Question

我已经为100个时期训练了自动编码器，批量大小为32。

当我使用model.fit（）时，我没有直接获得每个时期的整体准确性和损失，但是我看到了网络中各个层（例如density_1，density_2，reshape_1等）的准确性和损失。因此，我不确定该网络是否真的经过了100个时期的训练。

任何人都可以告诉我如何跟踪每个时期的总体网络准确性/损失吗？我的网络很复杂，有2个输入，2个输出以及介于它们之间的一些连接层。

def all_in_one_model(embeddedOutput,charMatrixList):
    embeddedInput = Input(shape=(1,400))
    #charInput = Input(shape=(40774,32,30))
    charInput = Input(shape=(32,30,1))
    p20 = Sequential()(charInput)
    p21 = Conv2D(32, (3, 3), strides=(1,1) , padding="same", activation="relu",input_shape=(32,30,1))(p20)
    print('p21 shape')
    print(p21.shape)
    #pp = Reshape((40774,32*32),name='reshape_1')(p21)
    pp = Reshape((32,960))(p21)
    p22 = LSTM(16, return_sequences=True,input_shape=(32,32))(pp)
    p23 = LSTM(16, return_sequences=True)(p22)
    p24 = Flatten()(p23)
    print('p24 shape')
    print(p24.shape)

    p10 = Sequential()(embeddedInput)
    p11 = LSTM(64, return_sequences=True, input_shape=(1,400))(p10) #Must be (400,1)
    p12 = LSTM(128, return_sequences=True)(p11)
    p13 = Flatten()(p12)
    print('p13 shape')
    print(p13.shape)

    x = Concatenate()([p13, p24])
    #print(x.shape)
    y = Dense(400,activation='relu',name = 'dense_1')(x)
    print('y shape')
    print(y.shape)
    pp = Reshape((1,400),name='reshape_2')(y)
    embOut = Dense(400,activation='relu',name = 'dense_2')(pp)
    print('Emb out shape')
    print(embOut.shape)
    #qq = Reshape((40774,960))(y)
    charMatOut = Dense(960,activation='relu',name = 'dense_3')(y)
    print('Char matrix output shape')
    print(charMatOut.shape)
    #charout = np.reshape(np.array(charMatOut),(32,30,1))
    charout = Reshape((32,30,1),name='reshape_3')(charMatOut)
    print('Reshaped Char matrix output shape')
    print(charout.shape)
    model = Model(inputs=[embeddedInput, charInput], outputs=[embOut, charout])
    model.summary()
    model.compile(loss='mean_squared_error', optimizer='adam',metrics=['accuracy'])
    plot_model(model, to_file = "model.png")
    return model


model = all_in_one_model(embeddedOutput,charMatrixList)
model.fit([npeo, npcm], [npeo, npcm], epochs=100, batch_size=32,verbose=1)

史诗1/100 40774/40774 [==============================]-188s 5ms / step-损耗：0.0031-density_2_loss：8.3277e- 04-reshape_3_loss：0.0022-density_2_acc：0.0020-reshape_3_acc：0.9976- ETA：17:16-损失：0.0058-density_2_loss：8.3114e-04-reshape_3_loss：0.0050-density_2_acc：0.0104-reshape_3_acc：0.9950-ETA：6:01-损失：0.0058-density_2_loss：8.3613e-04-reshape_3_loss：0.0050-density_2_acc：0.0024-reshape_3_acc：0.9948 992/40774 [..................................... .....]-ETA：5:28-损失：0.0058-density_2_loss：8.3601e-04-reshape_3_loss：0.0050-density_2_acc：0.0020-reshape_3_acc：0.9948-ETA：4:28-损失：0.0057-density_2_loss：8.3579e -04-reshape_3_loss：0.0049-density_2_acc：0.0027-reshape_3_acc：0.9948-ETA：4:16-损失：0.0057-density_2_loss：8.3530e-04-reshape_3_loss：0.0048-density_2_acc：0.0024-reshape_3_acc：0.9949-ETA：3:54-损失：0.0056-密度_2_损失：8.3492e-04-变形3_损失：0.0047-密度_2_acc：0.0023-形状_3_acc：0.9949 2688/40774

2输入2输出自动编码器可能未正确训练

0 个答案: