Python keras自定义公制和样本重量

我试图在Keras中定义一个考虑样本权重的自定义度量。拟合模型时，我使用样本权重，如下所示：

training_history = model.fit(
        train_data,
        train_labels,
        sample_weight = train_weights,
        epochs = num_epochs,
        batch_size = 128,
        validation_data = (validation_data, validatation_labels, validation_weights ),
    )

from keras import backend as K
import tensorflow as tf

def auc(true_labels, predictions, weights = None):
    auc = tf.metrics.auc(true_labels, predictions, weights = weights)[1]
    K.get_session().run(tf.local_variables_initializer())
    return auc

model.compile(
    optimizer = optimizer,
    loss = 'binary_crossentropy',
    metrics = ['accuracy', auc(sample_weights)]
)

我使用的自定义度量的一个示例是AUC（roc曲线下的面积），我将其定义如下：

training_history = model.fit(
        train_data,
        train_labels,
        sample_weight = train_weights,
        epochs = num_epochs,
        batch_size = 128,
        validation_data = (validation_data, validatation_labels, validation_weights ),
    )

from keras import backend as K
import tensorflow as tf

def auc(true_labels, predictions, weights = None):
    auc = tf.metrics.auc(true_labels, predictions, weights = weights)[1]
    K.get_session().run(tf.local_variables_initializer())
    return auc

model.compile(
    optimizer = optimizer,
    loss = 'binary_crossentropy',
    metrics = ['accuracy', auc(sample_weights)]
)

我在编译模型时使用了这个度量：

model.compile(
        optimizer = optimizer,
        loss = 'binary_crossentropy',
        metrics = ['accuracy', auc]
    )

---

但据我所知，该指标并未考虑样本权重。事实上，我通过将使用上面定义的自定义度量训练模型时看到的度量值与我自己从模型输出和样本权重计算得到的度量值进行比较来验证这一点，这确实产生了非常不同的结果。如何定义上面显示的auc度量以考虑样本权重？

您可以使用另一个函数包装度量，该函数将

样本权重作为参数：

def auc(weights):
    def metric(true_labels, predictions):
        auc = tf.metrics.auc(true_labels, predictions, weights=weights)[1]
        K.get_session().run(tf.local_variables_initializer())
        return auc
    return metric

然后定义一个额外的输入占位符，该占位符将接收样本权重：

sample_weights = Input(shape=(1,))

然后可以按如下方式编译您的模型：

training_history = model.fit(
        train_data,
        train_labels,
        sample_weight = train_weights,
        epochs = num_epochs,
        batch_size = 128,
        validation_data = (validation_data, validatation_labels, validation_weights ),
    )

from keras import backend as K
import tensorflow as tf

def auc(true_labels, predictions, weights = None):
    auc = tf.metrics.auc(true_labels, predictions, weights = weights)[1]
    K.get_session().run(tf.local_variables_initializer())
    return auc

model.compile(
    optimizer = optimizer,
    loss = 'binary_crossentropy',
    metrics = ['accuracy', auc(sample_weights)]
)

注意：未测试。
参见“加权指标”：