1

我的简化模型如下所示:

model = Sequential()
model.add(LSTM(12, input_shape=(1000,12)))
model.add(Dense(9, activation='sigmoid'))

我的训练数据具有以下形状:

(900,1000,12)

从输出层可以看出,我有 9 个输出,因此每个信号(长度为 1000)将被分类为一个或多个此输出(这是一个多标签分类)

我这样训练我的模型:

history = model.fit(X_train,y_train, batch_size=32, epochs=10,validation_data=(X_val,y_val),verbose=2)

所以到目前为止一切正常,但现在我想用 Lime 来解释分类

explainer = lime_tabular.RecurrentTabularExplainer(X_train, training_labels=y_train,feature_names=['1','2','3','4','5','6','7','8','9','10','11','12'],
                                                   discretize_continuous=True,
                                                   class_names=['a','b','c','d','e','f','g','h','i'],
                                                   discretizer='decile')

当我定义我的解释器时,我没有收到任何错误,但是当我尝试运行下面的代码时,它运行了很长时间才给我一个错误

exp=explainer.explain_instance(data_row=X[0].reshape(1,1000,12),classifier_fn= model)
exp.show_in_notebook()
NotImplementedError: LIME does not currently support classifier models without probability scores. 
If this conflicts with your use case, please let us know: https://github.com/datascienceinc/lime/issues/16

任何人都可以识别这个错误或看看有什么问题吗?

4

1 回答 1

1

你应该传入一个分类classifier_fnexplainer.explain_instance预测概率函数,它接受一个numpy数组并输出预测概率:在你的情况下model.predict_probamodel.predict如果它产生概率也可以)。

还要注意,在您的情况下,预测概率之和不等于 1,因为您sigmoid在最后一层应用了激活。考虑切换到softmax产生总和为 1 的概率

这里是完整的例子:

拟合一个虚拟模型

X = np.random.uniform(0,1, (50, 10, 12))
y = np.random.randint(0,1, (50, 9))

model = Sequential()
model.add(LSTM(12, input_shape=(10, 12)))
model.add(Dense(9, activation='softmax'))
model.compile('adam', 'categorical_crossentropy')
history = model.fit(X, y, epochs=3)

初始化解释器

from lime import lime_tabular

explainer = lime_tabular.RecurrentTabularExplainer(
    X, training_labels = y,
    feature_names = ['1','2','3','4','5','6','7','8','9','10','11','12'],
    discretize_continuous = True,
    class_names = ['a','b','c','d','e','f','g','h','i'],
    discretizer = 'decile')

解释实例:

exp = explainer.explain_instance(
    data_row = X[0].reshape(1,10,12),
    classifier_fn = model.predict)

exp.show_in_notebook()
于 2020-04-29T23:08:44.883 回答