keras CategoricalAccuracy速查

Accuracy metrics

tf.keras.metrics.Accuracy(name="accuracy", dtype=None)

Calculates how often predictions equal labels.

This metric creates two local variables, total and count that are used to compute the frequency with which y_pred matches y_true. This frequency is ultimately returned as binary accuracy: an idempotent operation that simply divides total by count.

If sample_weight is None, weights default to 1. Use sample_weight of 0 to mask values.

Arguments

• name: (Optional) string name of the metric instance.
• dtype: (Optional) data type of the metric result.

CategoricalAccuracy class

tf.keras.metrics.CategoricalAccuracy(name="categorical_accuracy", dtype=None)

Calculates how often predictions match one-hot labels.

You can provide logits of classes as y_pred, since argmax of logits and probabilities are same.

This metric creates two local variables, total and count that are used to compute the frequency with which y_pred matches y_true. This frequency is ultimately returned as categorical accuracy: an idempotent operation that simply divides total by count.

y_pred and y_true should be passed in as vectors of probabilities, rather than as labels. If necessary, use tf.one_hot to expand y_true as a vector.

If sample_weight is None, weights default to 1. Use sample_weight of 0 to mask values.

Arguments

• name: (Optional) string name of the metric instance.
• dtype: (Optional) data type of the metric result.

BinaryAccuracy class

tf.keras.metrics.BinaryAccuracy(
    name="binary_accuracy", dtype=None, threshold=0.5
)

Calculates how often predictions match binary labels.

This metric creates two local variables, total and count that are used to compute the frequency with which y_pred matches y_true. This frequency is ultimately returned as binary accuracy: an idempotent operation that simply divides total by count.

If sample_weight is None, weights default to 1. Use sample_weight of 0 to mask values.

Arguments

• name: (Optional) string name of the metric instance.
• dtype: (Optional) data type of the metric result.
• threshold: (Optional) Float representing the threshold for deciding whether prediction values are 1 or 0.

搬运画重点，compile的metrics仅用“accuracy”即可。loss这个 CategoricalAccuracy要one hot形式的label，sparse_categorical_accuracy直接label即可。

model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

Source

def categorical_accuracy(y_true, y_pred):
    return K.cast(K.equal(K.argmax(y_true, axis=-1),
                          K.argmax(y_pred, axis=-1)),
                  K.floatx())


def sparse_categorical_accuracy(y_true, y_pred):
    return K.cast(K.equal(K.max(y_true, axis=-1),
                          K.cast(K.argmax(y_pred, axis=-1), K.floatx())),
K.floatx())