有人了解RCE网络分类器
有人了解RCE网络分类器么 在线等 求高人指点
[解决办法]
The pseudo-code for a typical RCE or PRCE training procedure is shown below. In the
procedure, the italicized steps are for PRCE only. All others apply to both RCE and PRCE.
{ // Learn RCE/PRCE
Set λmin and λmax
do
{ // begin epoch
Reset Ck's
do
{ // learn vector
Input next vector and its associated class (classi
).
Compute the input vector's distance to each of the stored prototypes.
Compare distances to corresponding prototype's lambdas and determine firings.
Compute Dmin using the Dmin calculation procedure shown below.
If ((no prototypes of classi
exist) or (no prototypes fire))
store input vector with λ = Dmin
else for k = 1 to khighest stored
if (Pk firing and (class of Pk = classi
)) then Ck = Ck + 1
if (Pk firing and (class of Pk != classi
) and (λ of Pk != λmin )) then
{
if (distance to Pk > λmin ) then λ of Pk = distance to Pk
else
λ of Pk = λmin
}
}
} while more input vectors are available // learn vector
} while ((new prototypes were stored) or (any λ changed) // end of epoch
} // Learn RCE/PRCE
In the above pseudo-code, Dmin is calculated as:
Dmin = λmax
For k = 1 ...
During classification, the algorithm computes the distances from the input vector to each of the
prototype vectors stored during learning. If the distance to a prototype is less than the
prototypeís lambda, the input receives the prototype's class label. The result of the RCE
classification is a union of all firing classes.
这是我找到的rce算法的伪代码
