Partially fixed bayesian additive regression trees

Figures & data

Figure 1. The frequency of each variable used in the BART model. $X_1$ is an important variable. $X_6,\dots ,X_{10}$ are irrelevant variables.

Figure 2. Toy example for PFBART.

Figure 3. Boxplots of the RMSE values for each method across the 100 data sets.

Table 1. Settings for hyperparameter.

	ChangePrior	Prune
SET1	False	True
SET2	False	False
SET3	True	True
SET4	True	False

Table 2. UCI data sets information.

Data set name	Size	Covariate
Abalone	4170	8
Airfoil Self Noise	1500	5
Auto MPG	390	8
Bike Rental	730	10
Concrete Compressive Strength	1030	8
Energy Efficiency	760	8
Forest Fire	510	12
QSAR Aquatic Toxicity	540	8
QSAR Fish Toxicity	900	6
Real Estate Valuation	410	7
Strike	620	6
Tecator	240	13
Wine Quality	4890	11
Yacht Hydrodynamics	300	6

Figure 4. Relative RMSE for every covariate in UCI data sets.

Table 3. Hyper parameter combinations.

	ChangePrior	Prune	Swap
SET1	True	True	False
SET2	True	True	True
SET3	False	False	False
SET4	False	False	True
SET5	True	False	False
SET6	True	False	True
SET7	False	True	False
SET8	False	True	True

Figure 5. Relative RMSE for PFBART with Beijing house price data.