Qiu and See file AUTHORS for details

@Rosenbaum1983

e Figure #

Improved Calibration: Probability machines often provide better-calibrated predictions by capturing complex data relationships. Flexible Modelling: Unlike parametric methods like logistic regression, probability machines don’t rely on assumptions of additivity or linearity, allowing them to model intricate relationships that parametric models miss. Efficient Feature Selection: These machines automatically select features, making them ideal for high-dimensional datasets where manual selection is impractical. Handling Missing Data: Probability machines handle missing data robustly, minimizing the need for extensive data reprocessing and imputation. Simplified Data Exploration: By exploring complex data structures in a data-driven way, probability machines simplify model specification. For instance, tree-based models remain unaffected by adding squared or interaction terms, streamlining the modeling process.

Supervised mach

ty scores.

dict propensity scores

0.11

Non-farm Income Access

Figure 3.6: Comparison of Balance f

es with a depth between 1 and 5. The best tuning performance was found with shrinkage of 0.2 and 9 trees which were three splits 3 deep. As such, the tuning grid was redefined in a second iteration to use 0.1,0.15,0.2,0.25,0.3,0.35, and 0.4 with only 1000 trees with between 2 and 5 depth. The second fit, suggested a learning rate of 0.35 so the local area of 0.3,0.325,0.350,0.375, and 0.4 is searched in the final fit.

Figure 3.1: This

t”, package = “cobalt

PALCEHOLDER

Tree Visu

eq-ols-model

al Diagrams

Key Idea

usal infer

Causal inference

earners