tfr.keras.losses.OrdinalLoss

Computes the Ordinal loss between y_true and y_pred.

tfr.keras.losses.OrdinalLoss(
    reduction: tf.losses.Reduction = tf.losses.Reduction.AUTO,
    name: Optional[str] = None,
    ragged: bool = False,
    ordinal_size: int = 1,
    use_fraction_label: bool = False
)

In ordinal loss, y_pred is a 3D tensor with the last dimension equals to ordinal_size.

loss = -\sum_i=0^ordinal_size-1 I_{y_true > i} log(sigmoid(y_pred[i])) +
    I_{y_true <= i} log(1-sigmoid(y_pred[i]))

Standalone usage:

y_true = [[1., 0.]]
y_pred = [[[0.6, 0.2], [0.8, 0.3]]]
loss = tfr.keras.losses.OrdinalLoss(ordinal_size=2)
loss(y_true, y_pred).numpy()
1.6305413

# Using ragged tensors
y_true = tf.ragged.constant([[2., 1.], [0.]])
y_pred = tf.ragged.constant([[[0.6, 0.2], [0.8, 0.3]], [[0., -0.2]]])
loss = tfr.keras.losses.OrdinalLoss(ordinal_size=2, ragged=True)
loss(y_true, y_pred).numpy()
0.88809216

Usage with the compile() API:

model.compile(optimizer='sgd',
              loss=tfr.keras.losses.OrdinalLoss(ordinal_size=2))

Definition:

\[ \mathcal{L}(\{y\}, \{s\}) = - \sum_i\sum_{j=0}^{m-1} I_{y_i > j} \log(\text{sigmoid}(s_{i,j})) + I_{y_i \leq j} \log(1 - \text{sigmoid}(s_{i,j})) \]

Args
`reduction`	Type of `tf.keras.losses.Reduction` to apply to loss. Default value is `AUTO`. `AUTO` indicates that the reduction option will be determined by the usage context. For almost all cases this defaults to `SUM_OVER_BATCH_SIZE`. When used under a `tf.distribute.Strategy`, except via `Model.compile()` and `Model.fit()`, using `AUTO` or `SUM_OVER_BATCH_SIZE` will raise an error. Please see this custom training tutorial for more details.
`name`	Optional name for the instance.

Args

reduction Type of tf.keras.losses.Reduction to apply to loss. Default value is AUTO. AUTO indicates that the reduction option will be determined by the usage context. For almost all cases this defaults to SUM_OVER_BATCH_SIZE. When used under a tf.distribute.Strategy, except via Model.compile() and Model.fit(), using AUTO or SUM_OVER_BATCH_SIZE will raise an error. Please see this custom training tutorial for more details.

name Optional name for the instance.

Methods

`from_config`

@classmethod
from_config(
    config
)

Instantiates a Loss from its config (output of get_config()).

Args
`config`	Output of `get_config()`.

Returns
A `Loss` instance.

`get_config`

View source

get_config() -> Dict[str, Any]

Returns the config dictionary for a Loss instance.

`call`

View source

__call__(
    y_true: tfr.keras.model.TensorLike,
    y_pred: tfr.keras.model.TensorLike,
    sample_weight: Optional[utils.TensorLike] = None
) -> tf.Tensor

See tf.keras.losses.Loss.