log_likelihoods.Rnw

By using the personalised models one can obtain a likelihood for a random forest. Instead of using new data here the training data is used to compute the personalised models.

\begin{itemize}
	\item Obtain personalised models using out-of-bag prediction weights.
<<eval = FALSE>>=
weights <- predict(object, type = "weights", newdata = traindata, OOB = TRUE)
model <- lm(y ~ a, weights = weights, data = data)
@

	\item For each data point and the corresponding model compute the
log-likelihood contribution.
<<eval = FALSE>>=
y <- datapoint[, response]
yhat <- predict(model, newdata = datapoint)
llcontribution <- (yhat - y)^2
@

	\item Compute log-likelihood of the forest as the negative sum of the contributions.
<<eval = FALSE>>=
loglik <- -sum(llcontributions)
@

\end{itemize}