Continue to Site

Welcome to EDAboard.com

Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.

Register Log in

Back propagation algorithm

Status
Not open for further replies.
V

varshaturkar

Newbie level 3
Joined
Jan 23, 2011
Messages
4
Helped
1
Reputation
2
Reaction score
1
Trophy points
1,283
Activity points
1,310
Can any one tell me how to test the network after it is trained? How to use the weights which are generated during training
 

input the test signal to final updated weights, and check the output generated.....at the time of training do not use all the training pairs....some input-output pairs are kept for testing....

if the results are not correct means your network has memorized the patterns and need more training
 

I m getting good results for training. I m classifying the data into 3 classes so in training i m getting 3 weights one for each class. Shall I apply all three weights to all the test inputs and see the result? when i did this no matter what the input is according to weights i m getting the classes. can u pls help me
 

let me get it clearly...
you mean to say that you are getting three weight VECTORs say [Wx]for connections between the hidden layer and the output layer, ....classifying the data into three classes means you have three outputs of which only one is high at a time. right?
it
for each training pair input, all the 3 sets of vectors [ Wx] would have to be updated....
I am failing to see the problem....
"no matter what the input is according to weights i m getting the classes. "
I am not getting this.
You can try changing the sequence of application of training pair.
 

in input layer i have 3 neurons in hidden 3 and at output 3. so total weights will be 6. 3 between input and hidden and 3 between hidden and output.
I am applying training samples. and computed average wts for each class.
these average wts i am using for testing. i m giving training to the system randomly i.e. class 1 samples then class2 again class 1 and so on
 
by the archi that you are describing , their should be two weight matrix 1)for between the input and the hidden layer 2)for between the output and the hidden layer ., each of order (3x3). how are you getting 6 weights???
 

sorry its not 6 i m getting 2 matrices as u said. when i start testing i m applying the wt matrices which i get after training but problem is, in training samples if the last class is suppose "class 2" then in test phase for all classes it is giving "class 2". I think this is because the wts are getting adjusted according to last class at the end of training. I tried giving a big set of test samples having various classes not in sequence still the network is remembering the wts for last class in the training phase.
 

am starting fresh in BPA can anyone suggest a gud reference.On googling i ended up in many formulas n all din get head n tail out of it.
please help
 

Status
Not open for further replies.

Similar threads

Part and Inventory Search

Welcome to EDABoard.com

Sponsor

Back
Top