for i in np.arange(x1_s, x1_e + x1_stp, x1_stp):
for j in np.arange(x2_s, x2_e + x2_stp, x2_stp):
for k in np.arange(x3_s, x3_e + x3_stp, x3_stp):
# set settings for this test
o_settings[x1_name] = i
o_settings[x2_name] = j
o_settings[x3_name] = k
# initial variables
values = []
euclid_distance = []
test_name = x1_name + '(' + str(i) + ')' + ',' + \
x2_name + '(' + str(j) + ')' + ',' + \
x3_name + '(' + str(k) + ')'
print("Running test %s" % test_name)
# create histogram plot if true
if save_histograms:
hist_fig = plt.figure()
hist_ax = hist_fig.add_subplot(111)
# run optimization algorithm
for t in range(0, num_tests_per):
algorithm = o_algorithm(o_settings, o_function)
algorithm.run()
# save enf values
values.append(algorithm.get_best_x().get_fval())
# euclidean distance
squares = 0
for pos in algorithm.get_best_x().pos:
if o_function == rosenbrock_function.objective_function:
squares += (pos - 1)**2
elif o_function == easom_function.objective_function:
squares += (pos - np.pi)**2
else:
squares += pos**2
euclid_distance.append(np.sqrt(squares))
# save histogram if true
if save_histograms:
hist_ax.hist(values,
hist_num_bins,
range=(0, 1.5),
normed=True)
hist_fig.savefig(test_name + '.png')
plt.close(hist_fig)
# find average and save data
#avg = sum(values)/len(values)
#avg = median(values)
#avg = sum(euclid_distance)/len(euclid_distance)
avg = np.median(euclid_distance)
tests[test_name] = avg
X[n][1] = i
X[n][2] = j
X[n][3] = k
X[n][4] = i * i
X[n][5] = i * j
X[n][6] = i * k
X[n][7] = j * j
X[n][8] = j * k
X[n][9] = k * k
y[n] = avg
# increment test number
n += 1
fname = gen_filename(x1_name, x2_name, func_name)
write_xy_data(X, y, 'ga_3d_data.dat')
if debug:
print("\n*** DATA ***")
print("X")
print(X)
print("\ny")
print(y)
print("\ntests")
print(tests)
return (X, y)
def cmp_func_val_over_iterations(o_algorithm, settings, o_function):
x1_start = 0.3
x1_step = 0.1
x1_end = 0.6
x2_start = 0.25
x2_step = 0.25
x2_end = 0.75
x1_name = "selection_cutoff"
x2_name = "mutation_rate"
population_size = [50]
optimize_settings(settings)
fig, (ax1, ax2, ax3) = plt.subplots(3, sharex=True, sharey=True)
tests1 = []
tests2 = []
tests3 = []
for test in population_size:
for i, x1 in enumerate(np.arange(x1_start, x1_end + x1_step, x1_step)):
for j, x2 in enumerate(
np.arange(x2_start, x2_end + x2_step, x2_step)):
settings[x1_name] = x1
settings[x2_name] = x2
f = []
settings['population_size'] = test
algorithm = o_algorithm(settings, o_function)
while settings['num_iterations'] > algorithm.num_generations:
f.append(algorithm.get_best_f())
algorithm.do_loop()
if j == 0:
tests1.append(
"Selection Cutoff %4.2f Mutation Rate %4.2f" %
(x1, x2))
ax1.plot(range(1, len(f) + 1), f)
elif j == 1:
tests2.append(
"Selection Cutoff %4.2f Mutation Rate %4.2f" %
(x1, x2))
ax2.plot(range(1, len(f) + 1), f)
elif j == 2:
tests3.append(
"Selection Cutoff %4.2f Mutation Rate %4.2f" %
(x1, x2))
ax3.plot(range(1, len(f) + 1), f)
ax1.legend(tests1)
ax2.legend(tests2)
ax3.legend(tests3)
ax1.set_title(
'GA Comparison of Selection Cutoff & Mutation Rate on Ackley Function (50 particles)'
)
ax1.set_xlabel('Number of Iterations')
ax2.set_xlabel('Number of Iterations')
ax3.set_xlabel('Number of Iterations')
ax1.set_ylabel('Objective Function Value')
ax2.set_ylabel('Objective Function Value')
ax3.set_ylabel('Objective Function Value')
#ax2.ylabel('Objective Function Value')
#ax3.ylabel('Objective Function Value')
#plt.legend(tests)
plt.show()
def cmp_func_val_over_iterations(o_algorithm, settings, o_function):
x1_start = 0.25
x1_step = 0.25
x1_end = 0.75
x2_start = 0.25
x2_step = 0.25
x2_end = 0.75
x1_name = "cp"
x2_name = "cg"
varient = 'normal'
population_size = [50]
optimize_settings(settings)
fig, (ax1, ax2, ax3) = plt.subplots(3, sharex=True, sharey=True)
settings['velocity_type'] = varient
tests1 = []
tests2 = []
tests3 = []
for test in population_size:
for i, x1 in enumerate(np.arange(x1_start, x1_end+x1_step, x1_step)):
for j, x2 in enumerate(np.arange(x2_start, x2_end+x2_step, x2_step)):
settings[x1_name] = x1
settings[x2_name] = x2
f = []
settings['population_size'] = test
algorithm = o_algorithm(settings, o_function)
while settings['num_iterations'] > algorithm.num_iterations:
f.append(algorithm.get_best_f())
algorithm.do_loop()
if j == 0:
tests1.append("Cp %4.2f Cg %4.2f" % (x1, x2))
ax1.plot(range(1,len(f)+1), f)
elif j == 1:
tests2.append("Cp %4.2f Cg %4.2f" % (x1, x2))
ax2.plot(range(1,len(f)+1), f)
elif j == 2:
tests3.append("Cp %4.2f Cg %4.2f" % (x1, x2))
ax3.plot(range(1,len(f)+1), f)
ax1.legend(tests1)
ax2.legend(tests2)
ax3.legend(tests3)
varient = varient[0].upper() + varient[1:]
ax1.set_title(varient + ' PSO Comparison of Cp & Cg on Easom Function (50 particles)')
ax1.set_xlabel('Number of Iterations')
ax2.set_xlabel('Number of Iterations')
ax3.set_xlabel('Number of Iterations')
ax1.set_ylabel('Objective Function Value')
ax2.set_ylabel('Objective Function Value')
ax3.set_ylabel('Objective Function Value')
#ax2.ylabel('Objective Function Value')
#ax3.ylabel('Objective Function Value')
#plt.legend(tests)
plt.show()