Libraries: PNN Neural Network Class

New comment
 

PNN Neural Network Class:

The class realizes the probabilistic neural network (Probabilistic Neural Network - PNN)

Author: Yury Kulikov

 

Hi Yury Kulikov

        0 - network learning is completed and learning result can be checked through the class variable: mse – learning error, epoch – number of accomplished learning cycles;

 can you tell me   'mse – learning error '  general between witch interval ??  I test one and get the MSE=7.218702473434161e-008 ,is it all right ??

thank you very much! 


 
qingyouwei:
MSE less is better. But one must bear in mind that a very small error value can indicate retraining network.
 

Yurich:
MSE less is better. But one must bear in mind that a very small error value can indicate retraining network.

thaks very much.  But i can't get a small error value , in the class "class_pnn"  how did you exit leanning ? can let  it calculate a long time to get a small error values?

 


qingyouwei: 

... I test one and get the MSE=7.218702473434161e-008 ,is it all right ??

This is quite a small error.

Error and learning time depends on the prepared data for training. Preparing the data is a separate issue and it needs to explore before application of neural networks.

 
any example on how to use?
 

Nice. Even nicer that it works perfectly in MT4 too.

With a large number of samples the network does get rather large because it stores all the training samples. A useful addition would be a function to reduce the network by removing any samples that increase the total error.

Another useful addition would be the possibility of adding new training samples at a later date.

I'll contribute some code once I have figured out why I'm getting nan values for the mse from time to time. 

 
Jimmy Tee:

Nice. Even nicer that it works perfectly in MT4 too.

With a large number of samples the network does get rather large because it stores all the training samples. A useful addition would be a function to reduce the network by removing any samples that increase the total error.

Another useful addition would be the possibility of adding new training samples at a later date.

I'll contribute some code once I have figured out why I'm getting nan values for the mse from time to time. 

I also get mse = NaN time to time. This happens completely in random which is kind of strange. I will take a look at the code to see what could be the bug...

New comment
Reason: