/**
* Computes the transformed y variable using the procedure outlined in the following paper:
*
* @see HA Chipman, EI George, and RE McCulloch. BART: Bayesian Additive Regressive Trees. The
* Annals of Applied Statistics, 4(1): 266-298, 2010.
*/
protected void transformResponseVariable() {
super.transformResponseVariable();
y_min = StatToolbox.sample_minimum(y_orig);
y_max = StatToolbox.sample_maximum(y_orig);
y_range_sq = Math.pow(y_max - y_min, 2);
for (int i = 0; i < n; i++) {
y_trans[i] = transform_y(y_orig[i]);
}
}
/** Computes <code>hyper_sigsq_mu</code> and <code>hyper_lambda</code>. */
protected void calculateHyperparameters() {
hyper_mu_mu = 0;
hyper_sigsq_mu = Math.pow(YminAndYmaxHalfDiff / (hyper_k * Math.sqrt(num_trees)), 2);
if (sample_var_y == null) {
sample_var_y = StatToolbox.sample_variance(y_trans);
}
// calculate lambda from q
double ten_pctile_chisq_df_hyper_nu = 0;
ChiSquaredDistributionImpl chi_sq_dist = new ChiSquaredDistributionImpl(hyper_nu);
try {
ten_pctile_chisq_df_hyper_nu = chi_sq_dist.inverseCumulativeProbability(1 - hyper_q);
} catch (MathException e) {
System.err.println(
"Could not calculate inverse cum prob density for chi sq df = "
+ hyper_nu
+ " with q = "
+ hyper_q);
System.exit(0);
}
hyper_lambda = ten_pctile_chisq_df_hyper_nu / hyper_nu * sample_var_y;
}
