def run(self):
runner = MIMICRunner(problem=self.problem,
experiment_name=self.name,
seed=self.seed,
iteration_list=self.iterations,
population_sizes=self.population,
keep_percent_list=self.keep_pct,
max_attempts=self.attempts,
generate_curves=True,
output_directory=self.output)
runner.run()
def run(experiment_name, input_size=25):
SEED = 1
OUTPUT_DIRECTORY = './output'
# pop_size_list = list(range(input_size, input_size*6, input_size))
pop_size_list = [50, 200, 250, 500]
if (experiment_name == '4peaks'):
## Four Peaks
problem = generate_problem(experiment_name, seed=SEED, input_size=50)
mmc = MIMICRunner(problem=problem,
experiment_name=experiment_name,
output_directory=OUTPUT_DIRECTORY,
seed=SEED,
iteration_list=2**np.arange(10),
population_sizes=[50, 200, 250, 500],
max_attempts=100,
keep_percent_list=[0.1, 0.25, 0.5, 0.75],
use_fast_mimic=True)
df_run_stats, df_run_curves = mmc.run()
elif (experiment_name == 'flipflop'):
## Flip Flop
problem = generate_problem(experiment_name, seed=SEED, input_size=25)
mmc = MIMICRunner(problem=problem,
experiment_name=experiment_name,
output_directory=OUTPUT_DIRECTORY,
seed=SEED,
iteration_list=2**np.arange(10),
population_sizes=pop_size_list,
max_attempts=50,
keep_percent_list=[0.1, 0.25, 0.5, 0.75],
use_fast_mimic=True)
df_run_stats, df_run_curves = mmc.run()
elif (experiment_name == 'knapsack'):
## Knapsack
problem = generate_problem(experiment_name, seed=SEED, input_size=150)
mmc = MIMICRunner(problem=problem,
experiment_name=experiment_name,
output_directory=OUTPUT_DIRECTORY,
seed=SEED,
iteration_list=2**np.arange(10),
population_sizes=[200, 250, 750],
max_attempts=50,
keep_percent_list=[0.1, 0.25, 0.5, 0.75],
use_fast_mimic=True)
df_run_stats, df_run_curves = mmc.run()
else:
print('input error')
print('input size:', input_size)
def main():
experiment_name_ga = 'Max_K_GA'
OUTPUT_DIRECTORY = './output'
problem = MaxKColorGenerator.generate(seed=seed,
number_of_nodes=40,
max_connections_per_node=4)
ga = GARunner(problem=problem,
experiment_name=experiment_name_ga,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(12),
max_attempts=200,
population_sizes=[150, 200, 250, 300],
mutation_rates=[0.03, 0.1, 0.2, 0.3, 0.4])
ga.run()
experiment_name_mimic = 'Max_K_MIMIC'
mmc = MIMICRunner(problem=problem,
experiment_name=experiment_name_mimic,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(12),
max_attempts=200,
population_sizes=[150, 200, 250, 300],
keep_percent_list=[0.03, 0.1, 0.2, 0.3, 0.4])
mmc.run()
experiment_name_rhc = 'Max_K_RHC'
rhc = RHCRunner(problem=problem,
experiment_name=experiment_name_rhc,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(12),
max_attempts=5000,
restart_list=[25, 75, 100])
rhc.run()
experiment_name_sa = 'Max_K_SA'
sa = SARunner(
problem=problem,
experiment_name=experiment_name_sa,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(14),
max_attempts=200,
temperature_list=[1, 10, 50, 100, 250, 500, 1000, 2500, 5000, 10000])
sa.run()
output_directory=constants.OUTPUT_DIRECTORY,
seed=0,
iteration_list=constants.iteration_list,
max_attempts=100,
population_sizes=population_sizes,
mutation_rates=mutation_rates)
ga_run_stats, ga_run_curves = ga.run()
if algorithm_name == "MIMIC" and False:
population_sizes = [100]
keep_percent_list = [0.05, 0.1, 0.2, 0.3, 0.4]
mimic = MIMICRunner(problem=problem,
experiment_name=experiment_name,
output_directory=constants.OUTPUT_DIRECTORY,
seed=0,
iteration_list=[1, 2, 3, 4, 5, 10, 15, 20, 25, 30],
population_sizes=population_sizes,
max_attempts=100,
keep_percent_list=keep_percent_list,
use_fast_mimic=True)
mimic_run_stats, mimic_run_curves = mimic.run()
type = "curves"
dir = 'output/{0}-{1}-{2}-{3}-experiment'.format(problem_name, algorithm_name,
problem_size, t)
filename = '{3}__{0}-{3}-{1}-{4}-experiment__{2}_df.csv'.format(
problem_name, problem_size, type, algorithm_name, t)
path = '{0}/{1}'.format(dir, filename)
df = pd.read_csv(path)
max_row = df.iloc[df["Fitness"].idxmax()]
def run_traveling_salesman():
OUTPUT_DIRECTORY = './output/TSP_results/3'
SEED = 903387974
# Traveling Salesman
problem = generateTSP(seed=SEED, number_of_cities=22)
population = [350, 500]
prob_rates = [0.01, 0.1, 0.3, 0.5]
# Genetic Algorithm
ga = GARunner(problem=problem,
experiment_name="TSP_Maximize_GA_3",
seed=SEED,
iteration_list=2 ** np.arange(8),
max_attempts=1000,
population_sizes=population,
mutation_rates=prob_rates,
output_directory=OUTPUT_DIRECTORY)
df_ga_run_stats, df_ga_run_curves = ga.run()
ga_data_strings = {
'title': 'Genetic_Algorithms_22_Cities_3',
'Parameters': ['Mutation Rate', 'Population Size']
}
best_fitness = max(df_ga_run_stats['Fitness'])
print("=============")
print(best_fitness)
index_array = np.where(df_ga_run_stats['Fitness'] == best_fitness)
time = 0
for i in index_array:
print(df_ga_run_stats['Mutation Rate'][i])
print(df_ga_run_stats['Population Size'][i])
print(df_ga_run_stats['Time'][i])
print(df_ga_run_stats['Time'][i].sum()/len(df_ga_run_stats['Time'][i]))
avg_time = time / len(index_array)
print(avg_time)
print("============")
fig = plt.figure(dpi=100, figsize=(11, 8))
sns.lineplot(x='Iteration', y='Fitness', data=df_ga_run_curves).set_title('GA 22 Cities ' + str(best_fitness) + ' ' + str(avg_time))
fig.savefig('GA_22_Cities_Seaborn_3.png')
generate_graphs(df_ga_run_stats, df_ga_run_curves, ga_data_strings)
plt.show()
OUTPUT_DIRECTORY = './output/TSP_results/2'
# MIMIC Algorithm
mmc = MIMICRunner(problem=problem,
experiment_name="TSP_Maximize_MIMIC_2",
seed=69,
iteration_list=2 ** np.arange(8),
max_attempts=500,
population_sizes=population,
