def greedy_search(pages):
model = train_models(100, ['heuristic'], [classification.get_logistic_regression_model_liblinear])[0]
total_states_explored = 0
num_with_paths = 0
total_time = 0
for i in range(100):
start_time = datetime.datetime.now()
current_article = random.sample(pages, 1)[0]
print 'start article: ', current_article
print 'goal article: ', GOAL_ARTICLE
for j in range(10):
links = pages[current_article][1]
min_cost = 1000000
for link in links:
cost = classification.apply_model([pages[link][2]], model)
if cost < min_cost:
print 'current min cost: ', min_cost
min_cost = cost
current_article = link
if current_article == GOAL_ARTICLE:
break
end_time = datetime.datetime.now()
total_states_explored += j
print 'dist: ',j
if j == 9:
continue
num_with_paths += 1
total_time += int((end_time - start_time).microseconds)
print 'av states explored:', float(total_states_explored)/(i+1)
print 'percent with paths:', 100*float(num_with_paths)/(i+1), '%'
print 'av time:', float(total_time)/(i+1)
def test_models(num_testing_examples, models):
print 'generating testing data'
training_data = {}
for i in range(num_testing_examples):
if i % 10 == 0:
print 'generated', i, 'examples'
start_article = random.sample(pages, 1)[0]
search_prob = ucs.SearchProblem(pages, start_article, GOAL_ARTICLE)
ucs_prob = ucs.UniformCostSearch()
ucs_prob.solve(search_prob)
if ucs_prob.totalCost is None:
training_data[start_article] = INFINITE_COST
continue
num_actions = len(ucs_prob.actions)
training_data[start_article] = num_actions
x = []
y = []
for key, val in training_data.iteritems():
x.append(pages[key][2])
y.append(val)
results = {}
for i, model in enumerate(models):
print 'applying model', i
classifications = classification.apply_model(x, model)
correct_count = 0
reachable_count = 0
wrong_inf_count = 0
dist = 0
for j in range(len(y)):
if y[j] == INFINITE_COST:
continue
reachable_count += 1
if y[j] == classifications[j]:
correct_count += 1
else:
if y[j] == INFINITE_COST or classifications[j] == INFINITE_COST:
wrong_inf_count += 1
else:
dist += abs(y[j] - classifications[j])
results[model] = (correct_count, dist, wrong_inf_count, reachable_count)
print type(model)
print 'fully correct', 100 * float(correct_count) / reachable_count, '%'
print 'dist:', float(dist) / reachable_count
print 'wrong inf.s:', 100 * float(wrong_inf_count) / reachable_count, '%'
print ''
return results
def heuristic(link):
return classification.apply_model([pages[link][2]], model)
def h(v):
guess = 1000000*classification.apply_model([pages[pv[v]][2]], model)
#print guess
return guess