newrbe这个命令是作啥用的呀?

kevin19821

哪位高人给我讲解一下呀

[ 本帖最后由 eight 于 2007-10-28 16:29 编辑 ]

eight

原帖由 kevin19821 于 2007-10-28 14:54 发表

                               
登录/注册后可看大图

哪位高人给我讲解一下呀

>> help newrbe
NEWRBE Design an exact radial basis network.

   Synopsis

     net = newrbe(P,T,SPREAD)

   Description

     Radial basis networks can be used to approximate functions.
     NEWRBE very quickly designs a radial basis network with
     zero error on the design vectors.

     NEWRBE(P,T,SPREAD) takes two or three arguments,
     P      - RxQ matrix of Q input vectors.
     T      - SxQ matrix of Q target class vectors.
     SPREAD - of radial basis functions, default = 1.0.
     and returns a new exact radial basis network.

     The larger that SPREAD, is the smoother the function approximation
     will be. Too large a spread can cause numerical problems.

   Examples

     Here we design a radial basis network, given inputs P
     and targets T.

       P = [1 2 3];
       T = [2.0 4.1 5.9];
       net = newrbe(P,T);

     Here the network is simulated for a new input.

       P = 1.5;
       Y = sim(net,P)

   Algorithm

     NEWRBE creates a two layer network. The first layer has RADBAS
     neurons, and calculates its weighted inputs with DIST, and its
     net input with NETPROD.  The second layer has PURELIN neurons,
     and calculates its weighted input with DOTPROD and its net inputs
     with NETSUM. Both layer's have biases.

     NEWRBE sets the first layer weights to P', and the first
     layer biases are all set to 0.8326/SPREAD, resulting in
     radial basis functions that cross 0.5 at weighted inputs
     of +/- SPREAD.

     The second layer weights IW{2,1} and biases b{2} are found by
     simulating the first layer outputs A{1}, and then solving the
     following linear expression:

         [W{2,1} b{2}] * [A{1}; ones] = T

   See also sim, newrb, newgrnn, newpnn.

    Reference page in Help browser
       doc newrbe