keep_percent_list=prob_rates,
use_fast_mimic=False,
output_directory=OUTPUT_DIRECTORY)
# Proportion of samples to keep at each iteration
df_mmc_run_stats, df_mmc_run_curves = mmc.run()
mmc_data_strings = {
'title': 'MIMIC_22_Cities_2',
'Parameters': ['Population Size', 'Keep Percent']
}
print("=============")
best_fitness = max(df_mmc_run_stats['Fitness'])
print(best_fitness)
index_array = np.where(df_mmc_run_stats['Fitness'] == best_fitness)
print("Index:")
print(index_array)
time = 0
for i in index_array:
print(df_mmc_run_stats['Keep Percent'][i])
print(df_mmc_run_stats['Population Size'][i])
print(df_mmc_run_stats['Time'][i])
print(df_mmc_run_stats['Time'][i].sum()/len(df_mmc_run_stats['Time'][i]))
avg_time = time / len(index_array)
print(avg_time)
print("============")
fig = plt.figure(dpi=100, figsize=(11, 8))
sns.lineplot(x='Iteration', y='Fitness', data=df_mmc_run_curves).set_title('MIMIC 22 Cities ' + str(best_fitness) + ' ' + str(avg_time))
fig.savefig('MIMIC_22_Cities_Seaborn_2.png')
generate_graphs(df_mmc_run_stats, df_mmc_run_curves, mmc_data_strings)
plt.show()
# Randomized Hill Climbing
rhc = RHCRunner(problem=problem,
experiment_name="TSP_Maximize_RHC_2",
seed=SEED,
iteration_list=2 ** np.arange(8),
max_attempts=500,
restart_list=[5, 15, 25, 50, 75, 95],
output_directory=OUTPUT_DIRECTORY)
# Number of random restarts
df_rhc_run_stats, df_rhc_run_curves = rhc.run()
rhc_data_string = {
'title': 'RHC_22_Cities_2',
'Parameters': ['Restarts']
}
best_fitness = max(df_rhc_run_stats['Fitness'])
print("=============")
print(best_fitness)
index_array = np.where(df_rhc_run_stats['Fitness'] == best_fitness)
time = 0
for i in index_array:
print(df_rhc_run_stats['Restarts'][i])
print(df_rhc_run_stats['Time'][i])
print(df_rhc_run_stats['Time'][i].sum()/len(df_rhc_run_stats['Time'][i]))
avg_time = time / len(index_array)
print(avg_time)
print("============")
fig = plt.figure(dpi=100, figsize=(11, 8))
sns.lineplot(x='Iteration', y='Fitness', data=df_rhc_run_curves).set_title('RHC 22 Cities ' + str(best_fitness) + ' ' + str(avg_time))
fig.savefig('RHC_22_Cities_Seaborn_2.png')
generate_graphs(df_rhc_run_stats, df_rhc_run_curves, rhc_data_string)
plt.show()
# Simulated Annealing
sa = SARunner(problem=problem,
experiment_name="TSP_Maximize_SA_2",
seed=SEED,
iteration_list=2 ** np.arange(10),
max_attempts=1000,
temperature_list=[0.001, 0.05, 0.1, 0.25, 0.5, 0.8],
output_directory=OUTPUT_DIRECTORY)
# Temperature is just the decay
df_sa_run_stats, df_sa_run_curves = sa.run()
sa_data_string = {
'title': 'SA_22_Cities_2',
'Parameters': ['Temperature']
}
best_fitness = max(df_sa_run_stats['Fitness'])
print("=============")
print(best_fitness)
index_array = np.where(df_sa_run_stats['Fitness'] == best_fitness)
time = 0
for i in index_array:
print(df_sa_run_stats['Temperature'][i])
print(df_sa_run_stats['Time'][i])
print(df_sa_run_stats['Time'][i].sum() / len(df_sa_run_stats['Time'][i]))
avg_time = time / len(index_array)
print(avg_time)
print("============")
fig = plt.figure(dpi=100, figsize=(11, 8))
sns.lineplot(x='Iteration', y='Fitness', data=df_sa_run_curves).set_title('SA 22 Cities ' + str(best_fitness) + ' ' + str(avg_time))
fig.savefig('SA_22_Cities_Seaborn_2.png')
generate_graphs(df_sa_run_stats, df_sa_run_curves, sa_data_string)
plt.show()
def run_knapsack():
# Knapsack
OUTPUT_DIRECTORY = './output/Knapsack_results/4'
SEED = 903387974
problem = KnapsackGenerator.generate(seed=SEED, number_of_items_types=12, max_item_count=6, max_weight_per_item=26,
max_value_per_item=12, max_weight_pct=0.8)
population = [50, 130, 210, 500]
prob_rates = [0.05, 0.3, 0.5, 0.8]
# Genetic Algorithm
ga = GARunner(problem=problem,
experiment_name="Knapsack_Maximize_GA_4",
seed=SEED,
iteration_list=2 ** np.arange(6),
max_attempts=500,
population_sizes=population,
mutation_rates=prob_rates,
output_directory=OUTPUT_DIRECTORY)
df_ga_run_stats, df_ga_run_curves = ga.run()
ga_data_strings = {
'title': 'Genetic_Algorithms_Knapsack_4',
'Parameters': ['Mutation Rate', 'Population Size']
}
best_fitness = max(df_ga_run_stats['Fitness'])
print("=============")
print(best_fitness)
index_array = np.where(df_ga_run_stats['Fitness'] == best_fitness)
time = 0
for i in index_array:
print(df_ga_run_stats['Mutation Rate'][i])
print(df_ga_run_stats['Population Size'][i])
print(df_ga_run_stats['Time'][i])
print(df_ga_run_stats['Time'][i].sum()/len(df_ga_run_stats['Time'][i]))
avg_time = time / len(index_array)
print(avg_time)
print("============")
fig = plt.figure(dpi=100, figsize=(11, 8))
sns.lineplot(x='Iteration', y='Fitness', data=df_ga_run_curves).set_title('GA_Knapsack_' + str(best_fitness) + '_' + str(avg_time))
fig.savefig('GA_Knapsack_Seaborn_4.png')
generate_graphs(df_ga_run_stats, df_ga_run_curves, ga_data_strings)
plt.show()
# MIMIC Algorithm
mmc = MIMICRunner(problem=problem,
experiment_name="Knapsack_Maximize_MIMIC_4",
seed=SEED,
iteration_list=2 ** np.arange(6),
max_attempts=500,
population_sizes=population,
keep_percent_list=prob_rates,
use_fast_mimic=False,
output_directory=OUTPUT_DIRECTORY)
# Proportion of samples to keep at each iteration
df_mmc_run_stats, df_mmc_run_curves = mmc.run()
mmc_data_strings = {
'title': 'MIMIC_Knapsack_4',
'Parameters': ['Population Size', 'Keep Percent']
}
best_fitness = max(df_mmc_run_stats['Fitness'])
print("=============")
print(best_fitness)
index_array = np.where(df_mmc_run_stats['Fitness'] == best_fitness)
time = 0
for i in index_array:
print(df_mmc_run_stats['Keep Percent'][i])
print(df_mmc_run_stats['Population Size'][i])
print(df_mmc_run_stats['Time'][i])
