Neural Network Toolbox

trainb

Batch training with weight and bias learning rules.

Syntax

[net,TR,Ac,El] = trainb(net,Pd,Tl,Ai,Q,TS,VV,TV)

info = trainb(code)

Description

trainb is not called directly. Instead it is called by train for networks whose net.trainFcn property is set to 'trainb'.

trainb trains a network with weight and bias learning rules with batch updates. The weights and biases are updated at the end of an entire pass through the input data.

trainb(net,Pd,Tl,Ai,Q,TS,VV,TV) takes these inputs,

• net -- Neural network

  Pd -- Delayed inputs

  Tl -- Layer targets

  Ai -- Initial input conditions

  Q -- Batch size

  TS -- Time steps

  VV -- Empty matrix [] or structure of validation vectors

  TV -- Empty matrix [] or structure of test vectors

and returns,

• net -- Trained network

  TR -- Training record of various values over each epoch:

  • TR.epoch -- Epoch number

    TR.perf -- Training performance

    TR.vperf -- Validation performance

    TR.tperf -- Test performance

  Ac -- Collective layer outputs for last epoch.

  El -- Layer errors for last epoch

Training occurs according to the trainb's training parameters, shown here with their default values:

• net.trainParam.epochs      100 Maximum number of epochs to train

  net.trainParam.goal          0 Performance goal

  net.trainParam.max_fail      5 Maximum validation failures

  net.trainParam.show         25 Epochs between displays (NaN for no                                displays)

  net.trainParam.time        inf Maximum time to train in seconds

Dimensions for these variables are:

• Pd -- No x Ni x TS cell array, each element Pd{i,j,ts} is a Dij x Q matrix

  Tl -- Nl x TS cell array, each element P{i,ts} is a Vi x Q matrix or []

            Ai -- Nl x LD cell array, each element Ai{i,k} is an Si x Q matrix

where

• Ni = net.numInputs

            Nl = net.numLayers

            LD = net.numLayerDelays

  Ri = net.inputs{i}.size

  Si = net.layers{i}.size

  Vi = net.targets{i}.size

  Dij = Ri * length(net.inputWeights{i,j}.delays)

If VV or TV is not [], it must be a structure of vectors:

• VV.PD, TV.PD -- Validation/test delayed inputs

  VV.Tl, TV.Tl -- Validation/test layer targets

  VV.Ai, TV.Ai -- Validation/test initial input conditions

  VV.Q, TV.Q -- Validation/test batch size

  VV.TS, TV.TS -- Validation/test time steps

Validation vectors are used to stop training early if the network performance on the validation vectors fails to improve or remains the same for max_fail epochs in a row. Test vectors are used as a further check that the network is generalizing well, but do not have any effect on training.

trainb(CODE) returns useful information for each CODE string:

• 'pnames' -- Names of training parameters

  'pdefaults' -- Default training parameters

Network Use

    You can create a standard network that uses trainb by calling newlin.

To prepare a custom network to be trained with trainb

1. Set net.trainFcn to 'trainb'.

• (This will set NET.trainParam to trainb's default parameters.)

2. Set each NET.inputWeights{i,j}.learnFcn to a learning function.
3. Set each NET.layerWeights{i,j}.learnFcn to a learning function.
4. Set each NET.biases{i}.learnFcn to a learning function. (Weight and bias learning parameters will automatically be set to default values for the given learning function.)

To train the network

1. Set NET.trainParam properties to desired values.
2. Set weight and bias learning parameters to desired values.
3. Call train.

See newlin for training examples

Algorithm

Each weight and bias updates according to its learning function after each epoch (one pass through the entire set of input vectors).

Training stops when any of these conditions are met:

• The maximum number of epochs (repetitions) is reached.
• Performance has been minimized to the goal.
• The maximum amount of time has been exceeded.
• Validation performance has increase more than max_fail times since the last time it decreased (when using validation).

See Also

newp, newlin, train

train

trainbfg

© 1994-2005 The MathWorks, Inc.