Examples with SDVariable org.nd4j.autodiff.samediff.SDVariable used on opensource projects

Example 91 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class LossFunctions method l2.

/**
 * L2 loss function: i.e., sum of squared errors, L = sum_i (actual_i - predicted)^2
 *
 * @param outputName
 * @param predictions
 * @param label
 * @param weights
 * @param reduction
 * @param dimensions
 * @return
 */
public static LossInfo l2(String outputName, SDVariable predictions, SDVariable label, SDVariable weights, Reduction reduction, int... dimensions) {
    LossInfo.Builder b = validate("l2", predictions, label, reduction);
    SameDiff sd = predictions.getSameDiff();
    if (weights == null) {
        weights = sd.one("l2_loss_weights", SCALAR);
    }
    SDVariable diff = predictions.sub(label);
    String name = (reduction == Reduction.NONE ? outputName : null);
    SDVariable preReduceLoss = sd.square(diff).mul(name, weights);
    return doReduce(sd, outputName, false, b, reduction, preReduceLoss, label, weights, dimensions);
}

Example 92 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class LossFunctions method mcxent.

/**
 * Multi-Class Cross Entropy loss function:<br>
 * L = sum_i actual_i * log( predicted_i )
 *
 * @param outputName
 * @param predictions
 * @param label
 * @param weights
 * @param reduction
 * @param dimensions
 * @return
 */
public static LossInfo mcxent(String outputName, SDVariable predictions, SDVariable label, SDVariable weights, Reduction reduction, int... dimensions) {
    LossInfo.Builder b = validate("mcxent", predictions, label, reduction);
    SameDiff sd = predictions.getSameDiff();
    if (weights == null) {
        weights = sd.one("mcxent_loss_weights", SCALAR);
    }
    String name = (reduction == Reduction.NONE ? outputName : null);
    SDVariable weightedLogProd = sd.log(predictions).mul(label).mul(name, weights);
    return doReduce(sd, outputName, false, b, reduction, weightedLogProd, label, weights, dimensions);
}

Example 93 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class LossFunctions method l1.

/**
 * L1 loss - sum of absolute errors. L = sum_i abs(predicted_i - actual_i)
 *
 * @param outputName
 * @param predictions
 * @param label
 * @param weights
 * @param reduction
 * @param dimensions
 * @return
 */
public static LossInfo l1(String outputName, SDVariable predictions, SDVariable label, SDVariable weights, Reduction reduction, int... dimensions) {
    LossInfo.Builder b = validate("l1", predictions, label, reduction);
    SameDiff sd = predictions.getSameDiff();
    if (weights == null) {
        weights = sd.one("l1_loss_weights", SCALAR);
    }
    String name = (reduction == Reduction.NONE ? outputName : null);
    SDVariable preReduceLoss = sd.abs(predictions.sub(label)).mul(name, weights);
    return doReduce(sd, outputName, false, b, reduction, preReduceLoss, label, weights, dimensions);
}

Example 94 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class DifferentialFunctionFactory method batchNorm.

/**
 * Batch norm operation.
 */
public SDVariable batchNorm(SDVariable input, SDVariable mean, SDVariable variance, SDVariable gamma, SDVariable beta, boolean applyGamma, boolean applyBeta, double epsilon) {
    BatchNorm batchNorm = BatchNorm.builder().inputFunctions(new SDVariable[] { input, mean, variance, gamma, beta }).applyGamma(applyGamma).applyBeta(applyBeta).epsilon(epsilon).sameDiff(sameDiff()).build();
    val outputVars = batchNorm.outputVariables();
    return outputVars[0];
}

Example 95 with SDVariable

use of org.nd4j.autodiff.samediff.SDVariable in project nd4j by deeplearning4j.

the class DifferentialFunctionFactory method doGradChoose.

/**
 * @param func
 * @param input
 * @return
 */
public SDVariable doGradChoose(SDVariable func, SDVariable input) {
    validateDifferentialFunctionsameDiff(func);
    validateDifferentialFunctionsameDiff(input);
    SDVariable repeatedGrad = doRepeat(func, input);
    SDVariable resultRepeated = doRepeat(func.args()[0], input);
    SDVariable argMaxLocations = eq(input, resultRepeated);
    return div(mul(argMaxLocations, repeatedGrad), sum(argMaxLocations).outputVariables()[0]);
}