print(df_mmc_run_stats['Time'][i].sum()/len(df_mmc_run_stats['Time'][i]))
avg_time = time / len(index_array)
print(avg_time)
print("============")
fig = plt.figure(dpi=100, figsize=(11, 8))
sns.lineplot(x='Iteration', y='Fitness', data=df_mmc_run_curves).set_title('MIMIC_Knapsack_' + str(best_fitness) + '_' + str(avg_time))
fig.savefig('MIMIC_Knapsack_Seaborn_4.png')
generate_graphs(df_mmc_run_stats, df_mmc_run_curves, mmc_data_strings)
plt.show()
# Randomized Hill Climbing
rhc = RHCRunner(problem=problem,
experiment_name="Knapsack_Maximize_RHC_4",
seed=SEED,
iteration_list=2 ** np.arange(2),
max_attempts=250,
restart_list=[5, 15, 50],
output_directory=OUTPUT_DIRECTORY)
# Number of random restarts
df_rhc_run_stats, df_rhc_run_curves = rhc.run()
rhc_data_string = {
'title': 'RHC_Knapsack_4',
'Parameters': ['Restarts']
}
best_fitness = max(df_rhc_run_stats['Fitness'])
print("=============")
print(best_fitness)
index_array = np.where(df_rhc_run_stats['Fitness'] == best_fitness)
time = 0
for i in index_array:
print(df_rhc_run_stats['Restarts'][i])
print(df_rhc_run_stats['Time'][i])
print(df_rhc_run_stats['Time'][i].sum()/len(df_rhc_run_stats['Time'][i]))
avg_time = time / len(index_array)
print(avg_time)
print("============")
fig = plt.figure(dpi=100, figsize=(11, 8))
sns.lineplot(x='Iteration', y='Fitness', data=df_rhc_run_curves).set_title('RHC_Knapsack_' + str(best_fitness) + '_' + str(avg_time))
fig.savefig('RHC_Knapsack_Seaborn_4.png')
generate_graphs(df_rhc_run_stats, df_rhc_run_curves, rhc_data_string)
plt.show()
# Simulated Annealing
sa = SARunner(problem=problem,
experiment_name="Knapsack_Maximize_SA_4",
seed=SEED,
iteration_list=2 ** np.arange(8),
max_attempts=1000,
temperature_list=[0.001, 0.05, 0.1, 0.25, 0.5, 0.8],
output_directory=OUTPUT_DIRECTORY)
# Temperature is just the Geometric decay
df_sa_run_stats, df_sa_run_curves = sa.run()
sa_data_string = {
'title': 'SA_Knapsack_4',
'Parameters': ['Temperature']
}
best_fitness = max(df_sa_run_stats['Fitness'])
print("=============")
print(best_fitness)
index_array = np.where(df_sa_run_stats['Fitness'] == best_fitness)
time = 0
for i in index_array:
print(df_sa_run_stats['Temperature'][i])
print(df_sa_run_stats['Time'][i])
print(df_sa_run_stats['Time'][i].sum()/len(df_sa_run_stats['Time'][i]))
avg_time = time / len(index_array)
print(avg_time)
print("============")
fig = plt.figure(dpi=100, figsize=(11, 8))
sns.lineplot(x='Iteration', y='Fitness', data=df_sa_run_curves).set_title('SA_Knapsack_' + str(best_fitness) + '_' + str(avg_time))
fig.savefig('SA_Knapsack_Seaborn_4.png')
generate_graphs(df_sa_run_stats, df_sa_run_curves, sa_data_string)
plt.show()
def run_continuous_peaks(run_rhc, run_sa, run_ga, run_mimic, find_params):
np.random.seed(0)
continuous_peaks_problem = CustomContinuousPeaksGenerator.generate(1, 80)
rhc_run_iterations = None
rhc_run_average_restarts = None
rhc_run_average_time = None
sa_run_iterations = None
sa_run_average = None
sa_run_average_time = None
ga_run_iterations = None
ga_run_average = None
ga_run_average_time = None
mimic_run_iterations = None
mimic_run_average = None
mimic_run_average_time = None
if run_rhc:
out_dir = TRAVELING_SALES_PERSON_PATH + RHC_PATH
if find_params:
title_dic = {'fontsize': 6, 'fontweight': 'bold'}
fig, (ax1), = plt.subplots(1, 1, figsize=(3, 2))
ax1.set_title(
"Randomized Hill Climbing with Various Restarts \n Continous Peaks Problem",
title_dic)
ax1.set_ylabel("Average Fitness Score", title_dic)
ax1.tick_params(axis="x", labelsize=6)
ax1.tick_params(axis="y", labelsize=6)
ax1.set_xlabel("Iteration", title_dic)
num_restarts = [3, 5, 15, 25, 50]
restart_params = {3: [], 5: [], 15: [], 25: [], 50: []}
for param in num_restarts:
restarts = []
iterations = []
for i in range(param):
seed = np.random.randint(100)
# run RHC runner with param restarts.
rhc = RHCRunner(problem=continuous_peaks_problem,
experiment_name="Grid Search Restart " +
str(i) + " of max restarts " + str(param),
output_directory=out_dir,
seed=seed,
iteration_list=np.arange(1, 2200, 100),
max_attempts=500,
restart_list=[0])
rhc_run_stats, _ = rhc.run()
iterations = rhc_run_stats['Iteration'].tolist()
restarts.append(rhc_run_stats['Fitness'].tolist())
average_restarts = np.mean(np.asarray(restarts), axis=0)
ax1.plot(iterations,
average_restarts,
label=str(param) + " Restarts")
# Put a legend to the right of the current axis
box = ax1.get_position()
ax1.set_position([box.x0, box.y0, box.width * 0.99, box.height])
ax1.legend(loc='best', fontsize=6)
ax1.grid()
plt.tight_layout()
filename = "various restarts CPP problem.png"
filename = os.path.join(out_dir, filename)
plt.savefig(filename)
# Reproduce results from 3 restart run.
restarts = []
iterations = []
times = []
for i in range(15):
# run RHC runner with param restarts.
rhc = RHCRunner(problem=continuous_peaks_problem,
experiment_name="RHC CPP",
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 3500, 100),
max_attempts=500,
restart_list=[0])
rhc_run_stats, _ = rhc.run()
rhc_run_iterations = rhc_run_stats['Iteration'].tolist()
restarts.append(rhc_run_stats['Fitness'].tolist())
times.append(rhc_run_stats["Time"].iloc[-1] * 1000)
rhc_run_average_restarts = np.mean(np.asarray(restarts), axis=0)
rhc_run_average_time = np.mean(times)
if run_sa:
out_dir = TRAVELING_SALES_PERSON_PATH + SA_PATH
if find_params:
decay_list = [mlrose_hiive.GeomDecay, mlrose_hiive.ExpDecay]
decay_names = {
mlrose_hiive.GeomDecay: ('Geometric', 'solid'),
mlrose_hiive.ExpDecay: ('Exponential', 'dotted'),
mlrose_hiive.ArithDecay: ('Arithmetic', 'dashed')
