lPredictions;
if (!fromLogits) {
// add predictions range check for 0 - 1
diff --git a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/CategoricalHinge.java b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/CategoricalHinge.java
index f592c19f8bb..73837ed1756 100644
--- a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/CategoricalHinge.java
+++ b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/CategoricalHinge.java
@@ -25,7 +25,7 @@
* loss = maximum(neg - pos + 1, 0) where neg=maximum((1-labels)*predictions)
* and pos=sum(labels*predictions)
*
- *
labels values are expected to be 0 or 1.
+ * labels values are expected to be 0 or 1.
*
*
Standalone usage:
*
@@ -99,8 +99,8 @@ public CategoricalHinge(Ops tf, String name, Reduction reduction) {
/** {@inheritDoc} */
@Override
- public Operand call(
- Operand labels, Operand predictions, Operand sampleWeights) {
+ public Operand call(
+ Operand labels, Operand predictions, Operand sampleWeights) {
Operand losses = Losses.categoricalHinge(getTF(), labels, predictions);
return LossesHelper.computeWeightedLoss(getTF(), losses, getReduction(), sampleWeights);
}
diff --git a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/CosineSimilarity.java b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/CosineSimilarity.java
index 137c7025c04..0a18d93caf3 100644
--- a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/CosineSimilarity.java
+++ b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/CosineSimilarity.java
@@ -22,12 +22,13 @@
/**
* Computes the cosine similarity 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 -1indicate 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 or predictions is
- * a zero vector, cosine similarity will be 0 regardless of the proximity between predictions and
- * targets.
+ *
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 -1indicate 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 or
+ * predictions is a zero vector, cosine similarity will be 0 regardless of
+ * the proximity between predictions and targets.
*
*
loss = -sum(l2Norm(labels) * l2Norm(predictions))
*
@@ -71,7 +72,7 @@ public class CosineSimilarity extends Loss {
public static final int DEFAULT_AXIS = -1;
public static final Reduction DEFAULT_REDUCTION = Reduction.AUTO;
- private final int axis;
+ private final int[] axis;
/**
* Creates a Cosine Similarity Loss using {@link Class#getSimpleName()} as the loss name, an axis
@@ -107,6 +108,17 @@ public CosineSimilarity(Ops tf, int axis) {
this(tf, null, axis, DEFAULT_REDUCTION);
}
+ /**
+ * Creates a Cosine Similarity Loss using {@link Class#getSimpleName()} as the loss name, and a
+ * Loss Reduction of {@link #DEFAULT_REDUCTION}
+ *
+ * @param tf the TensorFlow Ops
+ * @param axis The dimension along which the cosine similarity is computed.
+ */
+ public CosineSimilarity(Ops tf, int[] axis) {
+
+ this(tf, null, axis, DEFAULT_REDUCTION);
+ }
/**
* Creates a Cosine Similarity Loss using a Loss Reduction of {@link #DEFAULT_REDUCTION}
@@ -120,6 +132,18 @@ public CosineSimilarity(Ops tf, String name, int axis) {
this(tf, name, axis, DEFAULT_REDUCTION);
}
+ /**
+ * Creates a Cosine Similarity Loss using a Loss Reduction of {@link #DEFAULT_REDUCTION}
+ *
+ * @param tf the TensorFlow Ops
+ * @param name the name of the loss
+ * @param axis The dimension along which the cosine similarity is computed.
+ */
+ public CosineSimilarity(Ops tf, String name, int[] axis) {
+
+ this(tf, name, axis, DEFAULT_REDUCTION);
+ }
+
/**
* Creates a Cosine Similarity Loss using {@link Class#getSimpleName()} as the loss name and an
* axis of {@link #DEFAULT_AXIS}
@@ -153,6 +177,18 @@ public CosineSimilarity(Ops tf, String name, Reduction reduction) {
*/
public CosineSimilarity(Ops tf, int axis, Reduction reduction) {
+ this(tf, null, new int[] {axis}, reduction);
+ }
+
+ /**
+ * Creates a Cosine Similarity Loss using {@link Class#getSimpleName()} as the loss name
+ *
+ * @param tf the TensorFlow Ops
+ * @param axis The dimension along which the cosine similarity is computed.
+ * @param reduction Type of Reduction to apply to the loss.
+ */
+ public CosineSimilarity(Ops tf, int[] axis, Reduction reduction) {
+
this(tf, null, axis, reduction);
}
@@ -165,15 +201,28 @@ public CosineSimilarity(Ops tf, int axis, Reduction reduction) {
* @param reduction Type of Reduction to apply to the loss.
