def ga_gp_rw_experiment(path_results):
uniformify_vals = [0.1, 0.5, .75, 0.9, 1.0, 1.0/0.9, 1.0/.75, 2.0, 10.0]
results = OrderedDict()
ga_length = 10
gp_depth = 2
tp_path = os.path.join(path_results, "ga_length_10", "TP.dat")
ga_tp = np.genfromtxt(tp_path)
tp_path = os.path.join(path_results, "depth_2", "TP.dat")
gp_tp = np.genfromtxt(tp_path)
print "doing probability of encounter experiment"
# Do the "probability of encounter" experiment first
reps = 100
steps = 50
fit_path = os.path.join(path_results, "ga_length_10", "fitness_vals.dat")
ga_fit = np.genfromtxt(fit_path)
gp_fit = [float(s) for s in
open(os.path.join(path_results, "depth_2", "all_fitness_values.dat")).readlines()]
inds = 0, len(gp_fit)-1
hc_encounters = [0.0, 0.0]
rw_encounters = [0.0, 0.0]
for rep_name, tp, fitvals in [["gp", gp_tp, gp_fit]]:
for rep in range(reps):
samples, fit_samples, best = random_walks.hillclimb(gp_tp, fitvals, steps, rw=False)
for i in range(2):
if inds[i] in samples:
hc_encounters[i] += 1.0 / reps
samples, fit_samples, best = random_walks.hillclimb(gp_tp, fitvals, steps, rw=True)
for i in range(2):
if inds[i] in samples:
rw_encounters[i] += 1.0 / reps
print "hc_encounters", hc_encounters
print "rw_encounters", rw_encounters
# now the GA v GP hillclimb experiments
reps = 30
for rep_name, tp, fitvals in [["ga", ga_tp, ga_fit],
["gp", gp_tp, gp_fit]]:
for uniformify_val in uniformify_vals:
for rep in range(reps):
tp_tmp = random_walks.uniformify(tp, uniformify_val)
samples, fit_samples, best = random_walks.hillclimb(tp_tmp, fitvals, steps, rw=True)
x = float(len(set(samples))) / len(samples)
results[rep_name, uniformify_val, rep] = x
return results, ga_fit, gp_fit
def rw_experiment_with_tp(tp):
N = len(tp)
walklen = int(math.ceil(math.sqrt(N)))
results = []
fitvals = np.zeros_like(tp[0]) # don't need them for this
reps = 30
for rep in range(reps):
samples, fit_samples, best = random_walks.hillclimb(tp, fitvals, walklen, rw=True)
prop_unique = float(len(set(samples))) / len(samples)
results.append(prop_unique)
return np.mean(results), np.std(results)
def ga_hc_experiment(path_results):
"""Run some hill-climbs on variations of a GA space. Report
performance."""
uniformify_vals = [
0.1, 0.5, .75, 0.9, 1.0, 1.0 / 0.9, 1.0 / .75, 2.0, 10.0
]
noise_vals = [0, 1, 10, 100, 1000]
results = OrderedDict()
ga_length = 10
tp_path = os.path.join(path_results, "ga_length_10", "TP.dat")
try:
ga_tp = np.genfromtxt(tp_path)
except:
ga_tp, _ = random_walks.generate_ga_tm(ga_length, pmut=1.0 / ga_length)
np.savetxt(tp_path, ga_tp)
fit_path = os.path.join(path_results, "ga_length_10", "fitness_vals.dat")
try:
ga_fit = np.genfromtxt(fit_path)
except:
ga_fit = random_walks.onemax_fitvals(ga_length)
np.savetxt(fit_path, ga_fit)
# just get mu(sigma()), don't bother with sigma(sigma())
mu_sigma_vals = [
random_walks.mu_sigma(random_walks.uniformify(ga_tp,
uniformify_val))[0]
for uniformify_val in uniformify_vals
]
reps = 30
steps = 50
for rep_name, tp, fitvals in [["ga", ga_tp, ga_fit]]:
for noise_val in noise_vals:
tmp_fit = random_walks.permute_vals(fitvals, noise_val)
for uniformify_val in uniformify_vals:
for rep in range(reps):
tp_tmp = random_walks.uniformify(tp, uniformify_val)
samples, fit_samples, best = random_walks.hillclimb(
tp_tmp, tmp_fit, steps, rw=False)
x = best
results[rep_name, uniformify_val, noise_val, rep] = x
return results, mu_sigma_vals
def ga_hc_experiment(path_results):
"""Run some hill-climbs on variations of a GA space. Report
performance."""
