How to test neural network with real world data after training it ? How to interpret output of ANN?

I am currently working with ANN for pattern recognition to identify different geological features.I am new to ANN.I am using Matlab R2011a Version 7.12.0 ANN tool box (nnstart) pattern recognition tool(nprtool).I trained ANN with input data (5x25 matrix) and target data (3x25 matrix).The algorithm itself dividing it into training,validation,test data.I followed the steps in tool box (nnstart) and finally the network was trained.Then it produced confusion plot and other.My query is how to test the data from the real world after training the network with the above input and target data?How to interpret that to which class the given input belongs from the result of ANN.Thank you all. Regards Suresh

[Shoib Sho]

Hello, The above sent link has some error can u resent the link again..My query is also same as above..

[Greg Heath]

 Although the content has been obviously been changed, what is 
there is appropriate.

Hope this helps.

Greg

Answer 1

[Shashank Prasanna]

This link should be helpful to you:

http://www.mathworks.com/help/nnet/gs/recognizing-patterns.html

Scroll down where it is demonstrated how to call the trained network with new inputs and how to test its performance as well.

[Suresh]

Dear sir, Thank you verymuch for providing me with the link.As per link "You can also save the network as net in the workspace. You can perform additional tests on it or put it to work on new inputs."I followed the procedure and saved the network in matlab workspace.Now how to proceed further? Thank you verymuch for spending your valuable time.

[Shashank Prasanna]

If you scroll a little further in the command line section you will see:

% Test the Network
outputs = net(inputs);
errors = gsubtract(targets,outputs);
performance = perform(net,targets,outputs)

You can use the net object just as described to predict for new inputs. Go through the whole document. You will find several examples in the documentation such as this, go through all of them.

[Suresh]

For exmple I am giving a 10x50 matrix as input and 3x50 as target data and I trained the network using nprtool.I saved the network as net,inputs,targets and output.Now If I want to test the real data 'P' (10x2) matrix ,to which I do not have any target data,what steps do i follow. outputtest=net(P);

[Pranava Koustubh Boppudi]

Suresh, the targets for training are used to help the neural network understand that these are the outputs you're looking for. They are not used while testing. Testing phase is when your previously trained network is now classifying new unseen data. If you want to validate your neural net for new data, you'll need targets. This is already being done by the neural network tool. It splits the dataset into training, validation and testing. That is how it validates itself. Hope this is clear.

[sunil kalyankar]

prassanna sir i am not getting from above the link i also having same problem how to use trained network to new inputs please help me thanks and regards

Answer 2

[mohamed mohamed]

hei , i had the same problem , my porblem was how to use the inputs and outputs in the NN model in matlab , and i found a way , maybe i am wrong but it worth to try, because at the end i got what i want , you can use timeseries with your data , for every step , the time will be the steps and u can use from work space block as well , and that is the way i used and i go the result from the neural network !