encog Neural Network - java - Training not working

Question

First of all, I'm extremely sorry if this question is already asked.., I've searched for weeks in this forum and found none similar to my question...

I'm new to neural networks!!

I've started a project in neural networks and this Internet shows me only one example = XOR - one java code - one concept when i search with any tags related to neural-network-java-encog!!

Creating Neural network -> BasicNeural network = new..... training => iteration(), network.Compute() etc...

I tried that code for my dataset of 2000 entries and 16 binary attributes. With Sigmoid function & BACKpropagation algorithm as mentioned in the XOR example. - No result.. just infinite iterations while training(aborted the execution after ~45 minutes)...

I altered the no. of hidden layers, no. of hidden neurons, learning rate and momentum in Backpropagation's constructor... and none got it working.

I even changed the Training method to resilientPropagation & Manhattan.

But, when i reduced my dataset to 100 records, it works fine.! 3 seconds max to get trained. Even if i change the dataset to 200 records with the same configuration(no.of neurons, layers, momentum etc), it didn't work... just worried tat i couldn't understand this code's behavior..

Where am I doing wrong.? I am an absolute beginner to these concepts and trying to understand !! So , kindly accept my apologies if my texts are irritating...

I don't know where to learn about the impact of every training method and when to use it. Is backpropagation method not suitable for this purpose or is it like, that type of code won't work for bigger number of records.??

Please help me with this issue !! hope im clear with my query Thanks in advance

Answer 1

Well since nobody helped here., I tried stopping those iterations after a specific time.(say 2000 times)

Now, when I checked the program, it got trained for some extent.. And I continued my work with those...