| 12345678910111213141516171819202122232425262728293031323334 |
- from sklearn.neural_network import MLPClassifier
- from datasets import Datasets
- from sklearn.model_selection import cross_val_score
- from sklearn.feature_extraction.text import CountVectorizer
- def main():
- # 加载ADFA-LD 数据
- x1, y1 = Datasets.load_adfa_normal()
- x2, y2 = Datasets.load_adfa_attack(r"Java_Meterpreter_\d+/UAD-Java-Meterpreter*")
- x = x1 + x2
- y = y1 + y2
- # 词袋特征
- cv = CountVectorizer(min_df=1)
- x = cv.fit_transform(x).toarray()
- mlp = MLPClassifier(hidden_layer_sizes=(150, 50), max_iter=10000, alpha=1e-4, solver='sgd', tol=1e-4,
- random_state=1, learning_rate_init=.01)
- # hidden_layer_sizes：第i个元素表示第i个隐藏层中的神经元数量。
- # slover：{‘lbfgs’，‘sgd’，‘adam’}，默认’adam’。权重优化的求解器：'lbfgs’是准牛顿方法族的优化器；'sgd’指的是随机梯度下降。'adam’是指由Kingma
- # alpha：L2惩罚（正则化项）参数
- # random_state：默认无随机数生成器的状态或种子
- # max_iter：最大迭代次数
- # verbose：是否将进度消息打印到stdout
- # 优化的容忍度，容差优化
- # learning_rate_init：初始学习率
- scores = cross_val_score(mlp, x, y, cv=10, scoring="accuracy")
- print(scores.mean()) # 0.9654934210526316
- if __name__ == "__main__":
- main()
|
