ISupportSdcaLoss 接口

定义

损失函数可能知道最佳双更新参考的接近形式解决方案:秒 (6.2) http://jmlr.org/papers/volume14/shalev-shwartz13a/shalev-shwartz13a.pdf

public interface ISupportSdcaLoss : Microsoft.ML.Trainers.ILossFunction<float,float>, Microsoft.ML.Trainers.IScalarLoss
type ISupportSdcaLoss = interface
    interface IScalarLoss
    interface ILossFunction<single, single>
Public Interface ISupportSdcaLoss
Implements ILossFunction(Of Single, Single), IScalarLoss
派生
实现

方法

ComputeDualUpdateInvariant(Single)

损失函数可能知道最佳双更新参考的接近形式解决方案:秒 (6.2) http://jmlr.org/papers/volume14/shalev-shwartz13a/shalev-shwartz13a.pdf

Derivative(Single, Single)

与输出相关的损失函数的派生

(继承自 IScalarLoss)
DualLoss(Single, Single)

训练示例的双重损失函数。 如果 f (x) 表示单个训练示例中的损失函数,则此函数返回 -f* (-x*) ,其中 f* (x*) 是 f (x) 的 Fenchel 串联。

DualUpdate(Single, Single, Single, Single, Int32)

在 SDCA 中计算双更新 (\Delta\alpha_i)

  • alpha:指定实例上的双变量
  • lambdaN:L2 const x 实例数
  • 缓存固定,由上述方法提示
Loss(TOutput, TLabel)

计算给定输出和地面真相的损失。 请注意,返回值具有 Double 类型,因为损失通常累积在多个实例上。

(继承自 ILossFunction<TOutput,TLabel>)

适用于