*/
public CosineSimilarity(Ops tf, String name, int axis, Reduction reduction) {
+ this(tf, name, new int[] {axis}, reduction);
+ }
+
+ /**
+ * Creates a Cosine Similarity Loss
+ *
+ * @param tf the TensorFlow Ops
+ * @param name the name of the loss
+ * @param axis The dimension along which the cosine similarity is computed.
+ * @param reduction Type of Reduction to apply to the loss.
+ */
+ public CosineSimilarity(Ops tf, String name, int[] axis, Reduction reduction) {
super(tf, name, reduction);
this.axis = axis;
}
/** {@inheritDoc} */
@Override
- public Operand call(
- Operand labels, Operand predictions, Operand sampleWeights) {
+ public Operand call(
+ Operand labels, Operand predictions, Operand sampleWeights) {
Operand losses = Losses.cosineSimilarity(getTF(), labels, predictions, axis);
+ losses = tf.math.neg(losses);
return LossesHelper.computeWeightedLoss(getTF(), losses, getReduction(), sampleWeights);
}
}
diff --git a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Hinge.java b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Hinge.java
index 88b4a7aa056..d4c350ef06c 100644
--- a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Hinge.java
+++ b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Hinge.java
@@ -18,15 +18,16 @@
import org.tensorflow.framework.losses.impl.LossesHelper;
import org.tensorflow.op.Ops;
import org.tensorflow.types.family.TNumber;
+
import static org.tensorflow.framework.utils.CastHelper.cast;
/**
* Computes the hinge loss between labels and predictions.
*
- * loss = maximum(1 - labels * predictions, 0)
.
+ * loss = maximum(1 - labels * predictions, 0).
*
- *
labels values are expected to be -1 or 1.
- * If binary (0 or 1) labels are provided, they will be converted to -1 or 1.
+ * labels values are expected to be -1 or 1. If binary (0 or 1) labels are provided,
+ * they will be converted to -1 or 1.
*
*
Standalone usage:
*
@@ -106,7 +107,7 @@ public Hinge(Ops tf, String name, Reduction reduction) {
* label values are not in the set [-1., 0., 1.].
*
* @param labels the truth values or labels, must be either -1, 0, or 1. Values are expected to be
- * -1 or 1. If binary (0 or 1) labels are provided they will be converted to -1 or 1.
+ * -1 or 1. If binary (0 or 1) labels are provided they will be converted to -1 or 1.
* @param predictions the predictions, values must be in the range [0. to 1.] inclusive.
* @param 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
@@ -116,21 +117,19 @@ public Hinge(Ops tf, String name, Reduction reduction) {
* predictions is scaled by the corresponding value of SampleWeights. (Note on dN-1: all loss
* functions reduce by 1 dimension, usually axis=-1.)
* @param The data type of the predictions, sampleWeights and loss.
- * @param The data type of the labels.
* @return the loss
* @throws IllegalArgumentException if the predictions are outside the range [0.-1.].
*/
@Override
- public Operand call(
- Operand labels, Operand predictions, Operand sampleWeights) {
- @SuppressWarnings("unchecked")
- Operand tLabels = predictions.type() == labels.type() ?
- (Operand)labels : cast(tf, labels, predictions.type());
- tLabels = LossesHelper.valueCheck(
+ public Operand call(
+ Operand labels, Operand predictions, Operand sampleWeights) {
+ Operand tLabels = cast(tf, labels, predictions.type());
+ tLabels =
+ LossesHelper.valueCheck(
getTF(),
"labels value check [-1, 0, 1]",
tLabels,
- cast(getTF(), getTF().constant(new int[] { -1, 0, 1}), predictions.type()));
+ cast(getTF(), getTF().constant(new int[] {-1, 0, 1}), predictions.type()));
Operand losses = Losses.hinge(getTF(), tLabels, predictions);
return LossesHelper.computeWeightedLoss(getTF(), losses, getReduction(), sampleWeights);
}
diff --git a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Huber.java b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Huber.java
index 6d3e3f0c2ac..b1aee1b0656 100644
--- a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Huber.java
+++ b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Huber.java
@@ -89,6 +89,7 @@ public Huber(Ops tf) {
* Loss#REDUCTION_DEFAULT}
*
* @param tf the TensorFlow Ops
+ * @param name the name of the loss, if null then {@link Class#getSimpleName()} is used.
*/
public Huber(Ops tf, String name) {
this(tf, name, DELTA_DEFAULT, Reduction.AUTO);
@@ -109,6 +110,7 @@ public Huber(Ops tf, Reduction reduction) {
* Creates a Huber Loss using {@link #DELTA_DEFAULT} as the delta
*
* @param tf the TensorFlow Ops
+ * @param name the name of the loss, if null then {@link Class#getSimpleName()} is used.
