function Y_hat = kernel_calibrate( X_new, X, Y, h, MAX_NUM_SAMPLES )
% Given a set of training data decision values (X) and their labels (Y),
% use kernel regression to estimate outcome probabilities (Y_hat) for a
% new set of data decision values (X_new)
%
% Inputs:
% X_new - new data (1-dimensional) points, usually classifier outputs
% X - training data points
% Y - training data labels (0/1)
% h - fixed bandwidth parameter
% MAX_NUM_SAMPLES - maximum number of samples to use (speeds up the
% kernel regression) (optional, default 5000)
%
% Y_hat - estimated outcome probabilities for each of the new points
% make sure vectors are column vectors
X_new = enforce_col_vector(X_new);
X = enforce_col_vector(X);
Y = enforce_col_vector(Y);
% remove NaNs rows from X/Y
M = [X Y];
M = remove_row_mvs(M);
X = M(:,1);
Y = M(:,2);
% make sure negative labels aren't -1
Y(Y < 0) = 0;
Y_hat = kernel_regress(X_new, X, Y, h);
end
function [A, rows_with_mvs] = remove_row_mvs(A)
missing = sum(isnan(A),2);
rows_with_mvs = find(missing > 0);
A(rows_with_mvs,:) = [];
end
function v = enforce_col_vector( v )
if size(v,1) < size(v,2)
v = v';
end
end