Loss Layers

Layer	Description
CategoricalAccuracy	0-1 loss function
CrossEntropy	Cross entropy between probability vectors
L1Norm	L1 vector norm
L2Norm2	Square of L2 vector norm
MeanAbsoluteError	Mean absolute error
MeanSquaredError	Mean squared error
TopKCategoricalAccuracy	Top-k prediction scores

CategoricalAccuracy

The CategoricalAccuracy Layer is a 0-1 loss function.

Requires two inputs, which are respectively interpreted as prediction scores and as a one-hot label vector. The output is one if the top entries in both inputs are in the same position and is otherwise zero. Ties are broken in favor of entries with smaller indices.

This is primarily intended for use as a metric since it is not differentiable.

Arguments: None

Back to Top

CrossEntropy

The CrossEntropy layer measures the probability and error between vectors.

Given a predicted distribution \(y\) and ground truth distribution \(\hat{y}\),

\[CE(y,\hat{y}) = - \sum\limits_{i} \hat{y}_i \log y_i\]

Arguments:

use_labels

(bool) Advanced option for distconv

Back to Top

L1Norm

The L1Norm layer is the L1 norm of a vector.

\[\lVert x\rVert_1 = \sum\limits_{i} | x_i |\]

Arguments: None

Back to Top

L2Norm2

The L2Norm2 layer is the square of L2 vector norm.

\[\lVert x\rVert_2^2 = \sum\limits_{i} x_i^2\]

Arguments: None

Back to Top

MeanAbsoluteError

The MeanAbsoluteError given a prediction \(y\) and ground truth \(\hat{y}\):

\[MAE(y,\hat{y}) = \frac{1}{n} \sum\limits_{i=1}^{n} | y_i - \hat{y}_i |\]

Arguments: None

Back to Top

MeanSquaredError

The MeanSquaredError layer given a prediction \(y\) and ground truth \(\hat{y}\):

\[MSE(y,\hat{y}) = \frac{1}{n} \sum\limits_{i=1}^{n} (y_i - \hat{y}_i)^2\]

Arguments: None

Back to Top

TopKCategoricalAccuracy

The TopKCategoricalAccuracy layer requires two inputs, which are respectively interpreted as prediction scores and as a one-hot label vector. The output is one if the corresponding label matches one of the top-k prediction scores and is otherwise zero. Ties in the top-k prediction scores are broken in favor of entries with smaller indices.

Arguments:

k

(int64)

Back to Top