def run(client, clf, print_prf=False, print_main_proportion=False):
attr_dict, label_dict, pullinfo_list_dict = load_data(ayonel_numerical_attr=ayonel_numerical_attr, ayonel_boolean_attr=ayonel_boolean_attr,
ayonel_categorical_attr_handler=ayonel_categorical_attr_handler)
ACC = 0.0
for org, repo in org_list:
input_X = attr_dict[org]
input_y = label_dict[org]
seg_point = int(len(input_X) * SEG_PROPORTION)
train_X = np.array(input_X[:seg_point])
train_y = np.array(input_y[:seg_point])
# X_sparse = coo_matrix(train_X)
#
# train_X, X_sparse, train_y = shuffle(train_X, X_sparse, train_y, random_state=0)
# train_X, train_y = AS().fit_sample(train_X, train_y)
test_X = np.array(input_X[seg_point:])
test_y = np.array(input_y[seg_point:])
train(clf, train_X, train_y)
accuracy = clf.score(test_X, test_y)
ACC += accuracy
predict_result = clf.predict(test_X).tolist()
actual_result = test_y.tolist()
precision, recall, F1 = precision_recall_f1(predict_result, actual_result)
print(accuracy, end='')
if print_prf:
print(",%f,%f,%f" % (precision, recall, F1), end='')
if print_main_proportion:
main_proportion = predict_result.count(1) / len(predict_result)
print(',%f' % (main_proportion if main_proportion > 0.5 else 1 - main_proportion), end='')
print()
print(ACC/len(org_list))
def run(clf, print_prf=False, print_main_proportion=False):
attr_dict, label_dict = load_data(gousios_attr_list=gousios_attr_list)
ACC = 0.0
for org, repo in org_list:
input_X = attr_dict[org]
input_y = label_dict[org]
# print(sorted(Counter(input_y).items()))
seg_point = int(len(input_X) * SEG_PROPORTION)
train_X = np.array(input_X[:seg_point])
train_y = np.array(input_y[:seg_point])
train_X, train_y = SMOTE().fit_sample(train_X, train_y)
test_X = np.array(input_X[seg_point:])
test_y = np.array(input_y[seg_point:])
train(clf, train_X, train_y)
actual_result = test_y.tolist()
predict_result = clf.predict(test_X).tolist()
accuracy = clf.score(test_X, test_y)
ACC += accuracy
print(accuracy, end='')
precision, recall, F1 = precision_recall_f1(predict_result, actual_result)
if print_prf:
print(",%f,%f,%f" % (precision, recall, F1), end='')
if print_main_proportion:
main_proportion = predict_result.count(1) / len(predict_result)
print(',%f' % (main_proportion if main_proportion > 0.5 else 1 - main_proportion), end='')
print()
print(str(ACC/len(org_list)))
def run_monthly(clf, print_time=False, print_acc=False, print_prf=False, print_prf_each=False, print_main_proportion=False, print_AUC=False, MonthGAP=1):
data_dict = load_data_monthly(gousios_attr_list=gousios_attr_list, MonthGAP=MonthGAP)
for org, repo in org_list:
outfile = open("../anova/" + repo + '.csv', "a", newline='')
writer = csv.writer(outfile)
train_cost_time = 0
test_cost_time = 0
total_start_time = time.time()
print(repo, end='')
batch_iter = data_dict[org]
train_batch = batch_iter.__next__()
train_X = np.array(train_batch[0])
train_y = np.array(train_batch[1])
predict_result = []
actual_result = []
predict_result_prob = []
samples = 0
for batch in batch_iter:
if len(batch[0]) == 0: # 测试集没有数据，直接预测下一batch
continue
test_X = np.array(batch[0])
test_y = np.array(batch[1])
train_start_time = time.time()
train(clf, train_X, train_y)
