personalised_models.Rnw

Using our approach it is possible to compute personalised models for each
individual. This is relevant to predict the treatment effect of a new patient.
This is an easy rule to decide if the treatment should be given to a person or
not.
\begin{itemize}
	\item Compute model-based forest.

	\item Obtain prediction weights for a new observation. The weights are
a measure of similarity between the new observation and every observation in
the training set. The weight vector hence has length $N$. Each weight is given
as the number of times the new observation is sorted into the same leaf node as
each of the training observations. 
<<eval = FALSE>>=
weights <- predict(object, type = "weights", newdata = newdata, OOB = FALSE)
@

	\item Compute base model using the weighted training set.
<<eval = FALSE>>=
model <- lm(y ~ a, weights = weights, data = data)
@

\end{itemize}