I have calculated statistical textural features for images that i want to classify. i have tried 'nprtool' of 'nnstart' of matlab to classify their standard dataset. after running this toolbox we get final result analysis for given dataset like confusion matrix, ROC etc. but if i want to know out of 100 dataset members which dataset member correctly classify and which is not,,, how i can get these details using this tool?
You probably just have to design the output layer carefully.
If you want a binary output the output layer should have only a single neuron which will either fire 0 or 1 (since you want to distinguish only between two classes). This is provided that your network is so designed that there is clear differences between the two classes of your input data.
I know this is a general answer but I hope this helps!
I'm agree , You have to consider the 100 outputs corresponding to the 100 members. If the outputs are binary values, they can directly be interpreted as the class of the considered data. If the outputs are real between 0 and 1, they can be interpreted as the probability of the given data to belong to a particular class.
have differentes ways, for example maybe the most common way is use a confusion matrix, but you need consider establish a treshold in the answer of NN, and then use Matrix Confussion.
Another option is used a graph to compare your test data and NN answers.
Use sim (simulation) function to find output of the given input. Use object created like net or net1 used when network was created, as input argument of this function.
For ex:
y=sim(net,p)
p is input for which output is required and y gives you output for each input
hi shubhangi thanks for reply... but i can see content of 'y' as a matrix row x col, where row indicate neurons and column indicate assignment of input vector to neuron.. i want to know how we can find out that "how much these neurons are closer to each other" as closeness of neurons help us to find number of clusters.
- Badges
- Science topic