Description: |
|
This quantlet computes different networks of the form single
layer feedforward perceptron. The quantlet can be used alone
or in connection with the library ISTA. The standalone version
also needs the parameter data. Just choose 0 for the input.
It is possible to split the data in a training and a test set.
The weight for the cases for the training of the net can be
chosen, the numbers of hidden units with ``from, stepwidth,
to'' and additional information concerning the weights of
the units. Different optional parameters can be chosen to
build the architektur of the network. The choice holds for
every single net. The default values are chosen in
order to solve a linear regression problem. The optional
parameters constits of 8 values. Boolean values for linear
output, entropy error function, log probability models and
for skip connections (direkt links). The fifth values is
the maximum value for the starting weights, the sixth is the
weight decay, the seventh the maximum number of iterations
and the the last value generates the output concerning the
architekur of the net if it is equal to one.
The output consits of the Error and MSE of the different nets
(MSE for test and trainings data separately if chosen) and
the R^2.
|