sN.MLPtrain {simpleNeural}

Trains a multilayer perceptron with 1 hidden layer

Description

Trains a multilayer perceptron with 1 hidden layer and a sigmoid activation function, using backpropagation and gradient descent. Don't forget to normalize the data first - sN.normalizeDF(), provided in the package, can be used to do so.

Usage

sN.MLPtrain(X, y, hidden_layer_size = 5, it = 50, lambda = 0.5,
  alpha = 0.5)

Arguments

X

Matrix of predictors

y

Vector of output (the ANN learns y=ANN(X)). Classes should be assigned an integer number, starting at 0 for the first class.

hidden_layer_size

Number of units in the hidden layer

it

Number of iterations for the gradient descent. The default value of 50 may be a little low in some cases. 100 to 1000 are generally sensible values.

lambda

Penalization for model coefficients (regularization parameter)

alpha

Speed multiplier (learning rate) for gradient descent

Value

The coefficients of the MLP, in a list (Theta1 between input and hidden layers, Theta2 between hidden and output layers)

References

M.W Gardner, S.R Dorling, Artificial neural networks (the multilayer perceptron)- a review of applications in the atmospheric sciences, Atmospheric Environment, Volume 32, Issues 14-15, 1 August 1998, Pages 2627-2636, ISSN 1352-2310, doi: 10.1016/S1352-2310(97)00447-0 [http://www.sciencedirect.com/science/article/pii/S1352231097004470]

Jain, A.K.; Jianchang Mao; Mohiuddin, K.M., "Artificial neural networks: a tutorial," Computer , vol.29, no.3, pp.31,44, Mar 1996. doi: 10.1109/2.485891 [http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=485891&isnumber=10412]

Rumelhart, David E., Geoffrey E. Hinton, and R. J. Williams. "Learning Internal Representations by Error Propagation". David E. Rumelhart, James L. McClelland, and the PDP research group (editors). Parallel distributed processing: Explorations in the microstructure of cognition, Volume 1: Foundations. MIT Press, 1986.

Examples

# NB: the provided examples are just here to help use the package's functions.
# In real use cases you should perform a proper validation (cross-validation,
# external validation data...)
data(UCI.BCD.Wisconsin);
X=as.matrix(sN.normalizeDF(as.data.frame(UCI.BCD.Wisconsin[,3:32])));
y=as.matrix(UCI.BCD.Wisconsin[,2]);
myMLP=sN.MLPtrain(X=X,y=y,hidden_layer_size=20,it=50,lambda=0.5,alpha=0.5);
myPrediction=sN.MLPpredict(nnModel=myMLP,X=X,raw=TRUE);
#library('verification');
#roc.area(y,myPrediction[,2]);