did i need target matrix for new input testing??

hi...i create neural network....the input matrix is with 18X20...(18= number of features and 20 = number of images for 2 classes)the target matrix is 2X20...after traning i want test the network with new image...the new enter images is matrix of 18X1.. i used this method for testing it:

simpleclassOutputs = sim(mynet11,input); figure();plotroc(T,simpleclassOutputs);

T= is target matrix which used in traning>> the error is ... Index exceeds matrix dimensions.

note: i used R2013a matlab and i am new in neural network

Error in plotconfusion>update_plot (line 396) y = y(:,known);

Error in plotconfusion (line 108) plotData = update_plot(param,fig,plotData,update_args{:});

Error in train (line 223) figure,plotconfusion(T,out);

my question is: how to sove this problem?? did i need target matrix for new input testing?? plz help me/

Answer 1

[Greg Heath]

If you have a new input you can use the net to obtain an output and the corresponding classification. However, without the corresponding target, you cannot determine, directly, if the classification is correct. Of course you could use the original data and design multiple classifiers of different types to estimate the probability of the classification being correct.

I think the roc plot needs output and target to be of the same size.

Hope this helps.

Thank you for formally accepting my answer

Greg

[Image Analyst]

primrose's 3 "Answers" moved here since they were not actually answers to the original posted question but actually replies to Greg.

thanks greg for your replay...i dont understand your answer can you explane your answer,,can i solve my problem or not>> ??? plz help me

plz can you help me??? after trainig network how can display the images matching which has high simlirty with testing images????????

hi greg ...this is my file ..can you help me plz

By the way, no file was attached in the "Answer".

Answer 2

[primrose khaleed]

this is file

[Greg Heath]

 class = vec2ind(output) will give a 1 or 2 for each column of output.

If you know the correct class you can count the errors for each class and obtain the class per cent error rates.

Since each class is defined by multiple examples, how do you want to display the correct answer as a single image? The mean of the class members or the member that is the most similar?

Your choice.

[primrose khaleed]

thank you very much...when i enter image from first class the output=1 and when enter image of second class the output=2...noow i want to used :

figure,plotconfusion(T,simpleclassOutputs);

but the there is error: Index exceeds matrix dimensions.

Error in plotconfusion>update_plot (line 396) y = y(:,known);

Error in plotconfusion (line 108) plotData = update_plot(param,fig,plotData,update_args{:});

Error in train (line 233) figure,plotconfusion(T,simpleclassOutputs);

can you solve this problem greg...i need your help...plz

Answer 3

[primrose khaleed]

hi greg ...can you see my file and check why is error...plz help me

[Greg Heath]

Sorry, I am not familiar with image functions.

[primrose khaleed]

thank you greg very much ....the i doing some processing in image this file contain the image processing :

[Image Analyst]

Did you read his comment, and his flag?

Answer 4

[primrose khaleed]

sorry image analysist....this way is correct?? thank you very much...when i enter image from first class the output=1 and when enter image of second class the output=2...noow i want to used :

    figure,plotconfusion(T,simpleclassOutputs);

but the there is error: Index exceeds matrix dimensions.

Error in plotconfusion>update_plot (line 396) y = y(:,known);

Error in plotconfusion (line 108) plotData = update_plot(param,fig,plotData,update_args{:});

Error in train (line 233) figure,plotconfusion(T,simpleclassOutputs);

can you solve this problem greg...i need your help...plz

[Greg Heath]

No, it is not correct. You used the ANSWER block instead of the COMMENT BLOCK.

plotconfusion is for matrices (not vectors) of the same size. Use confusion otherwise.

[primrose khaleed]

the problem of plotconfusion is solved when doing

plotconfusion(simpleclassOutputs2,T); instead of used plotconfusion(T,simpleclassOutputs2);

...but the problem when enter the image as test_image of first classs ...the first run of program classify as first class and the second run classify as second class ...how can solve this problem ..????plz help me