为什么 Epoch 显示相同的准确性？

Question

我正在尝试构建 IDS 入侵检测系统并尝试预测标签是良性还是 DDos。但我在各个时期都得到了相同的准确度。

代码：

        from tensorflow import keras
    import numpy as np
    import datetime
    import time
    from tensorflow.keras.optimizers import Adam
    from keras.models import Sequential
    from keras.layers import Dense, Dropout
    from keras import callbacks
    x=pd.DataFrame(X)
    x = x.values
    sample = x.shape[0]
    features = x.shape[1]
    #Train: convert 2D to 3D for input RNN
    x_train = np.reshape(x,(sample,features,1)) #shape  = (125973, 18, 1)
    #Test: convert 2D to 3D for input RNN
    x_test=pd.DataFrame(X_test)
    x_test = x_test.values
    x_test = np.reshape(x_test,(x_test.shape[0],x_test.shape[1],1))
Model = keras.Sequential([

        keras.layers.LSTM(80,input_shape=(features,x_train.shape[2]),
                          activation='sigmoid',recurrent_activation='hard_sigmoid'),
        keras.layers.Dense(1,activation="softmax")
    ])

Model.compile(optimizer='rmsprop',loss='mse', metrics=['accuracy'])

#Training the model

Model.fit(x_train, y, epochs=10, batch_size= 32) 
Model.summary()

# Final evaluation of the model
scores = Model.evaluate(x_test, y_test, verbose=0)
print('/n')
print("Accuracy: %.2f%%" % (scores[1]*100))



Epoch 1/10
1074/1074 [==============================] - 92s 83ms/step - loss: 0.4180 - accuracy: 0.5820
Epoch 2/10
1074/1074 [==============================] - 79s 74ms/step - loss: 0.4180 - accuracy: 0.5820
Epoch 3/10
1074/1074 [==============================] - 81s 76ms/step - loss: 0.4180 - accuracy: 0.5820

解决办法是什么？

Answer 1

因为"softmax"激活有1个神经元总是输出1。你的神经元不能调整它的输出来减少损失；它在数学上只能返回 1。