# Compute the acuarcy or error of the output?

1 view (last 30 days)
Salem on 21 Apr 2016
Commented: Greg Heath on 24 Apr 2016
I have two vectors Y and Yprd, each one is 1x602 double. Y contains the real data which represent the class label either one or zero. Yprd contains the prediction of the data which real numbers. Here is an example Y=[0 1 1 1 0] Yprd=[0.456 0.986 -0.008 0.987 0.0002] I would like to compute the accuracy of the model (or error) when at Yprd vector any values greater than 0.5 can be one and less than can zero.
##### 2 CommentsShowHide 1 older comment
Salem on 22 Apr 2016
I used neural network model, I built it myself as I want some specific operations.

Roger Stafford on 22 Apr 2016
Ymodel = 1*(Y>.5) + 0*(Y<=.5); % The model from the predictions (right half unnecessary)
p = sum(abs(Y-Ymodel))/size(Y,2); % Fractional error
Greg Heath on 24 Apr 2016
The usual convention for classifiers is to have c-dimensional {0,1} unit vectors for targets and nonnegative c-dimensional unit vectors for outputs
The relationship between the column vectors and the class indices are given by the functions
IND2VEC and VEC2IND
see their help and documentation.
Hope this helps.
Greg