Java Code Examples for org.nd4j.autodiff.samediff.SDVariable#markAsLoss()
The following examples show how to use
org.nd4j.autodiff.samediff.SDVariable#markAsLoss() .
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example. You may check out the related API usage on the sidebar.
Example 1
Source File: TransformOpValidation.java From deeplearning4j with Apache License 2.0 | 6 votes |
@Test public void testMergeAddBp() { Nd4j.getRandom().setSeed(12345); SameDiff sd = SameDiff.create(); SDVariable inputX = sd.var(Nd4j.rand(2, 3)); SDVariable inputY = sd.var(Nd4j.rand(2, 3)); SDVariable inputZ = sd.var(Nd4j.rand(2, 3)); SDVariable out = new MergeAddOp(sd, new SDVariable[]{inputX, inputY, inputZ}).outputVariable().std(true); out.markAsLoss(); String err = OpValidation.validate(new TestCase(sd) .gradientCheck(true)); assertNull(err); }
Example 2
Source File: ShapeOpValidation.java From deeplearning4j with Apache License 2.0 | 5 votes |
@Test public void testTriuOp() { SameDiff sd = SameDiff.create(); SDVariable input = sd.var(Nd4j.createFromArray(new double[][]{{1,2,3}, {4,5,6}, {7,8,9},{10,11,12}})); SDVariable out = new Triu(sd, input,-1).outputVariable(); out.markAsLoss(); INDArray expected = Nd4j.createFromArray(new double[][]{{1,2,3}, {4,5,6}, {0,8,9},{0,0,12}}); String err = OpValidation.validate(new TestCase(sd) .expectedOutput("triu", expected) .gradientCheck(true)); assertNull(err); }
Example 3
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 5 votes |
/** * Weighted cross entropy loss with logits<br> * * @param targets targets array (NUMERIC type) * @param inputs input array (NUMERIC type) * @param weights eights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @return output Loss variable (NUMERIC type) */ public SDVariable weightedCrossEntropyWithLogits(SDVariable targets, SDVariable inputs, SDVariable weights) { SDValidation.validateNumerical("weightedCrossEntropyWithLogits", "targets", targets); SDValidation.validateNumerical("weightedCrossEntropyWithLogits", "inputs", inputs); SDValidation.validateNumerical("weightedCrossEntropyWithLogits", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.WeightedCrossEntropyLoss(sd,targets, inputs, weights).outputVariable(); out.markAsLoss(); return out; }
Example 4
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 5 votes |
/** * Absolute difference loss: {@code sum_i abs( label[i] - predictions[i] )}<br> * * @param name name May be null. Name for the output variable * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @return output loss variable (NUMERIC type) */ public SDVariable absoluteDifference(String name, SDVariable label, SDVariable predictions, SDVariable weights) { SDValidation.validateNumerical("absoluteDifference", "label", label); SDValidation.validateNumerical("absoluteDifference", "predictions", predictions); SDValidation.validateNumerical("absoluteDifference", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.AbsoluteDifferenceLoss(sd,label, predictions, weights, org.nd4j.autodiff.loss.LossReduce.MEAN_BY_NONZERO_WEIGHT_COUNT).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 5
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Log loss, i.e., binary cross entropy loss, usually used for binary multi-label classification. Implements:<br> * {@code -1/numExamples * sum_i (labels[i] * log(predictions[i] + epsilon) + (1-labels[i]) * log(1-predictions[i] + epsilon))}<br> * * @param name name May be null. Name for the output variable * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @param lossReduce Reduction type for the loss. See {@link LossReduce} for more details. Default: {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} * @param epsilon epsilon * @return output Log loss (NUMERIC type) */ public SDVariable logLoss(String name, SDVariable label, SDVariable predictions, SDVariable weights, LossReduce lossReduce, double epsilon) { SDValidation.validateNumerical("logLoss", "label", label); SDValidation.validateNumerical("logLoss", "predictions", predictions); SDValidation.validateNumerical("logLoss", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.LogLoss(sd,label, predictions, weights, lossReduce, epsilon).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 6
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Mean squared error loss function. Implements {@code (label[i] - prediction[i])^2} - i.e., squared error on a per-element basis.<br> * When averaged (using {@link LossReduce#MEAN_BY_WEIGHT} or {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} (the default))<br> * this is the mean squared error loss function.<br> * * @param name name May be null. Name for the output variable * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @param lossReduce Reduction type for the loss. See {@link LossReduce} for more details. Default: {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} * @return output Loss variable (NUMERIC type) */ public SDVariable meanSquaredError(String name, SDVariable label, SDVariable predictions, SDVariable weights, LossReduce lossReduce) { SDValidation.validateNumerical("meanSquaredError", "label", label); SDValidation.validateNumerical("meanSquaredError", "predictions", predictions); SDValidation.validateNumerical("meanSquaredError", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.MeanSquaredErrorLoss(sd,label, predictions, weights, lossReduce).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 7
