def hill_climbing(model, choose_list=[]):
chooses = choose_list
feature_list = model.feature_list
visited = [
True if x in choose_list else False for x in range(len(feature_list))
]
for idx in range(len(choose_list), len(feature_list)):
chooseIndex = -1
best_score = 0.0
best_test_score = 0.0
chooses.append(-1)
for i in range(len(feature_list)):
if visited[i] == False:
chooses[idx] = i
feature = [feature_list[s] for s in chooses]
# print(len(feature_list))
model.feature_list = feature
train_nlpcc(model)
cur_score = dev_nlpcc(model)
test_score = test_nlpcc(model)
stst.record('./data/records.csv', cur_score, test_score, model)
if best_score < cur_score:
chooseIndex = i
best_score = cur_score
best_test_score = test_score
chooses[idx] = chooseIndex
visited[chooseIndex] = True
# feature = [ feature_list[s] for s in chooses]
print('Best Score: %.2f %%, %.2f%%,choose Feature %s' %
(best_score * 100, best_test_score * 100,
feature_list[chooseIndex].feature_name))
model.add(stst.NegativeFeature())
# train and test
train_file = './data/stsbenchmark/sts-train.csv'
dev_file = './data/stsbenchmark/sts-dev.csv'
test_file = './data/stsbenchmark/sts-test.csv'
# init the server and input the address
nlp = stst.StanfordNLP('http://localhost:9000')
# parse data
train_instances = stst.load_parse_data(train_file, nlp)
dev_instances = stst.load_parse_data(dev_file, nlp)
# train and test
model.train(train_instances, train_file)
model.test(dev_instances, dev_file)
# evaluation
dev_pearsonr = stst.eval_output_file(model.output_file)
print('Dev:', dev_pearsonr)
# test on new data set
test_instances = stst.load_parse_data(test_file, nlp)
model.test(test_instances, test_file)
test_pearsonr = stst.eval_output_file(model.output_file)
print('Test:', test_pearsonr)
recod_file = './data/records.csv'
stst.record(recod_file, dev_pearsonr, test_pearsonr, model)