def naive_vs_bitset_seqscout():
time_xp = 10
for i, (data, target, enable_i) in enumerate(datasets):
seq_scout(data,
target,
time_budget=time_xp,
enable_i=enable_i,
vertical=False,
iterations_limit=2**30)
seq_scout(data,
target,
time_budget=time_xp,
enable_i=enable_i,
vertical=True,
iterations_limit=2**30)
def number_iterations_optima():
iterations_limit = 100
# if we want to average
nb_launched = 5
data_final = {'cost': [], 'iterations': [], 'dataset_name': []}
for j, (data, target, enable_i) in enumerate(datasets[:-1]):
for i in range(nb_launched):
# we reset the count of iterations
seqscout.global_var.ITERATION_NUMBER = 0
result_ucb_opti = seq_scout(data,
target,
enable_i=enable_i,
time_budget=TIME_BUDGET_XP,
iterations_limit=iterations_limit,
vertical=False)
if len(result_ucb_opti) < TOP_K:
print("Too few seqscout: {}".format(len(result_ucb_opti)))
iterations = 1000
additional_iterations = seqscout.global_var.ITERATION_NUMBER - iterations_limit
for i in range(10):
data_add_generic(data_final,
cost=additional_iterations / iterations,
iterations=iterations,
dataset_name=datasets_names[j])
iterations += 2000
df = pd.DataFrame(data=data_final)
sns.set(rc={'figure.figsize': (8, 6.5)})
plt.clf()
ax = sns.lineplot(data=df, x='iterations', y='cost', hue='dataset_name')
plt.savefig('./number_iterations_optima/it_optima.png')
df.to_pickle('./number_iterations_optima/result')
if SHOW:
plt.show()
def optimizer(data, target, alg_name, time_budget, iteration_limit, enable_i):
if alg_name == 'misere':
results = misere(data,
target,
time_budget=time_budget,
iterations_limit=iteration_limit)
elif alg_name == 'BeamSearch':
results = beam_search(data,
target,
time_budget=time_budget,
iterations_limit=iteration_limit,
enable_i=enable_i)
elif alg_name == 'SeqScout':
results = seq_scout(data,
target,
time_budget=time_budget,
iterations_limit=iteration_limit,
enable_i=enable_i)
elif alg_name == 'MCTSExtent':
results = launch_mcts(data,
target,
time_budget=time_budget,
iterations_limit=iteration_limit)
else:
print('Error algo name')
return
sorted_patterns = PrioritySet(theta=THETA)
for result in results:
sorted_patterns.add(result[1], result[0])
results_post_opti = optimize_pattern(results, extract_items(data), data,
[], target, TOP_K, sorted_patterns,
enable_i)
return results, results_post_opti