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Huber loss function, used for robust regression. It is similar both squared error loss and absolute difference loss,<br> * though is less sensitive to outliers than squared error.<br> * Huber loss implements:<br> * <pre><br> * {@code L = 0.5 * (label[i] - predictions[i])^2 if abs(label[i] - predictions[i]) < delta}<br> * {@code L = delta * abs(label[i] - predictions[i]) - 0.5 * delta^2 otherwise}<br> * </pre><br> * * @param name name May be null. Name for the output variable * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @param lossReduce Reduction type for the loss. See {@link LossReduce} for more details. Default: {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} * @param delta Loss function delta value * @return output Huber loss (NUMERIC type) */ public SDVariable huberLoss(String name, SDVariable label, SDVariable predictions, SDVariable weights, LossReduce lossReduce, double delta) { SDValidation.validateNumerical("huberLoss", "label", label); SDValidation.validateNumerical("huberLoss", "predictions", predictions); SDValidation.validateNumerical("huberLoss", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.HuberLoss(sd,label, predictions, weights, lossReduce, delta).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 8
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Mean pairwise squared error.<br> * MPWSE loss calculates the difference between pairs of consecutive elements in the predictions and labels arrays.<br> * For example, if predictions = [p0, p1, p2] and labels are [l0, l1, l2] then MPWSE is:<br> * {@code [((p0-p1) - (l0-l1))^2 + ((p0-p2) - (l0-l2))^2 + ((p1-p2) - (l1-l2))^2] / 3}<br> * * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used. Must be either null, scalar, or have shape [batchSize] (NUMERIC type) * @param lossReduce Reduction type for the loss. See {@link LossReduce} for more details. Default: {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} * @return output Loss variable, scalar output (NUMERIC type) */ public SDVariable meanPairwiseSquaredError(SDVariable label, SDVariable predictions, SDVariable weights, LossReduce lossReduce) { SDValidation.validateNumerical("meanPairwiseSquaredError", "label", label); SDValidation.validateNumerical("meanPairwiseSquaredError", "predictions", predictions); SDValidation.validateNumerical("meanPairwiseSquaredError", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.MeanPairwiseSquaredErrorLoss(sd,label, predictions, weights, lossReduce).outputVariable(); out.markAsLoss(); return out; }
Example 9
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Hinge loss: a loss function used for training classifiers.<br> * Implements {@code L = max(0, 1 - t * predictions)} where t is the label values after internally converting to {-1,1}<br> * from the user specified {0,1}. Note that Labels should be provided with values {0,1}.<br> * * @param label Label array. Each value should be 0.0 or 1.0 (internally -1 to 1 is used) (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @param lossReduce Reduction type for the loss. See {@link LossReduce} for more details. Default: {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} * @return output Loss variable (NUMERIC type) */ public SDVariable hingeLoss(SDVariable label, SDVariable predictions, SDVariable weights, LossReduce lossReduce) { SDValidation.validateNumerical("hingeLoss", "label", label); SDValidation.validateNumerical("hingeLoss", "predictions", predictions); SDValidation.validateNumerical("hingeLoss", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.HingeLoss(sd,label, predictions, weights, lossReduce).outputVariable(); out.markAsLoss(); return out; }
Example 10
