tfa.losses.NpairsMultilabelLoss

View source on GitHub

Computes the npairs loss between multilabel data y_true and y_pred.

Npairs loss expects paired data where a pair is composed of samples from the same labels and each pairs in the minibatch have different labels. The loss takes each row of the pair-wise similarity matrix, y_pred, as logits and the remapped multi-class labels, y_true, as labels.

To deal with multilabel inputs, the count of label intersection is computed as follows:

L_{i,j} = | set_of_labels_for(i) \cap set_of_labels_for(j) |

Each row of the count based label matrix is further normalized so that each row sums to one.

y_true should be a binary indicator for classes. That is, if y_true[i, j] = 1, then ith sample is in jth class; if y_true[i, j] = 0, then ith sample is not in jth class.

The similarity matrix y_pred between two embedding matrices a and b with shape [batch_size, hidden_size] can be computed as follows:

# y_pred = a * b^T
y_pred = tf.matmul(a, b, transpose_a=False, transpose_b=True)

See: http://www.nec-labs.com/uploads/images/Department-Images/MediaAnalytics/papers/nips16_npairmetriclearning.pdf

name (Optional) name for the loss.

reduction (Optional) 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 with tf.distribute.Strategy, outside of built-in training loops such as tf.keras compile and 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 op.

Methods

from_config

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

Args
config Output of get_config().

Returns
A Loss instance.

get_config

Returns the config dictionary for a Loss instance.

__call__

Invokes the Loss instance.

Args
y_true Ground truth values. shape = [batch_size, d0, .. dN], except sparse loss functions such as sparse categorical crossentropy where shape = [batch_size, d0, .. dN-1]
y_pred The predicted values. shape = [batch_size, d0, .. dN]
sample_weight Optional sample_weight acts as a coefficient for the loss. If a scalar is provided, then the loss is simply scaled by the given value. If sample_weight is a tensor of size [batch_size], then the total loss for each sample of the batch is rescaled by the corresponding element in the sample_weight vector. If the shape of sample_weight is [batch_size, d0, .. dN-1] (or can be broadcasted to this shape), then each loss element of y_pred is scaled by the corresponding value of sample_weight. (Note ondN-1: all loss functions reduce by 1 dimension, usually axis=-1.)

Returns
Weighted loss float Tensor. If reduction is NONE, this has shape [batch_size, d0, .. dN-1]; otherwise, it is scalar. (Note dN-1 because all loss functions reduce by 1 dimension, usually axis=-1.)

Raises
ValueError If the shape of sample_weight is invalid.