train_cost_time += time.time() - train_start_time
actual_result += test_y.tolist() # 真实结果
test_start_time = time.time()
predict_result += clf.predict(test_X).tolist() # 预测结果
test_cost_time += time.time() - test_start_time
predict_result_prob += [x[0] for x in clf.predict_proba(test_X).tolist()]
samples += test_X.size
train_X = np.concatenate((train_X, test_X))
train_y = np.concatenate((train_y, test_y))
total_cost_time = time.time() - total_start_time
acc_num = 0
for i in range(len(actual_result)):
if actual_result[i] == predict_result[i]:
acc_num += 1
writer.writerow([0, 1])
else:
writer.writerow([0, 0])
if print_acc:
print(',%f' % (acc_num/len(actual_result)), end='')
if print_AUC:
y = np.array(actual_result)
pred = np.array(predict_result_prob)
fpr, tpr, thresholds = roc_curve(y, pred, pos_label=1)
AUC = auc(fpr, tpr)
print(',%f' % (AUC if AUC > 0.5 else 1-AUC), end='')
precision, recall, F1 = precision_recall_f1(predict_result, actual_result)
if print_prf:
print(",%f,%f,%f" % (precision, recall, F1), end='')
if print_prf_each:
merged_precision, merged_recall, merged_F1 = precision_recall_f1(predict_result, actual_result, POSITIVE=0)
rejected_precision, rejected_recall, rejected_F1 = precision_recall_f1(predict_result, actual_result, POSITIVE=1)
print(',%f,%f,%f,%f,%f,%f' % (merged_F1, merged_precision, merged_recall, rejected_F1,rejected_precision, rejected_recall ), end='')
if print_main_proportion:
main_proportion = predict_result.count(1) / len(predict_result)
print(',%f' % (main_proportion if main_proportion > 0.5 else 1 - main_proportion), end='')
if print_time:
print(',%f,%f,%f' % (train_cost_time, test_cost_time, total_cost_time), end='')
print()
def run_monthly(client,
clf,
print_time=True,
over_sample=False,
print_acc=False,
print_prf_each=False,
print_main_proportion=False,
print_AUC=False,
MonthGAP=1,
persistence=False):
data_dict, pullinfo_list_dict = load_data_monthly(
ayonel_numerical_attr=ayonel_numerical_attr,
ayonel_boolean_attr=ayonel_boolean_attr,
ayonel_categorical_attr_handler=ayonel_categorical_attr_handler,
MonthGAP=MonthGAP)
for org, repo in org_list:
train_cost_time = 0 # 训练耗时
test_cost_time = 0 # 预测耗时
total_start_time = time.time() # 每个项目开始时间
print(org, end='')
pullinfo_list = pullinfo_list_dict[org]
batch_iter = data_dict[org] # 获取数据集迭代器
train_batch = batch_iter.__next__() # 获取一轮数据
train_X = np.array(train_batch[0])
train_y = np.array(train_batch[1])
cursor = train_y.size # 游标，用于记录第一条开始预测pr的位置
predict_result = [] # 预测结果，0|1表示
predict_result_prob = [] # 预测结果，概率值
actual_result = [] # 实际结果，0|1表示
mean_accuracy = 0 # 平均acc
round = 1 # 轮次计数
for batch in batch_iter:
if len(batch[0]) == 0: # 测试集没有数据，直接预测下一batch
continue
test_X = np.array(batch[0])
test_y = np.array(batch[1])
# 过采样
if over_sample:
if train_y.tolist().count(0) <= 6 or train_y.tolist().count(
1) <= 6:
train(clf, train_X, train_y)
else:
resample_train_X, resample_train_y = SMOTE(
ratio='auto',
random_state=RANDOM_SEED).fit_sample(train_X, train_y)
train(clf, resample_train_X, resample_train_y)
else: # 正常算
if train_y.sum() != 0 and train_y.sum() != train_y.size:
train_start_time = time.time()
train(clf, train_X, train_y)