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Cosine distance loss: {@code 1 - cosineSimilarity(x,y)} or {@code 1 - sum_i label[i] * prediction[i]}, which is<br> * equivalent to cosine distance when both the predictions and labels are normalized.<br> * <b>Note</b>: This loss function assumes that both the predictions and labels are normalized to have unit l2 norm.<br> * If this is not the case, you should normalize them first by dividing by norm2(String, SDVariable, boolean, int...)<br> * along the cosine distance dimension (with keepDims=true).<br> * * @param name name May be null. Name for the output variable * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is use (NUMERIC type) * @param dimension Dimension to perform the cosine distance over * @return output Cosine distance loss (NUMERIC type) */ public SDVariable cosineDistance(String name, SDVariable label, SDVariable predictions, SDVariable weights, int dimension) { SDValidation.validateNumerical("cosineDistance", "label", label); SDValidation.validateNumerical("cosineDistance", "predictions", predictions); SDValidation.validateNumerical("cosineDistance", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.CosineDistanceLoss(sd,label, predictions, weights, org.nd4j.autodiff.loss.LossReduce.MEAN_BY_NONZERO_WEIGHT_COUNT, dimension).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 11
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * As per softmaxCrossEntropy(String, SDVariable, SDVariable, LossReduce) but the labels variable<br> * is represented as an integer array instead of the equivalent one-hot array.<br> * i.e., if logits are rank N, then labels have rank N-1<br> * * @param name name May be null. Name for the output variable * @param logits Logits array ("pre-softmax activations") (NUMERIC type) * @param labels Labels array. Must be an integer type. (INT type) * @return output Softmax cross entropy (NUMERIC type) */ public SDVariable sparseSoftmaxCrossEntropy(String name, SDVariable logits, SDVariable labels) { SDValidation.validateNumerical("sparseSoftmaxCrossEntropy", "logits", logits); SDValidation.validateInteger("sparseSoftmaxCrossEntropy", "labels", labels); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.SparseSoftmaxCrossEntropyLossWithLogits(sd,logits, labels).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 12
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Cosine distance loss: {@code 1 - cosineSimilarity(x,y)} or {@code 1 - sum_i label[i] * prediction[i]}, which is<br> * equivalent to cosine distance when both the predictions and labels are normalized.<br> * <b>Note</b>: This loss function assumes that both the predictions and labels are normalized to have unit l2 norm.<br> * If this is not the case, you should normalize them first by dividing by norm2(String, SDVariable, boolean, int...)<br> * along the cosine distance dimension (with keepDims=true).<br> * * @param name name May be null. Name for the output variable * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is use (NUMERIC type) * @param lossReduce Reduction type for the loss. See {@link LossReduce} for more details. Default: {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} * @param dimension Dimension to perform the cosine distance over * @return output Cosine distance loss (NUMERIC type) */ public SDVariable cosineDistance(String name, SDVariable label, SDVariable predictions, SDVariable weights, LossReduce lossReduce, int dimension) { SDValidation.validateNumerical("cosineDistance", "label", label); SDValidation.validateNumerical("cosineDistance", "predictions", predictions); SDValidation.validateNumerical("cosineDistance", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.CosineDistanceLoss(sd,label, predictions, weights, lossReduce, dimension).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 13
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Huber loss function, used for robust regression. It is similar both squared error loss and absolute difference loss,<br> * though is less sensitive to outliers than squared error.<br> * Huber loss implements:<br> * <pre><br> * {@code L = 0.5 * (label[i] - predictions[i])^2 if abs(label[i] - predictions[i]) < delta}<br> * {@code L = delta * abs(label[i] - predictions[i]) - 0.5 * delta^2 otherwise}<br> * </pre><br> * * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @param delta Loss function delta value * @return output Huber loss (NUMERIC type) */ public SDVariable huberLoss(SDVariable label, SDVariable predictions, SDVariable weights, double delta) { SDValidation.validateNumerical("huberLoss", "label", label); SDValidation.validateNumerical("huberLoss", "predictions", predictions); SDValidation.validateNumerical("huberLoss", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.HuberLoss(sd,label, predictions, weights, org.nd4j.autodiff.loss.LossReduce.MEAN_BY_NONZERO_WEIGHT_COUNT, delta).outputVariable(); out.markAsLoss(); return out; }
Example 14
