Losses

RelevanceLossBase

class ml4ir.base.model.losses.loss_base.RelevanceLossBase

Bases: object

Abstract class that defines the loss and final activation function used to train a RelevanceModel

get_loss_fn(**kwargs)

Returns the loss function _loss_fn()

Parameters:kwargs (dict) – Additional key value arguments can be passed as needed. For example, metadata features can be passed to compute custom losses
Returns:Loss function that computes the loss from predicted scores and true labels
Return type:function
get_final_activation_op()

Returns the final activation layer

Returns:Final activation function that is applied on the scores before computing losses
Return type:function

SigmoidCrossEntropy

class ml4ir.applications.ranking.model.losses.pointwise_losses.SigmoidCrossEntropy(loss_key='pointwise', scoring_type='')

Bases: ml4ir.applications.ranking.model.losses.loss_base.PointwiseLossBase

Instantiate a PointwiseLossBase object

Parameters:
  • loss_key (str) – Name of the loss function used
  • scoring_type (str) – Type of scoring function - pointwise, pairwise, groupwise
get_loss_fn(**kwargs)

Define a sigmoid cross entropy loss Additionally can pass in record positions to handle positional bias

Returns:Function to compute sigmoid cross entropy loss
Return type:function

Notes

Uses mask field to exclude padded records from contributing to the loss

get_final_activation_op(output_name)

Define a sigmoid activation function

Parameters:output_name (str) – Name of the output score node in the tensorflow model
Returns:Function to apply sigmoid activation to the output score
Return type:function

RankOneListNet

class ml4ir.applications.ranking.model.losses.listwise_losses.RankOneListNet(loss_key='listwise', scoring_type='')

Bases: ml4ir.applications.ranking.model.losses.listwise_losses.SoftmaxCrossEntropy

Instantiate a ListwiseLossBase object

Parameters:
  • loss_key (str) – Name of the loss function used
  • scoring_type (str) – Type of scoring function - pointwise, pairwise, groupwise
get_loss_fn(**kwargs)

Define a masked rank 1 ListNet loss. This loss is useful for multi-label classification when we have multiple click labels per document. This is because the loss breaks down the comparison between y_pred and y_true into individual binary assessments.

Ref -> https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/tr-2007-40.pdf

Returns:Function to compute top 1 listnet loss
Return type:function

Notes

Uses mask field to exclude padded records from contributing to the loss

CategoricalCrossEntropy

class ml4ir.applications.classification.model.losses.categorical_cross_entropy.CategoricalCrossEntropy

Bases: ml4ir.base.model.losses.loss_base.RelevanceLossBase

get_loss_fn(**kwargs)

Define a categorical cross entropy loss

Returns:Categorical cross entropy loss function
Return type:function
get_final_activation_op(output_name)

Define softmax activation function

Parameters:output_name (str) – Name of the output to use for final activation layer
Returns:Softmax activation function
Return type:function