train_cost_time += time.time() - train_start_time # 更新训练时间
else:
train_X = np.concatenate((train_X, test_X)) # 将测试集加入训练集中
train_y = np.concatenate((train_y, test_y)) # 将测试集加入训练集中
continue
actual_result += test_y.tolist() # 真实结果
test_start_time = time.time()
predict_result += clf.predict(test_X).tolist() # 预测结果
test_cost_time += time.time() - test_start_time # 更新预测时间
predict_result_prob += [
x[0] for x in clf.predict_proba(test_X).tolist()
]
mean_accuracy += clf.score(test_X, test_y)
train_X = np.concatenate((train_X, test_X)) # 将测试集加入训练集中
train_y = np.concatenate((train_y, test_y)) # 将测试集加入训练集中
round += 1
total_cost_time = time.time() - total_start_time
acc_num = 0 # 预测正确的pr数量
for i in range(len(actual_result)):
if actual_result[i] == predict_result[i]:
acc_num += 1
# 如果需要将结果入库
if persistence:
for i in range(len(predict_result)):
number = int(pullinfo_list[cursor + i]['number'])
data = {
'number': number,
'org': org,
'repo': repo,
'created_at':
float(pullinfo_list[cursor + i]['created_at']),
'closed_at': float(pullinfo_list[cursor + i]['closed_at']),
'title': pullinfo_list[cursor + i]['title'],
'submitted_by': pullinfo_list[cursor + i]['author'],
'merged': pullinfo_list[cursor + i]['merged'],
'predict_merged': True if predict_result[i] == 0 else False
}
client[persistence_db][persistence_col].insert(
data) # 该表为原型系统使用表
if print_acc: # 如果需要打印acc
print(',%f' % (acc_num / len(actual_result)), end='')
if print_AUC: # 如果需要打印AUC
y = np.array(actual_result)
pred = np.array(predict_result_prob)
fpr, tpr, thresholds = roc_curve(y, pred)
AUC = auc(fpr, tpr)
print(',%f' % (AUC if AUC > 0.5 else 1 - AUC), end='')
if print_prf_each: # 如果需要按照拒绝类、接受类打印精准率、召回率、F1
merged_precision, merged_recall, merged_F1 = precision_recall_f1(
predict_result, actual_result, POSITIVE=0)
rejected_precision, rejected_recall, rejected_F1 = precision_recall_f1(
predict_result, actual_result, POSITIVE=1)
print(',%f,%f,%f,%f,%f,%f' %
(merged_F1, merged_precision, merged_recall, rejected_F1,
rejected_precision, rejected_recall),
end='')
if print_main_proportion: # 如果需要打印主类占比
main_proportion = predict_result.count(1) / len(predict_result)
print(',%f' % (main_proportion if main_proportion > 0.5 else 1 -
main_proportion),
end='')
if print_time: # 如果需要打印时间统计
print(',%f,%f,%f' %
(train_cost_time, test_cost_time, total_cost_time),
end='')
print()
def run_monthly(client, clf, print_prf=False, print_prf_each=False, print_main_proportion=False, print_AUC=False, MonthGAP=1, persistence=False, ayonel_numerical_attr=None, ayonel_boolean_attr=None):
data_dict, pullinfo_list_dict = load_data_monthly(ayonel_numerical_attr=ayonel_numerical_attr, ayonel_boolean_attr=ayonel_boolean_attr,
ayonel_categorical_attr_handler=ayonel_categorical_attr_handler, MonthGAP=MonthGAP)
accuracy = 0
AUC = 0
mean_merged_precision, mean_merged_recall, mean_merged_F1, \
mean_rejected_precision, mean_rejected_recall, mean_rejected_F1 = 0, 0, 0, 0, 0, 0
for org, repo in org_list:
# print(org+",", end='')
pullinfo_list = pullinfo_list_dict[org]
batch_iter = data_dict[org]
train_batch = batch_iter.__next__()
train_X = np.array(train_batch[0])
train_y = np.array(train_batch[1])
cursor = train_y.size # 游标，用于记录第一条开始预测pr的位置
predict_result = []
