mae

(To be removed) Mean absolute error performance function

mae will be removed in a future release. For more information, see Transition Legacy Neural Network Code to dlnetwork Workflows.

For advice on updating your code, see Version History.

Syntax

perf = mae(E,Y,X)

dPerf_dx = mae('dx',E,Y,X,perf)

info = mae('code')

Description

Tip

To use mean absolute error with deep learning, use the trainnet function and set the loss function to "mae", or use the l1loss function for dlarray objects.

perf = mae(E,Y,X) takes a matrix or cell array of error vectors, E, and optionally a matrix or cell array of output vectors, Y, a vector of all weight and bias values, X, and returns network performance as the mean of absolute errors, perf.

dPerf_dx = mae('dx',E,Y,X,perf) returns the derivative of perf with respect to X.

info = mae('code') returns useful information for each code character vector:

• mae('name') returns the name of this function.

• mae('pnames') returns the names of the training parameters.

• mae('pdefaults') returns the default function parameters.

Examples

Calculate Network Performance with 'mae'

This example shows how to calculate the network performance as the mean of absolute errors.

Create and configure a perceptron to have one input and one neuron:

net = perceptron;
net = configure(net,0,0);

The network is given a batch of inputs P. The error is calculated by subtracting the output A from target T. Then the mean absolute error is calculated.

p = [-10 -5 0 5 10];
t = [0 0 1 1 1];
y = net(p)
e = t-y
perf = mae(e)

Note that mae can be called with only one argument because the other arguments are ignored. mae supports those arguments to conform to the standard performance function argument list.

Input Arguments

E — Errors

vector | matrix | cell array

Errors, specified as a vector, a matrix, or a cell array.

Y — Outputs

vector | matrix | cell array

Network outputs, specified as a vector, a matrix, or a cell array.

X — Weight and bias

vector

Weight and bias values, specified as a vector.

Output Arguments

perf — Network performance

scalar

Network performance as the mean of absolute errors, returned as a scalar.

dPerf_dx — Derivative of network performance

scalar

Derivative of perf with respect to X, returned as a scalar.

More About

Network Use

You can create a standard network that uses mae with perceptron.

To prepare a custom network to be trained with mae, set net.performFcn to 'mae'. This automatically sets net.performParam to the empty matrix [], because mae has no performance parameters.

In either case, calling train or adapt, results in mae being used to calculate performance.

Version History

Introduced before R2006a

mae will be removed

mae will be removed in a future release.

To learn more about using metrics and loss functions when you train neural networks, see these examples and topics:

• Deep Learning Metrics

• Custom Loss Functions

• Monitor Deep Learning Training Progress

• ROC Curve and Performance Metrics

For dlnetwork objects, specify mean absolute error (MAE) loss by setting the lossFcn of the trainnet function to "mae". For custom loss functions and custom training loops, use the l1loss function.

For more information, see Transition Legacy Neural Network Code to dlnetwork Workflows.