uniformify_vals = [0.1, 0.5, .75, 0.9, 1.0, 1.0/0.9, 1.0/.75, 2.0, 10.0]
noise_vals = [0, 1, 10, 100, 1000]
results = OrderedDict()
ga_length = 10
tp_path = os.path.join(path_results, "ga_length_10", "TP.dat")
try:
ga_tp = np.genfromtxt(tp_path)
except:
ga_tp, _ = random_walks.generate_ga_tm(ga_length, pmut=1.0/ga_length)
np.savetxt(tp_path, ga_tp)
fit_path = os.path.join(path_results, "ga_length_10", "fitness_vals.dat")
try:
ga_fit = np.genfromtxt(fit_path)
except:
ga_fit = random_walks.onemax_fitvals(ga_length)
np.savetxt(fit_path, ga_fit)
# just get mu(sigma()), don't bother with sigma(sigma())
mu_sigma_vals = [random_walks.mu_sigma(random_walks.uniformify(ga_tp, uniformify_val))[0]
for uniformify_val in uniformify_vals]
reps = 30
steps = 50
for rep_name, tp, fitvals in [["ga", ga_tp, ga_fit]]:
for noise_val in noise_vals:
tmp_fit = random_walks.permute_vals(fitvals, noise_val)
for uniformify_val in uniformify_vals:
for rep in range(reps):
tp_tmp = random_walks.uniformify(tp, uniformify_val)
samples, fit_samples, best = random_walks.hillclimb(tp_tmp, tmp_fit,
steps, rw=False)
x = best
results[rep_name, uniformify_val, noise_val, rep] = x
return results, mu_sigma_vals
def ga_gp_rw_experiment(path_results):
uniformify_vals = [
0.1, 0.5, .75, 0.9, 1.0, 1.0 / 0.9, 1.0 / .75, 2.0, 10.0
]
results = OrderedDict()
ga_length = 10
gp_depth = 2
tp_path = os.path.join(path_results, "ga_length_10", "TP.dat")
ga_tp = np.genfromtxt(tp_path)
tp_path = os.path.join(path_results, "depth_2", "TP.dat")
gp_tp = np.genfromtxt(tp_path)
print "doing probability of encounter experiment"
# Do the "probability of encounter" experiment first
reps = 100
steps = 50
fit_path = os.path.join(path_results, "ga_length_10", "fitness_vals.dat")
ga_fit = np.genfromtxt(fit_path)
gp_fit = [
float(s) for s in open(
os.path.join(path_results, "depth_2",
"all_fitness_values.dat")).readlines()
]
inds = 0, len(gp_fit) - 1
hc_encounters = [0.0, 0.0]
rw_encounters = [0.0, 0.0]
for rep_name, tp, fitvals in [["gp", gp_tp, gp_fit]]:
for rep in range(reps):
samples, fit_samples, best = random_walks.hillclimb(gp_tp,
fitvals,
steps,
rw=False)
for i in range(2):
if inds[i] in samples:
hc_encounters[i] += 1.0 / reps
samples, fit_samples, best = random_walks.hillclimb(gp_tp,
fitvals,
steps,
rw=True)
for i in range(2):
if inds[i] in samples:
rw_encounters[i] += 1.0 / reps
print "hc_encounters", hc_encounters
print "rw_encounters", rw_encounters
# now the GA v GP hillclimb experiments
reps = 30
for rep_name, tp, fitvals in [["ga", ga_tp, ga_fit], ["gp", gp_tp,
gp_fit]]:
for uniformify_val in uniformify_vals:
for rep in range(reps):
tp_tmp = random_walks.uniformify(tp, uniformify_val)
samples, fit_samples, best = random_walks.hillclimb(tp_tmp,
fitvals,
steps,
rw=True)
x = float(len(set(samples))) / len(samples)
results[rep_name, uniformify_val, rep] = x
return results, ga_fit, gp_fit
