tsp_model_builder: Build and cross-validate a TSP-based model
In prpatil/tdsm: Templated Deterministic Statistical Machines

Description Usage Arguments Details Value

This function takes training/test data and pairs generated via empirical control feature selection and builds a decision tree model. It also cross-validates to get an out-of-sample accuracy estimate

1 2	tsp_model_builder(train, train_outcome, train_covar, pairs, test, test_covar, npair, predtype)

`train`	p x n training data matrix
`train_outcome`	Outcome data of length n
`train_covar`	n x q additional covariates for training data (optional)
`pairs`	r x n matrix of TSP generated via empirical controls
`test`	p x m test data matrix, where p columns and column names match up with train
`test_covar`	m x s additional covariates for training data (necessary if train_covar specified; column names must match)
`npair`	Number of pairs desired in the model
`predtype`	Type of predictions to make - "class" if initial outcome is factor, "vector" if initial outcome is non-factor

This is a wrapper for a series of model-building steps. The main output of this function is the TSP decision tree model. We incorporate a second feature selection step (after empirical controls, done separately) that chooses from the candidate pairs. Pairs are chosen based on how much additional predictive value they provide on top of pairs already selected (and non-pair covariates, if specified). We also cross-validate this entire procedure five times to get an estimate of out-of-sample accuracy of our model.

A list contaiing the following attributes:

`tree`	The final decision tree built on training data
`p_train`	Model predictions on training data
`p_test`	Model predictions on test data
`final_names`	Pair names in final model
`pair_names`	Pair name aliases for pretty tree printing
`acc`	Out-of-sample accuracy calculated via cross-validation