使用GAN进行异常检测——可以进行网络流量的自学习哇，哥哥，人家是半监督，无监督的话，还是要VAE，SAE。

实验了效果，下面的还是图像的异常检测居多。

https://github.com/LeeDoYup/AnoGAN

https://github.com/tkwoo/anogan-keras

看了下，本质上是半监督学习，一开始是有分类模型的。代码如下，生产模型和判别模型：

### generator model define
def generator_model():
    inputs = Input((10,))
    fc1 = Dense(input_dim=10, units=128*7*7)(inputs)
    fc1 = BatchNormalization()(fc1)
    fc1 = LeakyReLU(0.2)(fc1)
    fc2 = Reshape((7, 7, 128), input_shape=(128*7*7,))(fc1)
    up1 = Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same')(fc2)
    conv1 = Conv2D(64, (3, 3), padding='same')(up1)
    conv1 = BatchNormalization()(conv1)
    conv1 = Activation('relu')(conv1)
    up2 = Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same')(conv1)
    conv2 = Conv2D(1, (5, 5), padding='same')(up2)
    outputs = Activation('tanh')(conv2)
     
    model = Model(inputs=[inputs], outputs=[outputs])
    return model
 
### discriminator model define
def discriminator_model():
    inputs = Input((28, 28, 1))
    conv1 = Conv2D(64, (5, 5), padding='same')(inputs)
    conv1 = LeakyReLU(0.2)(conv1)
    pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
    conv2 = Conv2D(128, (5, 5), padding='same')(pool1)
    conv2 = LeakyReLU(0.2)(conv2)
    pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
    fc1 = Flatten()(pool2)
    fc1 = Dense(1)(fc1)
    outputs = Activation('sigmoid')(fc1)
     
    model = Model(inputs=[inputs], outputs=[outputs])
    return model