In this article

Criterion

|
  • Share via
  •  Edit this article

Criterion

[class] L1NormLoss

(L1NormLoss &key (reduction :mean))

Implements L1Norm Loss:

l(x, y) = L = {l_1, ..., l_n}^\intercal, l_n = abs(x_n - y_n)

reduction is one of :mean or :sum.

[function] !l1norm

(!l1norm a b &key (reduction :mean))

Computes L1Norm

[class] MSELoss

(MSELoss &key (reduction :mean))

Implements MSE Loss:

l(x, y) = L = {l_1, ..., l_n}^\intercal, l_n = (x_n - y_n)^2

reduction is one of :mean or :sum.

[function] !mse

(!mse a b &key (reduction :mean))

Computes MSE Loss between a and b.

[class] CrossEntropyLoss

(CrossEntropyLoss &key (reduction :mean) (delta 1e-5))

Returns a tensor that measures the Cross-Entropy-Error between each element in the x and labels. labels are one-hot encoded.

L_i = -p_ilog(x_i + delta)

\begin{equation}
  out_i=
  \begin{cases}
    sum(L)  & \text{reduction = sum} \\
    mean(L) & \text{reduction = mean} \\
    L       & \text{otherwise}
  \end{cases}
\end{equation}

[function] !cross-entropy

(!cross-entropy x labels &key (reduction :mean) (delta 1e-5))
Computes Cross Entropy Loss between x and labels. labels are one-hot encoded.