* @param reduction Type of Reduction to apply to the loss.
*/
public Huber(Ops tf, String name, Reduction reduction) {
@@ -119,7 +121,7 @@ public Huber(Ops tf, String name, Reduction reduction) {
* Creates a Huber Loss
*
* @param tf the TensorFlow Ops
- * @param name the name of the loss
+ * @param name the name of the loss, if null then {@link Class#getSimpleName()} is used.
* @param delta the point where the Huber loss function changes from quadratic to linear.
* @param reduction Type of Reduction to apply to the loss.
*/
@@ -130,8 +132,8 @@ public Huber(Ops tf, String name, float delta, Reduction reduction) {
/** {@inheritDoc} */
@Override
- public Operand call(
- Operand labels, Operand predictions, Operand sampleWeights) {
+ public Operand call(
+ Operand labels, Operand predictions, Operand sampleWeights) {
Operand losses = Losses.huber(getTF(), labels, predictions, delta);
return LossesHelper.computeWeightedLoss(getTF(), losses, getReduction(), sampleWeights);
}
diff --git a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/KLDivergence.java b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/KLDivergence.java
index 8cf3db8d518..2aa1f72092b 100644
--- a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/KLDivergence.java
+++ b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/KLDivergence.java
@@ -99,8 +99,8 @@ public KLDivergence(Ops tf, String name, Reduction reduction) {
/** {@inheritDoc} */
@Override
- public Operand call(
- Operand labels, Operand predictions, Operand sampleWeights) {
+ public Operand call(
+ Operand labels, Operand predictions, Operand sampleWeights) {
Operand losses = Losses.kullbackLeiblerDivergence(getTF(), labels, predictions);
return LossesHelper.computeWeightedLoss(getTF(), losses, getReduction(), sampleWeights);
}
diff --git a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/LogCosh.java b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/LogCosh.java
index 1669669a768..a11d582e527 100644
--- a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/LogCosh.java
+++ b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/LogCosh.java
@@ -77,6 +77,7 @@ public LogCosh(Ops tf) {
* Creates a LogCosh Loss using a Loss Reduction of {@link Loss#REDUCTION_DEFAULT}
*
* @param tf the TensorFlow Ops
+ * @param name the name of the loss, if null then {@link Class#getSimpleName()} is used.
*/
public LogCosh(Ops tf, String name) {
this(tf, name, Reduction.AUTO);
@@ -96,7 +97,7 @@ public LogCosh(Ops tf, Reduction reduction) {
* Creates a LogCosh Loss
*
* @param tf the TensorFlow Ops
- * @param name the name of the loss
+ * @param name the name of the loss, if null then {@link Class#getSimpleName()} is used.
* @param reduction Type of Reduction to apply to the loss.
*/
public LogCosh(Ops tf, String name, Reduction reduction) {
@@ -105,8 +106,8 @@ public LogCosh(Ops tf, String name, Reduction reduction) {
/** {@inheritDoc} */
@Override
- public Operand call(
- Operand labels, Operand predictions, Operand sampleWeights) {
+ public Operand call(
+ Operand labels, Operand predictions, Operand sampleWeights) {
Operand losses = Losses.logCosh(getTF(), labels, predictions);
return LossesHelper.computeWeightedLoss(getTF(), losses, getReduction(), sampleWeights);
}
diff --git a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Loss.java b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Loss.java
index ae33d5dfa37..cdd35d28aba 100644
--- a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Loss.java
+++ b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Loss.java
@@ -25,7 +25,7 @@ public abstract class Loss {
protected final Reduction reduction;
/**
- * Creates a Loss using {@link Class#getSimpleName()} as the name and a Loss Reduction of {@link
+ * Creates a Loss using {@link Class#getSimpleName()} as the name and a Loss Reduction of {@link
* Loss#REDUCTION_DEFAULT}
*
* @param tf the TensorFlow Ops
@@ -62,10 +62,10 @@ protected Loss(Ops tf, String name, Reduction reduction) {
* @param labels the truth values or labels
* @param predictions the predictions
* @param The data type of the predictions and loss.
- * @param The data type of the labels.
* @return the loss
*/
- public Operand call(Operand labels, Operand predictions) {
+ public Operand call(
+ Operand labels, Operand predictions) {
return call(labels, predictions, null);
}
@@ -82,11 +82,10 @@ public Operand call(Operand labels,
* predictions is scaled by the corresponding value of SampleWeights. (Note on dN-1: all loss
* functions reduce by 1 dimension, usually axis=-1.)
