Getting scores in Naive Bayes and Support Vector Machine

I am using Naive Bayes and Support Vector Machine in a dataset with a binary dependent variable. I am using MatlabR2015a. I would like to get the scores when I use ScoreTransform equal to none and ScoreTransform with logit. I did some research on the subject, but I did not get results. How can I access to the scores?

Thanks in advance!

Elizabeth Vieira

Answer 1

[Brendan Hamm]

The scores are only the result from using the predict method of the aforementioned classes. They are the second output referred to as posterior.

% Load data
load fisheriris
X = meas(:,3:4);
Y = categorical(species); % 3 classes
% Remove versicolor so we have binary decision
idx = Y == 'versicolor';
X = X(~idx,:);
Y = Y(~idx,:);
Y = removecats(Y,'versicolor');
% Create a Binary Naive Bayes model w/ and w/out the score transform:
Mdl1 = fitcnb(X,Y);
Mdl2 = fitcnb(X,Y,'ScoreTransform','logit');

[grp1,score1,cost1] = predict(Mdl1,X);
[grp2,score2,cost2] = predict(Mdl2,X);

% Notice score1 has probabilities, score2 has been transformed using the logit function:
% logit(x) = 1/(1+exp(-x))
% The inverse logit is:
% ilogit(x) = -log(1/x-1);
score2Trans = -log(1./score2-1);
% Are these the same?
any(abs(score1 - score2Trans) > eps)
% voila score2Trans is within machine precision of score1.

[Elizabeth Vieira]

Dear Brendam, many thanks for you response.