}
# Arithmetic decay was way too slow. No progress was getting made.
decay_names = {
mlrose_hiive.GeomDecay: ('Geometric', 'solid'),
mlrose_hiive.ExpDecay: ('Exponential', 'dotted')
}
#mlrose_hiive.ArithDecay: ('Arithmetic', 'dashed')}
temperature_list = [1, 10, 50, 100, 250, 500, 1000]
parameters = []
for i in decay_list:
for j in temperature_list:
parameters.append((i, j))
title_dic = {'fontsize': 6, 'fontweight': 'bold'}
fig, (ax1), = plt.subplots(1, 1, figsize=(5, 2.5))
ax1.set_title(
"Simulated Annealing with Various Decay Schedules \n and Initial Temperature",
title_dic)
ax1.set_ylabel("Average Fitness Score", title_dic)
ax1.tick_params(axis="x", labelsize=6)
ax1.tick_params(axis="y", labelsize=6)
ax1.set_xlabel("Iteration", title_dic)
for entry in parameters:
runs = []
for i in range(5):
sa = SARunner(
problem=continuous_peaks_problem,
experiment_name=
"TSP Simulated Annealing with Decay Schedule" +
str(entry[0]) + "and Initial Temperature " +
str(entry[1]) + "Run " + str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 5000, 100),
max_attempts=200,
decay_list=[entry[0]],
temperature_list=[entry[1]])
sa_run_stats, _ = sa.run()
sa_run_iterations = sa_run_stats['Iteration'].tolist()
runs.append(sa_run_stats['Fitness'].tolist())
sa_run_average = np.mean(np.asarray(runs), axis=0)
ax1.plot(sa_run_iterations,
sa_run_average,
linestyle=decay_names[entry[0]][1],
label=decay_names[entry[0]][0] + " Init Temp:" +
str(entry[1]))
# Put a legend to the right of the current axis
box = ax1.get_position()
ax1.set_position([box.x0, box.y0, box.width * 0.99, box.height])
ax1.legend(loc='center left', bbox_to_anchor=(1, 0.5), fontsize=6)
ax1.grid()
plt.tight_layout()
filename = "decay_and_initial_temp.png"
filename = os.path.join(out_dir, filename)
plt.savefig(filename)
runs = []
times = []
for i in range(5):
sa = SARunner(
problem=continuous_peaks_problem,
experiment_name=
"TSP Simulated Annealing after Param Tuning Decay Schedule",
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 5000, 100),
max_attempts=1000,
decay_list=[mlrose_hiive.GeomDecay],
temperature_list=[50])
sa_run_stats, _ = sa.run()
sa_run_iterations = sa_run_stats['Iteration'].tolist()
times.append(sa_run_stats["Time"].iloc[-1] * 1000)
runs.append(sa_run_stats['Fitness'].tolist())
sa_run_average = np.mean(np.asarray(runs), axis=0)
sa_run_average_time = np.mean(times)
if run_ga:
out_dir = TRAVELING_SALES_PERSON_PATH + GA_PATH
if find_params:
population_sizes = [150, 200, 250]
pop_keys = {250: ('solid')}
# first iteration revealed a pop size of 150 was sufficient.
population_sizes = [250]
mutation_rates = [0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6]
#pop_keys = {
# 150: ('solid')
#}
parameters = []
for i in population_sizes:
for j in mutation_rates:
parameters.append((i, j))
title_dic = {'fontsize': 6, 'fontweight': 'bold'}
fig, (ax1), = plt.subplots(1, 1, figsize=(3, 2))
ax1.set_title("Genetic Algorithms with Various Mutation Rates",
title_dic)
ax1.set_ylabel("Average Fitness Score", title_dic)
ax1.tick_params(axis="x", labelsize=6)
ax1.tick_params(axis="y", labelsize=6)
ax1.set_xlabel("Iteration", title_dic)
for entry in parameters:
runs = []
for i in range(5):
ga = GARunner(problem=continuous_peaks_problem,
experiment_name="TSP GA with Population" +
str(entry[0]) + "and Mutation Split " +
str(entry[1]),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 500, 25),
max_attempts=150,
population_sizes=[entry[0]],
mutation_rates=[entry[1]])
ga_run_stats, _ = ga.run()
ga_run_iterations = ga_run_stats['Iteration'].tolist()
runs.append(ga_run_stats['Fitness'].tolist())
ga_run_average = np.mean(np.asarray(runs), axis=0)
ax1.plot(ga_run_iterations,
ga_run_average,
linestyle=pop_keys[entry[0]],
label="MutRate:" + str(entry[1]))
# Put a legend to the right of the current axis
box = ax1.get_position()
ax1.legend(loc='best', fontsize=6)
ax1.grid()
plt.tight_layout()
filename = "population_and_mutation_rates.png"
filename = os.path.join(out_dir, filename)
plt.savefig(filename)
runs = []
times = []
for i in range(5):
ga = GARunner(problem=continuous_peaks_problem,
experiment_name="CPP GA After Parameter Tuning" +
"Run " + str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 600, 25),
max_attempts=150,
population_sizes=[250],
mutation_rates=[0.3])
ga_run_stats, _ = ga.run()
ga_run_iterations = ga_run_stats['Iteration'].tolist()
times.append(ga_run_stats["Time"].iloc[-1] * 1000)
runs.append(ga_run_stats['Fitness'].tolist())
ga_run_average = np.mean(np.asarray(runs), axis=0)
ga_run_average_time = np.mean(times)
if run_mimic:
out_dir = TRAVELING_SALES_PERSON_PATH + MIMIC_PATH
if find_params:
population_sizes = [200, 250, 300]
# Second search, 200 provided the overall best.
population_sizes = [250]
keep_percent = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7]
pop_keys = {
250: ('solid'),
}
parameters = []
for i in population_sizes:
for j in keep_percent:
parameters.append((i, j))
title_dic = {'fontsize': 6, 'fontweight': 'bold'}
fig, (ax1), = plt.subplots(1, 1, figsize=(3, 2))
ax1.set_title(
"Mimic with 250 Population Size and Various \n Keep Percentages",
title_dic)
ax1.set_ylabel("Fitness Score", title_dic)
ax1.tick_params(axis="x", labelsize=6)
ax1.tick_params(axis="y", labelsize=6)
ax1.set_xlabel("Iteration", title_dic)
for entry in parameters:
runs = []
for i in range(5):
mimic = MIMICRunner(
problem=continuous_peaks_problem,
experiment_name="TSP Mimic with Population" +
str(entry[0]) + "and Keep Percent " + str(entry[1]),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 250, 25),
population_sizes=[entry[0]],
keep_percent_list=[entry[1]],
max_attempts=5,
generate_curves=True,
use_fast_mimic=True)
df_run_stats, _ = mimic.run()
mimic_run_iterations = df_run_stats['Iteration'].tolist()
runs.append(df_run_stats['Fitness'].tolist())
mimic_run_average = np.mean(np.asarray(runs), axis=0)
ax1.plot(mimic_run_iterations,
mimic_run_average,
linestyle=pop_keys[entry[0]],
label="Keep %:" + str(entry[1]))
# Put a legend to the right of the current axis
box = ax1.get_position()
ax1.set_position([box.x0, box.y0, box.width * 0.99, box.height])
ax1.legend(loc='best', fontsize=6)
ax1.grid()
plt.tight_layout()
filename = "population_and_keep_percent_rates.png"
filename = os.path.join(out_dir, filename)
plt.savefig(filename)
runs = []
times = []
for i in range(5):
mimic = MIMICRunner(problem=continuous_peaks_problem,
experiment_name="CPP Mimic" + "Run " + str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 150, 5),
population_sizes=[250],
keep_percent_list=[0.5],
max_attempts=50,
generate_curves=True,
use_fast_mimic=True)
mimic_run_stats, _ = mimic.run()
mimic_run_iterations = mimic_run_stats['Iteration'].tolist()
times.append(mimic_run_stats["Time"].iloc[-1] * 1000)
runs.append(mimic_run_stats['Fitness'].tolist())
mimic_run_average = np.mean(np.asarray(runs), axis=0)
mimic_run_average_time = np.mean(times)
title_dic = {'fontsize': 6, 'fontweight': 'bold'}
fig, (fitness_cp, time_cp) = plt.subplots(1, 2, figsize=(6, 2))
fitness_cp.set_title(
"Fitness vs Iterations\n Continuous Peak Problem N = 80", title_dic)
fitness_cp.set_ylabel("Average Fitness Score", title_dic)
fitness_cp.tick_params(axis="x", labelsize=6)
fitness_cp.tick_params(axis="y", labelsize=6)
fitness_cp.set_xlabel("Iteration", title_dic)
fitness_cp.plot(sa_run_iterations, sa_run_average, label="SA")
fitness_cp.plot(ga_run_iterations, ga_run_average, label="GA")
fitness_cp.plot(rhc_run_iterations, rhc_run_average_restarts, label="RHC")
fitness_cp.plot(mimic_run_iterations, mimic_run_average, label="MIMIC")
fitness_cp.plot(sa_run_iterations,
np.ones((len(sa_run_iterations), 1)) * 151,
label="Global\nOptimum",
linestyle='--',
color='k')
fitness_cp.legend(loc='best', fontsize=6)
time_cp.set_xlabel("Time (ms)", title_dic)
time_cp.set_title("Runtime \n Continuous Peaks Problem N = 80", title_dic)
time_cp.set_ylabel("Algorithm", title_dic)
time_cp.tick_params(axis="x", labelsize=6)
time_cp.tick_params(axis="y", labelsize=6)
x = ['RHC', 'SA', 'GA', 'MIMIC']
x_pos = [i for i, _ in enumerate(x)]
time_cp.set_yticks(x_pos)
time_cp.set_yticklabels(x)
time_cp.barh(x_pos, [
rhc_run_average_time, sa_run_average_time, ga_run_average_time,
mimic_run_average_time
])
time_cp.grid()
fitness_cp.grid()
plt.tight_layout()
path = TRAVELING_SALES_PERSON_PATH
filename = "fitness_vs_iteration.png"
filename = os.path.join(path, filename)
plt.savefig(filename)
title_dic = {'fontsize': 6, 'fontweight': 'bold'}
fig, (fitness_cp, time_cp) = plt.subplots(1, 2, figsize=(6, 2))
fitness_cp.set_title("Fitness vs Problem Size\n Continuous Peak Problem",
title_dic)
fitness_cp.set_ylabel("Average Fitness Score", title_dic)
fitness_cp.tick_params(axis="x", labelsize=6)
fitness_cp.tick_params(axis="y", labelsize=6)
fitness_cp.set_xlabel("Problem Size(N)", title_dic)
# Running on different problem set sizes
sizes = [50, 75, 100, 125, 150]
rhc_average_size = []
sa_average_size = []
ga_average_size = []
mimic_average_size = []
rhc_average_time = []
sa_average_time = []
ga_average_time = []
mimic_average_time = []
for size in sizes:
continuous_peaks_problem = CustomContinuousPeaksGenerator.generate(
1, size)
restarts = []
iterations = []
seeds = [44, 43, 29]
times = []
for i in range(3):
seed = seeds[i]