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Applies the softmax activation function to the input, then implement multi-class cross entropy:<br> * {@code -sum_classes label[i] * log(p[c])} where {@code p = softmax(logits)}<br> * If {@link LossReduce#NONE} is used, returned shape is [numExamples] out for [numExamples, numClasses] predicitons/labels;<br> * otherwise, the output is a scalar.<br> * <p><br> * When label smoothing is > 0, the following label smoothing is used:<br> * <pre><br> * {@code numClasses = labels.size(1);<br> * oneHotLabel = (1.0 - labelSmoothing) * oneHotLabels + labelSmoothing/numClasses}<br> * </pre><br> * * @param oneHotLabels Label array. Should be one-hot per example and same shape as predictions (for example, [mb, nOut]) (NUMERIC type) * @param logitPredictions Predictions array (pre-softmax) (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @param lossReduce Reduction type for the loss. See {@link LossReduce} for more details. Default: {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} * @param labelSmoothing Label smoothing value. Default value: 0 * @return output Loss variable (NUMERIC type) */ public SDVariable softmaxCrossEntropy(SDVariable oneHotLabels, SDVariable logitPredictions, SDVariable weights, LossReduce lossReduce, double labelSmoothing) { SDValidation.validateNumerical("softmaxCrossEntropy", "oneHotLabels", oneHotLabels); SDValidation.validateNumerical("softmaxCrossEntropy", "logitPredictions", logitPredictions); SDValidation.validateNumerical("softmaxCrossEntropy", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.SoftmaxCrossEntropyLoss(sd,oneHotLabels, logitPredictions, weights, lossReduce, labelSmoothing).outputVariable(); out.markAsLoss(); return out; }
Example 15
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Absolute difference loss: {@code sum_i abs( label[i] - predictions[i] )}<br> * * @param name name May be null. Name for the output variable * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @param lossReduce Reduction type for the loss. See {@link LossReduce} for more details. Default: {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} * @return output loss variable (NUMERIC type) */ public SDVariable absoluteDifference(String name, SDVariable label, SDVariable predictions, SDVariable weights, LossReduce lossReduce) { SDValidation.validateNumerical("absoluteDifference", "label", label); SDValidation.validateNumerical("absoluteDifference", "predictions", predictions); SDValidation.validateNumerical("absoluteDifference", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.AbsoluteDifferenceLoss(sd,label, predictions, weights, lossReduce).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 16
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Mean squared error loss function. Implements {@code (label[i] - prediction[i])^2} - i.e., squared error on a per-element basis.<br> * When averaged (using {@link LossReduce#MEAN_BY_WEIGHT} or {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} (the default))<br> * this is the mean squared error loss function.<br> * * @param label Label array (NUMERIC type) * @param predictions Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @return output Loss variable (NUMERIC type) */ public SDVariable meanSquaredError(SDVariable label, SDVariable predictions, SDVariable weights) { SDValidation.validateNumerical("meanSquaredError", "label", label); SDValidation.validateNumerical("meanSquaredError", "predictions", predictions); SDValidation.validateNumerical("meanSquaredError", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.MeanSquaredErrorLoss(sd,label, predictions, weights, org.nd4j.autodiff.loss.LossReduce.MEAN_BY_NONZERO_WEIGHT_COUNT).outputVariable(); out.markAsLoss(); return out; }
Example 17
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Sigmoid cross entropy: applies the sigmoid activation function on the input logits (input "pre-sigmoid preductions")<br> * and implements the binary cross entropy loss function. This implementation is numerically more stable than using<br> * standard (but separate) sigmoid activation function and log loss (binary cross entropy) loss function.<br> * Implements:<br> * {@code -1/numExamples * sum_i (labels[i] * log(sigmoid(logits[i])) + (1-labels[i]) * log(1-sigmoid(logits[i])))}<br> * though this is done in a mathematically equivalent but more numerical stable form.<br> * <br> * When label smoothing is > 0, the following label smoothing is used:<br> * <pre><br> * {@code numClasses = labels.size(1);<br> * label = (1.0 - labelSmoothing) * label + 0.5 * labelSmoothing}<br> * </pre><br> * * @param label Label array (NUMERIC type) * @param predictionLogits Predictions array (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @return output Loss variable (NUMERIC type) */ public SDVariable sigmoidCrossEntropy(SDVariable label, SDVariable predictionLogits, SDVariable weights) { SDValidation.validateNumerical("sigmoidCrossEntropy", "label", label); SDValidation.validateNumerical("sigmoidCrossEntropy", "predictionLogits", predictionLogits); SDValidation.validateNumerical("sigmoidCrossEntropy", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.SigmoidCrossEntropyLoss(sd,label, predictionLogits, weights, org.nd4j.autodiff.loss.LossReduce.MEAN_BY_NONZERO_WEIGHT_COUNT, 0.0).outputVariable(); out.markAsLoss(); return out; }
