Hi Martin or other experts,
I am confused about the meaning of the values of the ‘res_signed_decision_values.nii’.
I read in a previous post : “This will be equivalent to a Haxby-style correlation-based classification analysis and will provide the difference in Fisher z-transformed correlation coefficient between patterns of the same condition and patterns of different conditions.”
But the comment in the function “res_signed_decision_values.m” says:
% Calculate signed decision values. This function outputs decision values
% of the classifier with negative sign for incorrect predictions and
% positive sign for correct predictions. The output can be understood as
% accuracy weighted by decision values, with an expected value of 0. In
% other words, it uses the size of the decision value as evidence weight.
% The more obviously correct samples will receive a higher weight than less
% obviously correct samples. This can be useful also when there are only
% very few test samples available to get a more continuous results measure.
I don’t understand what is the incorrect or correct prediction in the case of difference of pattern-correlation ?