msereg (Neural Network Toolbox)

Neural Network Toolbox

msereg

Mean squared error with regularization performance function

Syntax

perf = msereg(E,X,PP)

perf = msereg(E,net,PP)

info = msereg(code)

Description

msereg is a network performance function. It measures network performance as the weight sum of two factors: the mean squared error and the mean squared weight and bias values.

msereg(E,X,PP) takes from three arguments,

E -- Matrix or cell array of error vector(s)

X -- Vector of all weight and bias values

PP -- Performance parameter

where PP defines one performance parameters,

PP.ratio -- Relative importance of errors vs. weight and bias values

and returns the sum of mean squared errors (times PP.ratio) with the mean squared weight and bias values (times 1--PP.ratio).

The errors E can be given in cell array form,

E -- Nt x TS cell array, each element E{i,ts} is an Vi x Q matrix or []

or as a matrix,

E -- (sum of Vi) x Q matrix

where

Nt = net.numTargets

TS = Number of time steps

Q = Batch size

Vi = net.targets{i}.size

msereg(E,net) takes an alternate argument to X and PP,

net -- Neural network from which X and PP can be obtained

msereg(code) returns useful information for each code string:

'deriv' -- Name of derivative function

'name' -- Full name

'pnames' -- Names of training parameters

'pdefaults' -- Default training parameters

Examples

Here a two-layer feed-forward is created with a one-element input ranging from -2 to 2, four hidden tansig neurons, and one purelin output neuron.

net = newff([-2 2],[4 1]
{'tansig','purelin'},'trainlm','learngdm','msereg');

Here the network is given a batch of inputs P. The error is calculated by subtracting the output A from target T. Then the mean squared error is calculated using a ratio of 20/(20+1). (Errors are 20 times as important as weight and bias values).

p = [-2 -1 0 1 2];
t = [0 1 1 1 0];
y = sim(net,p)
e = t-y
net.performParam.ratio = 20/(20+1);
perf = msereg(e,net)

Network Use

You can create a standard network that uses msereg with newff, newcf, or newelm.

To prepare a custom network to be trained with msereg, set net.performFcn to 'msereg'. This will automatically set net.performParam to msereg's default performance parameters.

In either case, calling train or adapt will result in msereg being used to calculate performance.

See newff or newcf for examples.

See Also

mse, mae, dmsereg

mse negdist