# run RHC runner with param restarts.
rhc = RHCRunner(problem=continuous_peaks_problem,
experiment_name="RHC CPP",
output_directory=out_dir,
seed=seed,
iteration_list=np.arange(1, 2200, 100),
max_attempts=500,
restart_list=[0])
rhc_run_stats, _ = rhc.run()
rhc_run_iterations = rhc_run_stats['Iteration'].tolist()
restarts.append(rhc_run_stats['Fitness'].tolist())
times.append(rhc_run_stats["Time"].iloc[-1] * 1000)
rhc_run_average_restarts = np.mean(np.asarray(restarts), axis=0)
rhc_run_average_time = np.mean(times)
rhc_average_size.append(rhc_run_average_restarts[-1])
rhc_average_time.append(rhc_run_average_time)
runs = []
times = []
for i in range(5):
sa = SARunner(
problem=continuous_peaks_problem,
experiment_name=
"TSP Simulated Annealing after Param Tuning Decay Schedule",
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 5000, 100),
max_attempts=1000,
decay_list=[mlrose_hiive.GeomDecay],
temperature_list=[1000])
sa_run_stats, _ = sa.run()
sa_run_iterations = sa_run_stats['Iteration'].tolist()
times.append(sa_run_stats["Time"].iloc[-1] * 1000)
runs.append(sa_run_stats['Fitness'].tolist())
sa_run_average = np.mean(np.asarray(runs), axis=0)
sa_run_average_time = np.mean(times)
sa_average_size.append(sa_run_average[-1])
sa_average_time.append(sa_run_average_time)
runs = []
times = []
for i in range(5):
ga = GARunner(problem=continuous_peaks_problem,
experiment_name="CPP GA After Parameter Tuning",
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 500, 25),
max_attempts=150,
population_sizes=[150],
mutation_rates=[0.3])
ga_run_stats, _ = ga.run()
ga_run_iterations = ga_run_stats['Iteration'].tolist()
times.append(ga_run_stats["Time"].iloc[-1] * 1000)
runs.append(ga_run_stats['Fitness'].tolist())
ga_run_average = np.mean(np.asarray(runs), axis=0)
ga_run_average_time = np.mean(times)
ga_average_size.append(ga_run_average[-1])
ga_average_time.append(ga_run_average_time)
runs = []
times = []
for i in range(5):
mimic = MIMICRunner(problem=continuous_peaks_problem,
experiment_name="CPP Mimic",
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 150, 5),
population_sizes=[250],
keep_percent_list=[0.5],
max_attempts=50,
generate_curves=True,
use_fast_mimic=True)
mimic_run_stats, _ = mimic.run()
mimic_run_iterations = mimic_run_stats['Iteration'].tolist()
times.append(mimic_run_stats["Time"].iloc[-1] * 1000)
runs.append(mimic_run_stats['Fitness'].tolist())
mimic_run_average = np.mean(np.asarray(runs), axis=0)
mimic_run_average_time = np.mean(times)
mimic_average_size.append(mimic_run_average[-1])
mimic_average_time.append(mimic_run_average_time)
fitness_cp.plot(sizes, sa_average_size, label="SA")
fitness_cp.plot(sizes, ga_average_size, label="GA")
fitness_cp.plot(sizes, rhc_average_size, label="RHC")
fitness_cp.plot(sizes, mimic_average_size, label="MIMIC")
fitness_cp.legend(loc='best', fontsize=6)
time_cp.set_ylabel("Time (ms)", title_dic)
time_cp.set_title("Runtime vs Problem Size \n Continuous Peaks Problem",
title_dic)
time_cp.set_xlabel("Problem Size(N)", title_dic)
time_cp.tick_params(axis="x", labelsize=6)
time_cp.tick_params(axis="y", labelsize=6)
time_cp.plot(sizes, sa_average_time, label="SA")
time_cp.plot(sizes, ga_average_time, label="GA")
time_cp.plot(sizes, rhc_average_time, label="RHC")
time_cp.plot(sizes, mimic_average_time, label="MIMIC")
time_cp.legend(loc='best', fontsize=6)
time_cp.grid()
fitness_cp.grid()
plt.tight_layout()
path = TRAVELING_SALES_PERSON_PATH
filename = "fitness_vs_iteration_various_problem_sizes.png"
filename = os.path.join(path, filename)
plt.savefig(filename)
print("fitness evaluations: {}".format(eval_count))
rhc_evals_list.append(eval_count)
eval_count = 0
mimic_evals_list = []
df_mimic_stats_list = []
df_mimic_curves_list = []
for r in range(mimic_repeats):
problem.set_mimic_fast_mode(True)
mimic = MIMICRunner(problem=problem,
experiment_name="mimic_test",
output_directory="./results/",
seed=r,
iteration_list=2 ** np.arange(18),
max_attempts=200,
use_fast_mimic=True,
keep_percent_list=[0.05],
population_sizes=[200])
df_mimic_stats, df_mimic_curves = mimic.run()
df_mimic_stats_list.append(df_mimic_stats)
df_mimic_curves_list.append(df_mimic_curves)
print("fitness evaluations: {}".format(eval_count))
mimic_evals_list.append(eval_count)
eval_count = 0
#
# fitness vs iterations
def main():
experiment_name_ga = 'TSP_Maximize_GA'
OUTPUT_DIRECTORY = './output'
problem = TSPGenerator.generate(seed=seed,
number_of_cities=50,
maximize=True)
ga = GARunner(problem=problem,
experiment_name=experiment_name_ga,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(12),
max_attempts=200,
population_sizes=[150, 200, 250, 300],
mutation_rates=[0.01, 0.02, 0.03, 0.1, 0.2, 0.3, 0.4])
ga.run()
experiment_name_mimic = 'TSP_Maximize_MIMIC'
mmc = MIMICRunner(problem=problem,
experiment_name=experiment_name_mimic,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(12),
max_attempts=200,
population_sizes=[150, 200, 250, 300],
keep_percent_list=[0.01, 0.02, 0.03, 0.1, 0.2, 0.3, 0.4])
mmc.run()
experiment_name_rhc = 'TSP_Maximize_RHC'
rhc = RHCRunner(problem=problem,
experiment_name=experiment_name_rhc,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(12),
max_attempts=200,
restart_list=[25, 75, 100])
rhc.run()
experiment_name_sa = 'TSP_Maximize_SA'
sa = SARunner(
problem=problem,
experiment_name=experiment_name_sa,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(14),
max_attempts=200,
temperature_list=[1, 10, 50, 100, 250, 500, 1000, 2500, 5000, 10000])
sa.run()
OUTPUT_DIRECTORY = './output/N_CITIES'
for i in range(10, 100):
problem = TSPGenerator.generate(seed=seed,
number_of_cities=i,
maximize=True)
experiment_name_cities_ga = str(i) + '_' + 'City_Fitness_ga'
ga = GARunner(problem=problem,
experiment_name=experiment_name_cities_ga,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(12),
max_attempts=200,
population_sizes=[300],
mutation_rates=[0.03])
ga.run()
experiment_name_cities_MIMIC = str(i) + '_' + 'City_Fitness_MIMIC'
mmc = MIMICRunner(problem=problem,