Example 18
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Log poisson loss: a loss function used for training classifiers.<br> * Implements {@code L = exp(c) - z * c} where c is log(predictions) and z is labels.<br> * * @param name name May be null. Name for the output variable * @param label Label array. Each value should be 0.0 or 1.0 (NUMERIC type) * @param predictions Predictions array (has to be log(x) of actual predictions) (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @param full Boolean flag. true for logPoissonFull, false for logPoisson * @return output Loss variable (NUMERIC type) */ public SDVariable logPoisson(String name, SDVariable label, SDVariable predictions, SDVariable weights, boolean full) { SDValidation.validateNumerical("logPoisson", "label", label); SDValidation.validateNumerical("logPoisson", "predictions", predictions); SDValidation.validateNumerical("logPoisson", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.LogPoissonLoss(sd,label, predictions, weights, org.nd4j.autodiff.loss.LossReduce.MEAN_BY_NONZERO_WEIGHT_COUNT, full).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 19
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Applies the softmax activation function to the input, then implement multi-class cross entropy:<br> * {@code -sum_classes label[i] * log(p[c])} where {@code p = softmax(logits)}<br> * If {@link LossReduce#NONE} is used, returned shape is [numExamples] out for [numExamples, numClasses] predicitons/labels;<br> * otherwise, the output is a scalar.<br> * <p><br> * When label smoothing is > 0, the following label smoothing is used:<br> * <pre><br> * {@code numClasses = labels.size(1);<br> * oneHotLabel = (1.0 - labelSmoothing) * oneHotLabels + labelSmoothing/numClasses}<br> * </pre><br> * * @param name name May be null. Name for the output variable * @param oneHotLabels Label array. Should be one-hot per example and same shape as predictions (for example, [mb, nOut]) (NUMERIC type) * @param logitPredictions Predictions array (pre-softmax) (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @return output Loss variable (NUMERIC type) */ public SDVariable softmaxCrossEntropy(String name, SDVariable oneHotLabels, SDVariable logitPredictions, SDVariable weights) { SDValidation.validateNumerical("softmaxCrossEntropy", "oneHotLabels", oneHotLabels); SDValidation.validateNumerical("softmaxCrossEntropy", "logitPredictions", logitPredictions); SDValidation.validateNumerical("softmaxCrossEntropy", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.SoftmaxCrossEntropyLoss(sd,oneHotLabels, logitPredictions, weights, org.nd4j.autodiff.loss.LossReduce.MEAN_BY_NONZERO_WEIGHT_COUNT, 0.0).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }
Example 20
Source File: SDLoss.java From deeplearning4j with Apache License 2.0 | 3 votes |
/** * Applies the softmax activation function to the input, then implement multi-class cross entropy:<br> * {@code -sum_classes label[i] * log(p[c])} where {@code p = softmax(logits)}<br> * If {@link LossReduce#NONE} is used, returned shape is [numExamples] out for [numExamples, numClasses] predicitons/labels;<br> * otherwise, the output is a scalar.<br> * <p><br> * When label smoothing is > 0, the following label smoothing is used:<br> * <pre><br> * {@code numClasses = labels.size(1);<br> * oneHotLabel = (1.0 - labelSmoothing) * oneHotLabels + labelSmoothing/numClasses}<br> * </pre><br> * * @param name name May be null. Name for the output variable * @param oneHotLabels Label array. Should be one-hot per example and same shape as predictions (for example, [mb, nOut]) (NUMERIC type) * @param logitPredictions Predictions array (pre-softmax) (NUMERIC type) * @param weights Weights array. May be null. If null, a weight of 1.0 is used (NUMERIC type) * @param lossReduce Reduction type for the loss. See {@link LossReduce} for more details. Default: {@link LossReduce#MEAN_BY_NONZERO_WEIGHT_COUNT} * @param labelSmoothing Label smoothing value. Default value: 0 * @return output Loss variable (NUMERIC type) */ public SDVariable softmaxCrossEntropy(String name, SDVariable oneHotLabels, SDVariable logitPredictions, SDVariable weights, LossReduce lossReduce, double labelSmoothing) { SDValidation.validateNumerical("softmaxCrossEntropy", "oneHotLabels", oneHotLabels); SDValidation.validateNumerical("softmaxCrossEntropy", "logitPredictions", logitPredictions); SDValidation.validateNumerical("softmaxCrossEntropy", "weights", weights); SDVariable out = new org.nd4j.linalg.api.ops.impl.loss.SoftmaxCrossEntropyLoss(sd,oneHotLabels, logitPredictions, weights, lossReduce, labelSmoothing).outputVariable(); out.markAsLoss(); return sd.updateVariableNameAndReference(out, name); }