Metrics

public class Metrics

Helper class with built-in metrics functions.

Constants

float L2_NORM_EPSILON

Public Constructors

Metrics()

Public Methods

static <T extends TNumber> Operand<T>
topKCategoricalAccuracy(Ops tf, Operand<? extends TNumber> labels, Operand<T> predictions, long k)
Computes how often targets are in the top K predictions.

Inherited Methods

Constants

public static final float L2_NORM_EPSILON

Constant Value: 1.0E-12

Public Constructors

public Metrics ()

Public Methods

public static Operand<T> topKCategoricalAccuracy (Ops tf, Operand<? extends TNumber> labels, Operand<T> predictions, long k)

Computes how often targets are in the top K predictions.

Standalone usage: 

     Operand<TInt32> labels = tf.constant(new int[][]
                                    { {0, 0, 1}, {0, 1, 0} });
     Operand<TFloat32> predictions = tf.constant(new float[][]
                                    { {0.1f, 0.9f, 0.8f}, {0.05f, 0.95f, 0f} });
     Operand<TFloat32> m = Metrics.topKCategoricalAccuracy(
                                    labels, predictions, 3)
     //m.shape().toString == "[2]"

Parameters
tf the TensorFlow Ops.
labels the ground truth values.
predictions The prediction values.
k Number of top elements to look at for computing accuracy.
Returns
  • the Operand for the Top K categorical accuracy value.