def cross_validation(scaled_X,y,classifier,K_num=2):
'''
Doing cross-validation to get the split of training and testing data with the best performance.
The spliting range from 2 to 9
'''
cross_scores = []
if classifier =='knn':
knn = KNeighborsClassifier(n_neighbors=K_num)
for i in range(2,10):
scores = cross_val_score(knn,scaled_X,y,cv =i )
cross_scores.append(scores.mean()) #Record the scores of cross validation
plt.plot(range(2,10),cross_scores)
plt.show()
elif classifier =='adacost':
Adacost=AdaCostClassifier(n_estimators=100)
for i in range(2,10):
scores = cross_val_score(Adacost,scaled_X,y,cv =i )
cross_scores.append(scores.mean())
plt.plot(range(2,10),cross_scores)
plt.show()
elif classifier =='svm':
svm=LinearSVC()
for i in range(2,10):
scores = cross_val_score(svm,scaled_X,y,cv =i )
cross_scores.append(scores.mean())
# plt.plot(range(2,10),cross_scores)
# plt.show()
return cross_scores.index(max(cross_scores))
def one_out(scaled_X, y, As_Gunshot, y_A):
'''
Leave one Africa data for testing and rest of data for training
'''
score = []
for i in range(len(As_Gunshot)):
test_x = As_Gunshot[i][np.newaxis, :]
test_y = y_A[i] #The candidates to leave out
X_train = np.delete(As_Gunshot, i, 0)
Y_train = np.delete(y_A, i)
X = np.vstack((scaled_X, X_train)).astype(
np.float64) #Combine the rest Africa data with other data
Y = np.hstack((y, Y_train)).astype(np.float64)
#split = cross_validation(X,Y,classifier = 'svm')
X_train, X_test, y_train, y_test = data_split(X, Y)
Adacost = AdaCostClassifier(n_estimators=100)
Adacost.fit(X_train[:-10], y_train[:-10])
test_x = np.vstack((test_x, X_train[-10:]))
test_y = np.hstack(
(test_y, y_train[-10:])
) #Test on the chosen Africa data and 10 other data without trained
print('Test Prediction of Adacost = ', Adacost.predict(test_x))
for i in range(len(test_x)):
if Adacost.predict(test_x[i][np.newaxis, :]) == test_y[i]:
score.append(1)
else:
score.append(0)
print('The out put of one_out is',
round(sum(score) / len(As_Gunshot) / 11, 4)) #Get the overall score
def Adacost(X_train,X_test,y_train,y_test):
Adacost = AdaCostClassifier(n_estimators=100)
t=time.time()
Adacost.fit(X_train,y_train)
t2 = time.time()
print(round(t2-t, 2), 'Seconds to train Adacost...')
# Check the score of the Adacost
print('Test Accuracy of Adacost = ', round(Adacost.score(X_test, y_test), 4))
t0 = timer()
y_pred = Adacost.predict(X_test)
print("done in %0.3fs" % (timer() - t0))
target_names = ['not_gunshot','gunshot']
print(classification_report(y_test, y_pred, labels = [-1,1],target_names=target_names))
print(confusion_matrix(y_test, y_pred, labels=[-1,1]))
acu_curve(y_pred,y_test)
df.loc[df.Class == 0, 'Class'] = 1
print(df.shape)
print(df.Class.value_counts())
return df.drop('Class', axis=1), df['Class']
if __name__ == '__main__':
# X_train, y_train = load_creditcard_data()＃输出预测数据时用
X, y= load_creditcard_data()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0) #分割数据集，将部分训练集用作测试集
# X_test= pd.read_csv('creditcard_test.csv' ＃输出预测数据时用
# ZZ=load_creditcardtest_data()
clf = AdaCostClassifier(n_estimators=100)
clf.fit(X_train, y_train)
y_pred = clf.predict(X_test)
print(pd.Series(y_pred).value_counts())
print('recall=',recall_score(y_test, y_pred, pos_label=-1),' '
'precision=',precision_score(y_test, y_pred, pos_label=-1),' '
'f1_score=',f1_score(y_test, y_pred, pos_label=-1), )
# answer = pd.read_csv(open('sample_Submission.csv')) # 输出预测csv
# for i in range(y_pred.shape[0]):
# predict = y_pred[i]
# if predict==1:
# answer.loc[i, "class"] = "1"
# else:
# answer.loc[i, "class"] = "0"
# answer.to_csv('submission1.csv', index=False)