experiment_name=experiment_name_cities_MIMIC,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(12),
max_attempts=200,
population_sizes=[300],
keep_percent_list=[0.2])
mmc.run()
experiment_name_cities_RHC = str(i) + '_' + 'City_Fitness_RHC'
rhc = RHCRunner(problem=problem,
experiment_name=experiment_name_cities_RHC,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(12),
max_attempts=200,
restart_list=[25])
rhc.run()
experiment_name_cities_sa = str(i) + '_' + 'City_Fitness_SA'
sa = SARunner(problem=problem,
experiment_name=experiment_name_cities_sa,
output_directory=OUTPUT_DIRECTORY,
seed=seed,
iteration_list=2**np.arange(14),
max_attempts=200,
temperature_list=[10])
sa.run()
def run_knapsack_experiment(run_rhc, run_sa, run_ga, run_mimic, find_params):
title_dic = {'fontsize': 7, 'fontweight': 'bold'}
axis_label_size = 7
tuning_figure_size = (3.5, 2)
np.random.seed(0)
problem = CustomKnapSackGenerator.generate(
1,
number_of_items_types=100,
max_item_count=50,
max_weight_per_item=100,
max_value_per_item=50,
max_weight_pct=0.6,
multiply_by_max_item_count=True,
crossover=one_point_crossover.OnePointCrossOver,
mutator=change_one_mutator.ChangeOneMutator)
rhc_run_iterations = None
rhc_run_average_restarts = None
rhc_run_average_time = None
sa_run_iterations = None
sa_run_average = None
sa_run_average_time = None
ga_run_iterations = None
ga_run_average = None
ga_run_average_time = None
mimic_run_iterations = None
mimic_run_average = None
mimic_run_average_time = None
if run_rhc:
out_dir = KNAPSACK_PATH + RHC_PATH
if find_params:
fig, (ax1), = plt.subplots(1, 1, figsize=tuning_figure_size)
ax1.set_title(
"Randomized Hill Climbing with Various Restarts \n Continous Peaks Problem",
title_dic)
ax1.set_ylabel("Average Fitness Score", title_dic)
ax1.tick_params(axis="x", labelsize=axis_label_size)
ax1.tick_params(axis="y", labelsize=axis_label_size)
ax1.set_xlabel("Iteration", title_dic)
num_restarts = [3, 5, 15, 25, 50]
restart_params = {3: [], 5: [], 15: [], 25: [], 50: []}
for param in num_restarts:
restarts = []
iterations = []
for i in range(param):
seed = np.random.randint(100)
# run RHC runner with param restarts.
rhc = RHCRunner(problem=problem,
experiment_name="Grid Search Restart " +
str(i) + " of max restarts " + str(param),
output_directory=out_dir,
seed=seed,
iteration_list=np.arange(1, 1000, 100),
max_attempts=500,
restart_list=[0])
rhc_run_stats, _ = rhc.run()
iterations = rhc_run_stats['Iteration'].tolist()
fitness = np.asarray(rhc_run_stats['Fitness'].tolist())
restarts.append(fitness)
average_restarts = np.mean(np.asarray(restarts), axis=0)
ax1.plot(iterations,
average_restarts,
label=str(param) + " Restarts")
# Put a legend to the right of the current axis
box = ax1.get_position()
ax1.set_position([box.x0, box.y0, box.width * 0.99, box.height])
ax1.legend(loc='best', fontsize=6)
ax1.grid()
plt.tight_layout()
filename = "various restarts CPP problem.png"
filename = os.path.join(out_dir, filename)
plt.savefig(filename)
# Reproduce results from 3 restart run.
restarts = []
iterations = []
times = []
for i in range(15):
# run RHC runner with param restarts.
rhc = RHCRunner(problem=problem,
experiment_name="RHC QUEENS" + str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 1000, 100),
max_attempts=500,
restart_list=[0])
rhc_run_stats, _ = rhc.run()
rhc_run_iterations = rhc_run_stats['Iteration'].tolist()
fitness = np.asarray(rhc_run_stats['Fitness'].tolist())
restarts.append(fitness)
times.append(rhc_run_stats["Time"].iloc[-1] * 1000)
rhc_run_average_restarts = np.mean(np.asarray(restarts), axis=0)
rhc_run_average_time = np.mean(times)
if run_sa:
out_dir = KNAPSACK_PATH + SA_PATH
if find_params:
#decay_list = [mlrose_hiive.GeomDecay, mlrose_hiive.ExpDecay]
# Round two: Exp Decay was better
decay_list = [mlrose_hiive.ExpDecay]
decay_names = {
mlrose_hiive.GeomDecay: ('Geometric', 'dotted'),
mlrose_hiive.ExpDecay: ('Exponential', 'solid'),
mlrose_hiive.ArithDecay: ('Arithmetic', 'dashed')
}
temperature_list = [500, 1000, 2500, 3500, 4500, 5000]
parameters = []
for i in decay_list:
for j in temperature_list:
parameters.append((i, j))
fig, (ax1), = plt.subplots(1, 1, figsize=tuning_figure_size)
ax1.set_title(
"Simulated Annealing with Exponential Decay \n Schedule and Initial Temperature - Knapsack",
title_dic)
ax1.set_ylabel("Average Fitness Score", title_dic)
ax1.tick_params(axis="x", labelsize=axis_label_size)
ax1.tick_params(axis="y", labelsize=axis_label_size)
ax1.set_xlabel("Iteration", title_dic)
for entry in parameters:
runs = []
for i in range(5):
sa = SARunner(
problem=problem,
experiment_name=
"KnapSack Simulated Annealing with Decay Schedule" +
str(entry[0]) + "and Initial Temperature " +
str(entry[1]) + "Run " + str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 2000, 100),
max_attempts=1000,
decay_list=[entry[0]],
temperature_list=[entry[1]])
sa_run_stats, _ = sa.run()
sa_run_iterations = sa_run_stats['Iteration'].tolist()
fitness = np.asarray(sa_run_stats['Fitness'].tolist())
runs.append(fitness)
sa_run_average = np.mean(np.asarray(runs), axis=0)
ax1.plot(sa_run_iterations,
sa_run_average,
linestyle=decay_names[entry[0]][1],
label="Init Temp:" + str(entry[1]))
ax1.legend(loc='best', fontsize=6)
ax1.grid()
plt.tight_layout()
filename = "decay_and_initial_temp.png"
filename = os.path.join(out_dir, filename)
plt.savefig(filename)
# Run 5 times and average
runs = []
times = []
for i in range(5):
sa = SARunner(
problem=problem,
experiment_name=
"Knapsack Simulated Annealing after Param Tuning Decay Schedule "
+ str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 1500, 100),
max_attempts=1000,
decay_list=[mlrose_hiive.ExpDecay],
temperature_list=[3500])
sa_run_stats, _ = sa.run()
sa_run_iterations = sa_run_stats['Iteration'].tolist()
times.append(sa_run_stats["Time"].iloc[-1] * 1000)
fitness = np.asarray(sa_run_stats['Fitness'].tolist())
runs.append(fitness)
sa_run_average = np.mean(np.asarray(runs), axis=0)
sa_run_average_time = np.mean(times)
if run_ga:
out_dir = KNAPSACK_PATH + GA_PATH
if find_params:
population_sizes = [25, 50, 75, 100]
