Contents

init(learningRate:alpha:epsilon:centered:momentum:gradientScale:regularizationScale:gradientClipping:regularizationFunction:)

Returns a new RMSProp optimizer object with gradient clipped by value or clipped by norm.

Declaration

init(learningRate: Float = 1e-2, alpha: Float = 0.99, epsilon: Float = 1e-8, centered: Bool, momentum: Float = 0, gradientScale: Float, regularizationScale: Float, gradientClipping: BNNS.GradientClipping, regularizationFunction: BNNSOptimizerRegularizationFunction)

Parameters

  • learningRate:

    A value that specifies the learning rate.

  • alpha:

    A constant that specifies smoothing, in the range 0 to 1.

  • epsilon:

    A term that the optimizer adds to the denominator.

  • centered:

    A Boolean value that specifies whether to use the centered variant.

  • momentum:

    The rate of momentum decay.

  • gradientScale:

    A value that specifies the gradient scaling factor.

  • regularizationScale:

    A value that specifies the regularization scaling factor.

  • gradientClipping:

    The gradient clipping function and bounds.

  • regularizationFunction:

    The variable that specifies the regularization function.

See Also

Creating an RMSProp Optimizer