predict_result_prob = []
actual_result = []
mean_accuracy = 0
for batch in batch_iter:
if len(batch[0]) == 0: # 测试集没有数据，直接预测下一batch
continue
test_X = np.array(batch[0])
test_y = np.array(batch[1])
# 正常训练
train(clf, train_X, train_y)
actual_result += test_y.tolist() # 真实结果
predict_result += clf.predict(test_X).tolist() # 预测结果
predict_result_prob += [x[0] for x in clf.predict_proba(test_X).tolist()]
mean_accuracy += clf.score(test_X, test_y)
train_X = np.concatenate((train_X, test_X))
train_y = np.concatenate((train_y, test_y))
acc_num = 0
for i in range(len(actual_result)):
if actual_result[i] == predict_result[i]:
acc_num += 1
# 需要将结果入库
if persistence:
for i in range(len(predict_result)):
number = int(pullinfo_list[cursor + i]['number'])
data = {
'number': number,
'org': org,
'repo': repo,
'created_at': float(pullinfo_list[cursor + i]['created_at']),
'closed_at': float(pullinfo_list[cursor + i]['closed_at']),
'title': pullinfo_list[cursor + i]['title'],
'submitted_by': pullinfo_list[cursor + i]['author'],
'merged': pullinfo_list[cursor + i]['merged'],
'predict_merged': True if predict_result[i] == 0 else False
}
client[persistence_db][persistence_col].insert(data)
accuracy += acc_num / len(actual_result)
if print_prf_each:
try:
merged_precision, merged_recall, merged_F1 = precision_recall_f1(predict_result, actual_result, POSITIVE=0)
if mean_merged_precision == -1 or mean_merged_recall == -1 or mean_merged_F1 == -1:
pass
else:
mean_merged_precision += merged_precision
mean_merged_recall += merged_recall
mean_merged_F1 += merged_F1
except ValueError:
mean_merged_precision, mean_merged_recall, mean_merged_F1 = -1, -1, -1
try:
rejected_precision, rejected_recall, rejected_F1 = precision_recall_f1(predict_result, actual_result, POSITIVE=1)
if mean_rejected_precision == -1 or mean_rejected_recall == -1 or mean_rejected_F1 == -1:
pass
else:
mean_rejected_precision += rejected_precision
mean_rejected_recall += rejected_recall
mean_rejected_F1 += rejected_F1
except ValueError:
mean_rejected_precision, mean_rejected_recall, mean_rejected_F1 = -1, -1, -1
# print(',%f,%f,%f,%f,%f,%f' % (merged_F1, merged_precision, merged_recall, rejected_F1,rejected_precision, rejected_recall ), end='')
if print_main_proportion:
main_proportion = predict_result.count(1) / len(predict_result)
print(',%f' % (main_proportion if main_proportion > 0.5 else 1 - main_proportion), end='')
if print_AUC:
y = np.array(actual_result)
pred = np.array(predict_result_prob)
fpr, tpr, thresholds = roc_curve(y, pred)
this_AUC = auc(fpr, tpr)
AUC += this_AUC if this_AUC > 0.5 else 1 - this_AUC
# print(',%f' % (AUC if AUC > 0.5 else 1 - AUC), end='')
# print()
if mean_merged_precision == -1 or mean_merged_recall == -1 or mean_merged_F1 == -1 \
or mean_rejected_precision == -1 or mean_rejected_recall == -1 or mean_rejected_F1 == -1:
return accuracy/len(org_list), AUC/len(org_list), -1, -1, -1, -1, -1, -1
else:
return accuracy / len(org_list), AUC / len(org_list), \
mean_merged_precision/len(org_list), \
mean_merged_recall/len(org_list), \
mean_merged_F1/len(org_list), \
mean_rejected_precision/len(org_list), \
mean_rejected_recall/len(org_list), \
mean_rejected_recall/len(org_list)