# first iteration revealed a pop size of 75 was sufficient.
population_sizes = [75]
pop_keys = {100: ('dashed'), 50: ('dashed'), 75: ('solid')}
mutation_rates = [
0.1,
0.3,
0.6,
0.9,
]
parameters = []
for i in population_sizes:
for j in mutation_rates:
parameters.append((i, j))
fig, (ax1), = plt.subplots(1, 1, figsize=tuning_figure_size)
ax1.set_title(
"Genetic Algorithms with Population Size 75 & \n Various Mutation Rates - Knapsack",
title_dic)
ax1.set_ylabel("Average Fitness Score", title_dic)
ax1.tick_params(axis="x", labelsize=axis_label_size)
ax1.tick_params(axis="y", labelsize=axis_label_size)
ax1.set_xlabel("Iteration", title_dic)
for entry in parameters:
runs = []
for i in range(5):
ga = GARunner(
problem=problem,
experiment_name="Knapsack GA with Population" +
str(entry[0]) + "and Mutation Split " + str(entry[1]),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 1500, 25),
max_attempts=150,
population_sizes=[entry[0]],
mutation_rates=[entry[1]])
ga_run_stats, _ = ga.run()
fitness = np.asarray(ga_run_stats['Fitness'].tolist())
ga_run_iterations = ga_run_stats['Iteration'].tolist()
runs.append(fitness)
ga_run_average = np.mean(np.asarray(runs), axis=0)
ax1.plot(ga_run_iterations,
ga_run_average,
linestyle=pop_keys[entry[0]],
label="MutRate:" + str(entry[1]))
ax1.legend(loc='best', fontsize=6, ncol=2)
ax1.grid()
plt.tight_layout()
filename = "population_and_mutation_rates.png"
filename = os.path.join(out_dir, filename)
plt.savefig(filename)
runs = []
times = []
for i in range(5):
ga = GARunner(problem=problem,
experiment_name="CPP GA After Parameter Tuning" +
"Run " + str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 1500, 25),
max_attempts=150,
population_sizes=[75],
mutation_rates=[0.9])
ga_run_stats, _ = ga.run()
ga_run_iterations = ga_run_stats['Iteration'].tolist()
times.append(ga_run_stats["Time"].iloc[-1] * 1000)
fitness = np.asarray(ga_run_stats['Fitness'].tolist())
runs.append(fitness)
ga_run_average = np.mean(np.asarray(runs), axis=0)
ga_run_average_time = np.mean(times)
if run_mimic:
out_dir = KNAPSACK_PATH + MIMIC_PATH
if find_params:
population_sizes = [25, 50, 100, 150, 200]
# Pop size of 100 was best tradeoff with time.
population_sizes = [100]
keep_percent = [
0.3,
0.4,
0.5,
0.6,
0.7,
0.8,
0.9,
]
pop_keys = {100: ('solid')}
parameters = []
for i in population_sizes:
for j in keep_percent:
parameters.append((i, j))
fig, (ax1), = plt.subplots(1, 1, figsize=tuning_figure_size)
ax1.set_title(
"Mimic with 100 Population Size and Various \n Keep Percentages - Knapsack",
title_dic)
ax1.set_ylabel("Average Fitness Score", title_dic)
ax1.tick_params(axis="x", labelsize=axis_label_size)
ax1.tick_params(axis="y", labelsize=axis_label_size)
ax1.set_xlabel("Iteration", title_dic)
for entry in parameters:
runs = []
for i in range(5):
mimic = MIMICRunner(
problem=problem,
experiment_name="Knapsack Mimic with Population" +
str(entry[0]) + "and Keep Percent " + str(entry[1]) +
"run " + str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 50, 1),
population_sizes=[entry[0]],
keep_percent_list=[entry[1]],
max_attempts=5,
generate_curves=True,
use_fast_mimic=True)
df_run_stats, _ = mimic.run()
mimic_run_iterations = df_run_stats['Iteration'].tolist()
fitness = np.asarray(df_run_stats['Fitness'].tolist())
runs.append(fitness)
mimic_run_average = np.mean(np.asarray(runs), axis=0)
ax1.plot(mimic_run_iterations,
mimic_run_average,
linestyle=pop_keys[entry[0]],
label="Keep %:" + str(entry[1]))
# Put a legend to the right of the current axis
ax1.legend(loc='best', fontsize=6)
ax1.grid()
plt.tight_layout()
filename = "population_and_keep_percent_rates.png"
filename = os.path.join(out_dir, filename)
plt.savefig(filename)
runs = []
times = []
for i in range(5):
mimic = MIMICRunner(problem=problem,
experiment_name="Knapsack Mimic" + "Run " +
str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 150, 5),
population_sizes=[100],
keep_percent_list=[0.5],
max_attempts=5,
generate_curves=True,
use_fast_mimic=True)
mimic_run_stats, _ = mimic.run()
mimic_run_iterations = mimic_run_stats['Iteration'].tolist()
times.append(mimic_run_stats["Time"].iloc[-1] * 1000)
fitness = np.asarray(mimic_run_stats['Fitness'].tolist())
runs.append(fitness)
mimic_run_average = np.mean(np.asarray(runs), axis=0)
mimic_run_average_time = np.mean(times)
fig, (fitness_cp, time_cp) = plt.subplots(1, 2, figsize=(6, 2))
fitness_cp.set_title("Fitness vs Iterations\n Knapsack N = 100", title_dic)
fitness_cp.set_ylabel("Average Fitness Score", title_dic)
fitness_cp.tick_params(axis="x", labelsize=axis_label_size)
fitness_cp.tick_params(axis="y", labelsize=axis_label_size)
fitness_cp.set_xlabel("Iteration", title_dic)
fitness_cp.plot(sa_run_iterations, sa_run_average, label="SA")
fitness_cp.plot(ga_run_iterations, ga_run_average, label="GA")
fitness_cp.plot(rhc_run_iterations, rhc_run_average_restarts, label="RHC")
fitness_cp.plot(mimic_run_iterations, mimic_run_average, label="MIMIC")
fitness_cp.legend(loc='best', fontsize=6)
time_cp.set_xlabel("Log Time (ms)", title_dic)
time_cp.set_title("Runtime \n Knapsack N = 100", title_dic)
time_cp.set_ylabel("Algorithm", title_dic)
time_cp.tick_params(axis="x", labelsize=axis_label_size)
time_cp.tick_params(axis="y", labelsize=axis_label_size)
x = ['RHC', 'SA', 'GA', 'MIMIC']
x_pos = [i for i, _ in enumerate(x)]
time_cp.set_yticks(x_pos)
time_cp.set_yticklabels(x)
time_cp.set_xscale('log')
time_cp.barh(x_pos, [
rhc_run_average_time, sa_run_average_time, ga_run_average_time,
mimic_run_average_time
])
time_cp.grid()
fitness_cp.grid()
plt.tight_layout()
path = KNAPSACK_PATH
filename = "fitness_vs_iteration.png"
filename = os.path.join(path, filename)
plt.savefig(filename)
fig, (fitness_cp, time_cp) = plt.subplots(1, 2, figsize=(6, 2))
fitness_cp.set_title("Fitness vs Problem Size\n Knapsack Problem",
title_dic)
fitness_cp.set_ylabel("Average Fitness Score", title_dic)
fitness_cp.tick_params(axis="x", labelsize=axis_label_size)
fitness_cp.tick_params(axis="y", labelsize=axis_label_size)
fitness_cp.set_xlabel("Problem Size(N)", title_dic)
# Running on different problem set sizes
sizes = [50, 75, 100, 125, 150]
rhc_average_size = []
sa_average_size = []
ga_average_size = []
mimic_average_size = []
rhc_average_time = []
sa_average_time = []
ga_average_time = []
mimic_average_time = []
out_dir = KNAPSACK_PATH
for size in sizes:
problem = CustomKnapSackGenerator.generate(
1,
number_of_items_types=size,
max_item_count=50,
max_weight_per_item=100,
max_value_per_item=50,
max_weight_pct=0.6,
multiply_by_max_item_count=True,
crossover=one_point_crossover.OnePointCrossOver,
mutator=change_one_mutator.ChangeOneMutator)
restarts = []
iterations = []
times = []
for i in range(15):