* @param The data type of the predictions, sampleWeights and loss.
- * @param The data type of the labels.
* @return the loss
*/
- public abstract Operand call(
- Operand labels, Operand predictions, Operand sampleWeights);
+ public abstract Operand call(
+ Operand labels, Operand predictions, Operand sampleWeights);
/**
* Gets the TensorFlow Ops
diff --git a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Losses.java b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Losses.java
index 0d25bd5e7e2..9aa94cf7fcf 100644
--- a/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Losses.java
+++ b/tensorflow-framework/src/main/java/org/tensorflow/framework/losses/Losses.java
@@ -48,11 +48,10 @@ public class Losses {
* @param labels the labels
* @param predictions the predictions
* @param the data type of the predictions and result
- * @param the data type of the labels
* @return the mean absolute error
*/
- public static Operand meanAbsoluteError(
- Ops tf, Operand labels, Operand predictions) {
+ public static Operand meanAbsoluteError(
+ Ops tf, Operand labels, Operand predictions) {
Operand tLabels = cast(tf, labels, predictions.type());
LossTuple ops = LossesHelper.squeezeOrExpandDimensions(tf, tLabels, predictions, null);
predictions = ops.getTarget();
@@ -70,11 +69,10 @@ public static Operand meanAbsoluteErro
* @param labels the labels
* @param predictions the predictions
* @param the data type of the predictions and result
- * @param the data type of the labels
* @return the mean squared error
*/
- public static Operand meanSquaredError(
- Ops tf, Operand labels, Operand predictions) {
+ public static Operand meanSquaredError(
+ Ops tf, Operand labels, Operand predictions) {
Operand tLabels = cast(tf, labels, predictions.type());
LossTuple ops = LossesHelper.squeezeOrExpandDimensions(tf, tLabels, predictions, null);
predictions = ops.getTarget();
@@ -91,11 +89,10 @@ public static Operand meanSquaredError
* @param labels the labels
* @param predictions the predictions
* @param the data type of the predictions and result
- * @param the data type of the labels
* @return the mean absolute percentage error
*/
- public static Operand meanAbsolutePercentageError(
- Ops tf, Operand labels, Operand predictions) {
+ public static Operand meanAbsolutePercentageError(
+ Ops tf, Operand labels, Operand predictions) {
Class predictionType = predictions.type();
Operand tLabels = cast(tf, labels, predictionType);
LossTuple ops = LossesHelper.squeezeOrExpandDimensions(tf, tLabels, predictions, null);
@@ -105,8 +102,10 @@ public static Operand meanAbsolutePerc
tf.math.abs(
tf.math.div(
tf.math.sub(tLabels, predictions),
- tf.math.maximum(tf.math.abs(tLabels), cast(tf, tf.constant(EPSILON), predictionType))));
- return tf.math.mul(cast(tf, tf.constant(100), predictionType), tf.math.mean(diff, tf.constant(-1)));
+ tf.math.maximum(
+ tf.math.abs(tLabels), cast(tf, tf.constant(EPSILON), predictionType))));
+ return tf.math.mul(
+ cast(tf, tf.constant(100), predictionType), tf.math.mean(diff, tf.constant(-1)));
}
/**
@@ -118,11 +117,10 @@ public static Operand meanAbsolutePerc
* @param labels the labels
* @param predictions the predictions
* @param the data type of the predictions and result
- * @param the data type of the labels
* @return the mean squared logarithmic percentage error
*/
- public static Operand meanSquaredLogarithmicError(
- Ops tf, Operand labels, Operand predictions) {
+ public static Operand meanSquaredLogarithmicError(
+ Ops tf, Operand labels, Operand predictions) {
Class predictionType = predictions.type();
Operand tLabels = cast(tf, labels, predictionType);
LossTuple ops = LossesHelper.squeezeOrExpandDimensions(tf, tLabels, predictions, null);
@@ -152,8 +150,12 @@ public static Operand meanSquaredLogar
* @param the data type of the predictions and labels
* @return the binary crossentropy loss.
