neural (version 1.4.2.1)

mlptrain: MLP neural network

Description

A simple MLP neural network that is suitable for classification tasks.

Usage

mlptrain(inp,neurons,out,weight=c(),dist=c(),alfa=0.2,it=200,online=TRUE,
		permute=TRUE,thresh=0,dthresh=0.1,actfns=c(),diffact=c(),visual=TRUE, ...)

Arguments

inp

a matrix that contains one input data in each row.

neurons

a numeric vector with length equals to the number of layers in the network, and the ith layer will contains neurons[i] neuron.

out

a matrix that contains one output data in each row.

weight

the starting weights of the network.

dist

the starting distortions of the network.

alfa

the learning-rate parameter of the back-propagation algorithm.

it

the maximum number of training iterations.

online

if TRUE the algorithm will operate in sequential mode of back-propagation,if FALSE the algorithm will operate in batch mode of back-propagation.

permute

if TRUE the algorithm will use a random permutation of the input data in each epoch.

thresh

the maximal difference between the desired response and the actual response that is regarded as zero.

dthresh

if the difference between the desired response and the actual response is lesser than this value, the corresponding neuron is drawn in red, otherwise it is drawn in green.

actfns

a list, which contains the numeric code of the activation functions, or the activation function itself. The length of the vector must be the same as the length of the neurons vector, and each element of the vector must be between 1 and 4 or a func

diffact

a list, which contains the differential of the activation functions. Only need to use if you use your own activation functions.

visual

a logical value, that switches on/off the graphical user interface.

...

currently not used.

Value

• a list with 5 arguments:
• weight
  the weights of the network.
• dist
  the distortions of the network.
• neurons
  a numeric vector with length equals to the number of layers in the network, and the ith layer will contains neurons[i] neuron.
• actfns
  a list, that contains the activation functions. The length of the list is equal with the number of active layers.
• diffact
  a list, which contains the differential of the activation functions. The length of the list is equal with the number of active layers.

Details

The function creates an MLP neural network on the basis of the function parameters. After the creation of the network it is trained with the back-propagation algorithm using the inp and out parameters. The inp and out parameters has to be the same number of rows, otherwise the function will stop with an error message. If you use the weight or dist argument, than that variables won't be determined by random. This could be useful if you want to retrain your network. In that case use both of this two arguments. From this vesion of the package there is the chance to use your own activation functions, by using the actfns argument. If you do this, don't forget to set the differential of the activation functions in the diffact argument at the same order and the same position where you are using the new activation function. (No need of using the diffact argument if you're using the preset activation functions.) The function has a graphical user interface that can be switched on and off using the visual argument. If the graphical interface is on, the activation functions can be set in manually. If the activation functions are not set then each of them will be automatically the logistic function. The result of the function are the parameters of the trained MLP neural network. Use the mlp function for information recall.

See Also

`mlp' for recall; `rbftrain' and `rbf' for training an RBF network.

Examples

x<-matrix(c(1,1,0,0,1,0,1,0),4,2)
	y<-matrix(c(0,1,1,0),4,1)
	neurons<-4
	data<-mlptrain(x,neurons,y,it=4000);
	mlp(x,data$weight,data$dist,data$neurons,data$actfns)