Class CosineSimilarity<T extends TNumber>

java.lang.Object
org.tensorflow.framework.metrics.BaseMetric
org.tensorflow.framework.metrics.Mean<T>
org.tensorflow.framework.metrics.CosineSimilarity<T>
Type Parameters:
T - The data type for the metric result.
All Implemented Interfaces:
Metric
public class CosineSimilarity<T extends TNumber> extends Mean<T>
A metric that computes the cosine similarity metric between labels and predictions.

Note that it is a number between -1 and 1. When it is a negative number between -1 and 0, 0 indicates orthogonality and values closer to -1 indicate greater similarity. The values closer to 1 indicate greater dissimilarity. This makes it usable as a loss function in a setting where you try to maximize the proximity between predictions and targets. If either labels and predictions is a zero vector, cosine similarity will be 0 regardless of the proximity between predictions and targets. 

loss = -sum(l2_norm(y_true) * l2_norm(y_pred))
See Also:

  • Field Details

  • Constructor Details

    • CosineSimilarity

      public CosineSimilarity(long seed, Class<T> type)
      Creates a metric that computes the cosine similarity metric between labels and predictions with a default axis, DEFAULT_AXIS and using Class.getSimpleName() for the metric name.
      Parameters:
      seed - the seed for random number generation. An initializer created with a given seed will always produce the same random tensor for a given shape and data type.
      type - the type for the variables and result

    • CosineSimilarity

      public CosineSimilarity(int axis, long seed, Class<T> type)
      Creates a metric that computes the cosine similarity metric between labels and predictions using Class.getSimpleName() for the metric name.
      Parameters:
      axis - The dimension along which the cosine similarity is computed.
      seed - the seed for random number generation. An initializer created with a given seed will always produce the same random tensor for a given shape and data type.
      type - the type for the variables and result

    • CosineSimilarity

      public CosineSimilarity(int[] axis, long seed, Class<T> type)
      Creates a CosineSimilarity metric using Class.getSimpleName() for the metric name.
      Parameters:
      axis - The dimension along which the cosine similarity is computed.
      seed - the seed for random number generation. An initializer created with a given seed will always produce the same random tensor for a given shape and data type.
      type - the type for the variables and result

    • CosineSimilarity

      public CosineSimilarity(String name, long seed, Class<T> type)
      Creates a metric that computes the cosine similarity metric between labels and predictions with a default axis, DEFAULT_AXIS
      Parameters:
      name - the name of this metric, if null then metric name is Class.getSimpleName().
      seed - the seed for random number generation. An initializer created with a given seed will always produce the same random tensor for a given shape and data type.
      type - the type for the variables and result

    • CosineSimilarity

      public CosineSimilarity(String name, int axis, long seed, Class<T> type)
      Creates a metric that computes the cosine similarity metric between labels and predictions.
      Parameters:
      name - the name of this metric, if null then metric name is Class.getSimpleName().
      axis - The dimension along which the cosine similarity is computed.
      seed - the seed for random number generation. An initializer created with a given seed will always produce the same random tensor for a given shape and data type.
      type - the type for the variables and result

    • CosineSimilarity

      public CosineSimilarity(String name, int[] axis, long seed, Class<T> type)
      Creates a CosineSimilarity metric
      Parameters:
      name - the name of this metric, if null then metric name is Class.getSimpleName().
      axis - The dimension along which the cosine similarity is computed.
      seed - the seed for random number generation. An initializer created with a given seed will always produce the same random tensor for a given shape and data type.
      type - the type for the variables and result

  • Method Details

    • call

      public <U extends TNumber> Operand<U> call(Ops tf, Operand<? extends TNumber> labels, Operand<? extends TNumber> predictions, Class<U> resultType)
      Computes the cosine similarity loss between labels and predictions.
      Parameters:
      tf - the TensorFlow Ops encapsulating a Graph environment.
      labels - the truth values or labels
      predictions - the predictions
      Returns:
      the cosine similarity loss
      Throws:
      IllegalArgumentException - if the TensorFlow Ops scope does not encapsulate a Graph environment.

    • getLoss

      public org.tensorflow.framework.metrics.impl.LossMetric getLoss()
      Gets the loss function.
      Returns:
      the loss function.

    • setLoss

      protected void setLoss(org.tensorflow.framework.metrics.impl.LossMetric loss)
      Sets the AbstractLoss function for this wrapper.
      Parameters:
      loss - the loss function.

    • updateStateList

      public List<Op> updateStateList(Ops tf, Operand<? extends TNumber> labels, Operand<? extends TNumber> predictions, Operand<? extends TNumber> sampleWeights)
      Creates Operations that update the state of the mean metric, by calling the loss function and passing the loss to the Mean metric to calculate the weighted mean of the loss over many iterations.
      Specified by:
      updateStateList in interface Metric
      Overrides:
      updateStateList in class BaseMetric
      Parameters:
      tf - the TensorFlow Ops encapsulating a Graph environment.
      labels - the truth values or labels
      predictions - the predictions
      sampleWeights - Optional sampleWeights acts as a coefficient for the loss. If a scalar is provided, then the loss is simply scaled by the given value. If sampleWeights 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 sampleWeights vector. If the shape of sampleWeights is [batch_size, d0, .. dN-1] (or can be broadcasted to this shape), then each loss element of predictions is scaled by the corresponding value of sampleWeights. (Note on dN-1: all loss functions reduce by 1 dimension, usually axis=-1.)
      Returns:
      a List of control operations that updates the Mean state variables.
      Throws:
      IllegalArgumentException - if the TensorFlow Ops scope does not encapsulate a Graph environment.

    • init

      protected void init(Ops tf)
      Initialize the TensorFlow Ops
      Specified by:
      init in class BaseMetric
      Parameters:
      tf - the TensorFlow Ops encapsulating a Graph environment.

    • resetStates

      public Op resetStates(Ops tf)
      Resets any state variables to their initial values
      Parameters:
      tf - the TensorFlow Ops encapsulating a Graph environment.
      Returns:
      the operation for doing the reset

    • updateStateList

      public List<Op> updateStateList(Ops tf, Operand<? extends TNumber> values, Operand<? extends TNumber> sampleWeights)
      Updates the metric variables based on the inputs. At least one input arg required for values, an optional additional input for the sampleWeights
      Specified by:
      updateStateList in interface Metric
      Overrides:
      updateStateList in class BaseMetric
      Parameters:
      tf - the TensorFlow Ops encapsulating a Graph environment.
      values - the inputs to be passed to update state, this may not be null
      sampleWeights - sample weights to be applied to values, will default to 1 if null.
      Returns:
      the result with a control dependency on update state Operands
      Throws:
      IllegalArgumentException - if values is null

    • result

      public <U extends TNumber> Operand<U> result(Ops tf, Class<U> type)
      Gets the current result of the metric
      Type Parameters:
      U - the date type for the result
      Parameters:
      tf - the TensorFlow Ops encapsulating a Graph environment.
      type - the data type for the result
      Returns:
      the result, possibly with control dependencies

    • getTotal

      public Variable<T> getTotal()
      Gets the total variable
      Returns:
      the total variable

    • getCount

      public Variable<T> getCount()
      Gets the count variable
      Returns:
      the count variable

    • getInternalType

      public Class<T> getInternalType()
      Gets the type for the variables
      Returns:
      the type for the variables