如何从dbn分类器训练中绘制实时折线图

Question

我正在开发一个使用DBN算法的人脸识别系统。训练数据时，系统会根据n纪元产生错误。我想做一个基于n纪元的错误曲线图

训练代码

    classifier = SupervisedDBNClassification (hidden_layers_structure=[200, 200],
    learning_rate_rbm=0.0001, 
    learning_rate=0.01,
    n_epochs_rbm=10,
    n_iter_backprop=100,
    batch_size=32,
    activation_function='relu',
    dropout_p = 0.2)

如果我们运行该代码，它将生成

    >> Epoch 84 finished    ANN training loss 0.681700
    >> Epoch 85 finished    ANN training loss 0.682314
    >> Epoch 86 finished    ANN training loss 0.680272
    >> Epoch 87 finished    ANN training loss 0.680542

Answer 1

使用history = classifier.fit(x_train, y_train)

# list all data in history
print(history.history.keys())

您可以访问以下所有信息来绘制折线图

history.history['acc'])
history.history['val_acc']
history.history['loss'])
history.history['val_loss']

有关更多信息，请查看此link