在Keras中绘制val_loss和loss得到的数字不正确

Question

我正在运行下面的一段代码，以便在Keras中绘制列车和测试损失曲线。

train_x, test_x, train_y, test_y = train_test_split(features, target_classes , test_size=0.30)

epochs = 4
batch_size = 256

for i in range(epochs):
    print("epoch Value", i)

    ix_train = np.random.choice(train_x.shape[0], size=batch_size)

    score = model.fit(
        train_x[ix_train], train_y[ix_train]
        , epochs=1
        , validation_data=(test_x, test_y)
    )

scores.append(score)

//此代码用于绘制val_loss和train_loss

for i in range(0, len(scores)):
    val_loss_change.append(scores[i].history['val_loss'])
    loss_change.append(scores[i].history['loss'])

plt.plot(val_loss_change, label='val_loss')
plt.plot(loss_change, label='train_loss')
plt.legend(loc='upper right')
plt.show()
plt.savefig('LossVal_loss')

生成绘图时，打印的值看起来并不真实。请参阅附图LossVal_loss。我已经从Python Debug prompt复制了val_loss_change和loss_change。

loss_change=：[4.3783984780311584，3.9744645059108734，3.921104222536087，3.5381810665130615，3.3796855211257935，3.161308079957962，2.9224385917186737，2.80639386177063，2.5576193928718567，2.1081390380859375]

val_loss_change =：[4.315125052134196，4.105147279103597，4.0108651924133305，3.9794070688883463，4.025013980865478，4.060481491088868，4.1542660458882645，4.011785678863525，3.989632488886515，4.240501753489176]

当我简单地从pyhton Debug提示符中复制val_loss_change和loss_change并创建一个新的python文件并尝试运行它时，如下代码所示。绘制的图是正确的。请参阅附图LossVal_loss1

val_loss_change = [[4.315125052134196], [4.105147279103597], [4.0108651924133305], [3.9794070688883463], [4.025013980865478], [4.060481491088868], [4.1542660458882645], [4.011785678863525], [3.989632488886515], [4.240501753489176]]
loss_change = [[4.3783984780311584], [3.9744645059108734], [3.921104222536087], [3.5381810665130615], [3.3796855211257935], [3.161308079957962], [2.9224385917186737], [2.80639386177063], [2.5576193928718567], [2.1081390380859375]]

plt.plot(val_loss_change, label='val_loss')
plt.plot(loss_change, label='train_loss')
plt.legend(loc='upper right')
plt.show()
plt.savefig('LossVal_loss1')





Can anyone tell what goes wrong in 1st code?
I want to run fit function multiple times and then plot a curve for loss and val_loss.




  [1 Correct Figure from 2nd code ]: https://i.stack.imgur.com/fVg4o.png
  [2 Incorrect Figure from 1st Code ]: https://i.stack.imgur.com/dcRSE.png

Answer 1

您正在一次又一次地使用1个纪元拟合模型。将 epochs 更改为epochs的总值，并删除额外的循环。我在手机上，所以我现在不能测试代码。

要获得更好的动态图形，请使用TensorBoard