*/
- public static Operand binaryCrossentropy(
- Ops tf, Operand labels, Operand predictions, boolean fromLogits, float labelSmoothing) {
+ public static Operand binaryCrossentropy(
+ Ops tf,
+ Operand labels,
+ Operand predictions,
+ boolean fromLogits,
+ float labelSmoothing) {
Operand tLabels = cast(tf, labels, predictions.type());
LossTuple ops = LossesHelper.squeezeOrExpandDimensions(tf, tLabels, predictions, null);
predictions = ops.getTarget();
@@ -181,7 +183,7 @@ private static Operand binaryCrossentropyHelper(
Ops tf, Operand target, Operand output, boolean fromLogits) {
if (fromLogits) return tf.nn.sigmoidCrossEntropyWithLogits(target, output);
- /* TODO - skip this loggic for now. It requires walking back the inputs which is not yet possible
+ /* TODO - skip this logic for now. It requires walking back the inputs which is not yet possible
if (!(output instanceof Variable) && (!tf.scope().env().isEager())) {
// TODO - this does not work
// TODO output = backtrackIdentity(output);
@@ -218,16 +220,17 @@ private static Operand binaryCrossentropyHelper(
* @param labels true targets
* @param predictions the predictions
* @param fromLogits Whether to interpret predictions as a tensor of logit values
- * @param labelSmoothing Float in [0, 1]. When > 0, label values are smoothed, meaning the
- * confidence on label values are relaxed. e.g. labelSmoothing=0.2 means that we will use a
- * value of 0.1 for label 0 and 0.9 for label 1
+ * @param labelSmoothing Float in [0, 1]. When > 0, label values are
+ * smoothed, meaning the confidence on label values are relaxed. e.g. labelSmoothing=0.2
+ * means that we will use a value of 0.1 for label 0 and
+ * 0.9 for label 1
* @param axis the
* @param the data type of the predictions and labels
* @return the categorical crossentropy loss.
*/
- public static Operand categoricalCrossentropy(
+ public static Operand categoricalCrossentropy(
Ops tf,
- Operand labels,
+ Operand labels,
Operand predictions,
boolean fromLogits,
float labelSmoothing,
@@ -283,8 +286,8 @@ public static Operand categoricalCross
* @param the data type of the predictions and labels
* @return the categorical hinge loss
*/
- public static Operand categoricalHinge(
- Ops tf, Operand labels, Operand predictions) {
+ public static Operand categoricalHinge(
+ Ops tf, Operand labels, Operand predictions) {
Class predictionType = predictions.type();
Operand tLabels = cast(tf, labels, predictionType);
LossTuple lossTuple = LossesHelper.squeezeOrExpandDimensions(tf, tLabels, predictions, null);
@@ -329,8 +332,8 @@ public static Operand categoricalHinge
* @param the data type of the predictions and labels
* @return the cosine similarity loss
*/
- public static Operand cosineSimilarity(
- Ops tf, Operand labels, Operand predictions, int axis) {
+ public static Operand cosineSimilarity(
+ Ops tf, Operand labels, Operand predictions, int[] axis) {
Operand tLabels = cast(tf, labels, predictions.type());
LossTuple lossTuple = LossesHelper.squeezeOrExpandDimensions(tf, tLabels, predictions, null);
predictions = lossTuple.getTarget();
@@ -339,8 +342,7 @@ public static Operand cosineSimilarity
tLabels = l2Normalize(tf, tLabels, axis);
predictions = l2Normalize(tf, predictions, axis);
Operand mathMul = tf.math.mul(tLabels, predictions);
- Operand sum = tf.reduceSum(mathMul, tf.constant(axis), ReduceSum.keepDims(Boolean.FALSE));
- return tf.math.neg(sum);
+ return tf.reduceSum(mathMul, tf.constant(axis), ReduceSum.keepDims(Boolean.FALSE));
}
/**
@@ -355,8 +357,8 @@ public static Operand cosineSimilarity
* @param the data type of the predictions and labels
* @return the hinge loss
*/
- public static Operand hinge(
- Ops tf, Operand labels, Operand predictions) {
+ public static Operand hinge(
+ Ops tf, Operand labels, Operand predictions) {
Class predictionType = predictions.type();
Operand tLabels = cast(tf, labels, predictionType);
LossTuple lossTuple = LossesHelper.squeezeOrExpandDimensions(tf, tLabels, predictions, null);
@@ -391,8 +393,8 @@ public static Operand hinge(
* @param the data type of the predictions and labels
* @return the Huber loss
*/
- public static Operand huber(
- Ops tf, Operand labels, Operand predictions, float delta) {
+ public static Operand huber(
+ Ops tf, Operand labels, Operand predictions, float delta) {
Class predictionType = predictions.type();
Operand tLabels = cast(tf, labels, predictionType);
LossTuple lossTuple = LossesHelper.squeezeOrExpandDimensions(tf, tLabels, predictions, null);
@@ -422,8 +424,8 @@ public static Operand huber(
* @see Kullback?Leibler
* divergence
*/
- public static Operand