After training the neural network,and getting the bais and weigts values,the prediction from manual calculation is not correct. How to re-normalize the weights and bias values.
Dear Mr Luca
I trained FFMLP with Levenberg-Marquard back propagation. After training the network when i used the weights and bias values for my manual calculation the prediction is not correct. I use the same input (normalize values) for my calculation.
I got some reply for mapminmax problem but it is not clear.
Waiting for your reply.
Regards
A Ahmad
I also have similar problem and haven't yet solved. The following are two responses I got from various communities :-
1) "MAPMINMAX is not used correctly: The parameters obtained from the training input should be used on the test input. The inverse parameters obtained from the training target should be used on the test output."
2) "Heuristic method are not searching the domain completely and they are not formal, if you didn't design it and you had been used its template, it would be different on data some times. If you even run ANN or any other heuristic methods on same big data maybe you would saw these differences too. They had some conditions for stop the search, when you have the minimal error result or when your steps run out."
Attached below are the links for reference.
https://www.researchgate.net/post/Why_is_there_a_difference_between_output_of_neural_network_by_inbuilt_test_function_checks_accuracy_of_network_and_custom_designed_test_function2
https://se.mathworks.com/matlabcentral/answers/196322-why-is-there-a-difference-between-output-of-neural-network-by-inbuilt-test-function-ann-toolbox-an
Dear Mr Luca
My NN model is ST-4-5-1(sigmiod function b/w input and hidden layer and Tanh b/w hidden and output layer,with 4 input and 5 hidden neuron and one output). the overall R value is 0.95.
Now i am working on your guidelines.
Thanks
Dear Mr Luca
Thank you for your guidelines. I try to use the real data without normalization but it didnt work.
Please guide me how to use the train neural network weights and bias values for the prediction by manual calculation.
Best Reagrds
Dear Mr Luca
Please find the file. I used the trained weights and bais value for prediction but couldnt get the right value.
Regards
- Badges
- Science topic
- Similar topics
- Correction