Wrong classification with RProp in neural network demo

Wrong classification with RProp in neural network demo - Printable Version

+- Forums (http://mitov.com/forum)
+-- Forum: VCL Components (/forum-6.html)
+--- Forum: IntelligenceLab (/forum-39.html)
+--- Thread: Wrong classification with RProp in neural network demo (/thread-309.html)

Wrong classification with RProp in neural network demo - benjamin - 04-25-2012 01:37 AM

Hi all,

am I the only one who gets wrong classifications with the RProp training method in the neural network demo.

In 28% of the cases the classification ist wrong. Backpropagation works fine for me. I attached some screenshots with the wrong classification.

Is this an error in the component or is this normal for RProp?

greetings
Benjamin

RE: Wrong classification with RProp in neural network demo - Dave - 04-25-2012 04:27 AM

Hi! You're right, you are re-training a network (or overtraining).

Regards,
Dave

RE: Wrong classification with RProp in neural network demo - benjamin - 05-01-2012 07:13 PM

Hi

If I understand it right the problem with re-training should be solved if I add

Code:

ILNeuralNetwork1->Clear();

at the beginning of the training.

Unfortunately this solves the problem not completely. The wrong classifications become rare but they still occur.

Greetings
Benjamin