# run RHC runner with param restarts.
rhc = RHCRunner(problem=problem,
experiment_name="RHC QUEENS" + str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 1000, 100),
max_attempts=500,
restart_list=[0])
rhc_run_stats, _ = rhc.run()
rhc_run_iterations = rhc_run_stats['Iteration'].tolist()
restarts.append(rhc_run_stats['Fitness'].tolist())
times.append(rhc_run_stats["Time"].iloc[-1] * 1000)
rhc_run_average_restarts = np.mean(np.asarray(restarts), axis=0)
rhc_run_average_time = np.mean(times)
rhc_average_size.append(rhc_run_average_restarts[-1])
rhc_average_time.append(rhc_run_average_time)
runs = []
times = []
for i in range(5):
sa = SARunner(
problem=problem,
experiment_name=
"Knapsack Simulated Annealing after Param Tuning Decay Schedule "
+ str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 1500, 100),
max_attempts=1000,
decay_list=[mlrose_hiive.ExpDecay],
temperature_list=[3500])
sa_run_stats, _ = sa.run()
sa_run_iterations = sa_run_stats['Iteration'].tolist()
times.append(sa_run_stats["Time"].iloc[-1] * 1000)
runs.append(sa_run_stats['Fitness'].tolist())
sa_run_average = np.mean(np.asarray(runs), axis=0)
sa_run_average_time = np.mean(times)
sa_average_size.append(sa_run_average[-1])
sa_average_time.append(sa_run_average_time)
runs = []
times = []
for i in range(5):
ga = GARunner(problem=problem,
experiment_name="CPP GA After Parameter Tuning" +
"Run " + str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 1500, 25),
max_attempts=150,
population_sizes=[75],
mutation_rates=[0.9])
ga_run_stats, _ = ga.run()
ga_run_iterations = ga_run_stats['Iteration'].tolist()
times.append(ga_run_stats["Time"].iloc[-1] * 1000)
runs.append(ga_run_stats['Fitness'].tolist())
ga_run_average = np.mean(np.asarray(runs), axis=0)
ga_run_average_time = np.mean(times)
ga_average_size.append(ga_run_average[-1])
ga_average_time.append(ga_run_average_time)
runs = []
times = []
for i in range(5):
mimic = MIMICRunner(problem=problem,
experiment_name="Knapsack Mimic" + "Run " +
str(i),
output_directory=out_dir,
seed=np.random.randint(100),
iteration_list=np.arange(1, 150, 5),
population_sizes=[100],
keep_percent_list=[0.5],
max_attempts=5,
generate_curves=True,
use_fast_mimic=True)
mimic_run_stats, _ = mimic.run()
mimic_run_iterations = mimic_run_stats['Iteration'].tolist()
times.append(mimic_run_stats["Time"].iloc[-1] * 1000)
runs.append(mimic_run_stats['Fitness'].tolist())
mimic_run_average = np.mean(np.asarray(runs), axis=0)
mimic_run_average_time = np.mean(times)
mimic_average_size.append(mimic_run_average[-1])
mimic_average_time.append(mimic_run_average_time)
fitness_cp.plot(sizes, sa_average_size, label="SA")
fitness_cp.plot(sizes, ga_average_size, label="GA")
fitness_cp.plot(sizes, rhc_average_size, label="RHC")
fitness_cp.plot(sizes, mimic_average_size, label="MIMIC")
fitness_cp.legend(loc='best', fontsize=6)
time_cp.set_ylabel("Log Time (ms)", title_dic)
time_cp.set_title("Runtime vs Problem Size \n Knapsack Problem", title_dic)
time_cp.set_xlabel("Problem Size(N)", title_dic)
time_cp.tick_params(axis="x", labelsize=axis_label_size)
time_cp.tick_params(axis="y", labelsize=axis_label_size)
time_cp.set_yscale('log')
time_cp.plot(sizes, sa_average_time, label="SA")
time_cp.plot(sizes, ga_average_time, label="GA")
time_cp.plot(sizes, rhc_average_time, label="RHC")
time_cp.plot(sizes, mimic_average_time, label="MIMIC")
time_cp.legend(loc='best', fontsize=6)
time_cp.grid()
fitness_cp.grid()
plt.tight_layout()
path = KNAPSACK_PATH
filename = "fitness_vs_iteration_various_problem_sizes.png"
filename = os.path.join(path, filename)
plt.savefig(filename)
experiment_name=experiment_name,
output_directory=constants.OUTPUT_DIRECTORY,
seed=0,
iteration_list=constants.iteration_list,
max_attempts=2000,
temperature_list=[500],
decay_list=[ExpDecay])
sa_run_stats, sa_run_curves = sa.run()
if algorithm_name == "GA" and False:
ga = GARunner(problem=problem,
experiment_name=experiment_name,
output_directory=constants.OUTPUT_DIRECTORY,
seed=0,
iteration_list=2**np.arange(16),
max_attempts=2000,
population_sizes=[10],
mutation_rates=[0.05])
ga_run_stats, ga_run_curves = ga.run()
if algorithm_name == "MIMIC" and False:
mimic = MIMICRunner(
problem=problem,
experiment_name=experiment_name,
output_directory=constants.OUTPUT_DIRECTORY,
seed=0,
iteration_list=2**np.arange(16),
population_sizes=[10],
max_attempts=500,
keep_percent_list=[0.05],
use_fast_mimic=True)
mimic_run_stats, mimic_run_curves = mimic.run()