gpe_trees: Learner Functions Generators for gpe

In pre: Prediction Rule Ensembles

Learner Functions Generators for gpe

Description

Functions to get "learner" functions for gpe.

Usage

gpe_trees(
  ...,
  remove_duplicates_complements = TRUE,
  mtry = Inf,
  ntrees = 500,
  maxdepth = 3L,
  learnrate = 0.01,
  parallel = FALSE,
  use_grad = TRUE,
  tree.control = ctree_control(mtry = mtry, maxdepth = maxdepth)
)

gpe_linear(..., winsfrac = 0.025, normalize = TRUE)

gpe_earth(
  ...,
  degree = 3,
  nk = 8,
  normalize = TRUE,
  ntrain = 100,
  learnrate = 0.1,
  cor_thresh = 0.99
)

Arguments

...	Currently not used.
remove_duplicates_complements	TRUE. Should rules with complementary or duplicate support be removed?
mtry	Number of input variables randomly sampled as candidates at each node for random forest like algorithms. The argument is passed to the tree methods in the partykit package.
ntrees	Number of trees to fit. Will not have an effect if tree.control is used.
maxdepth	Maximum depth of trees. Will not have an effect if tree.control is used.
learnrate	Learning rate for methods. Corresponds to the \nu parameter in Friedman & Popescu (2008).
parallel	TRUE. Should basis functions be found in parallel?
use_grad	TRUE. Should binary outcomes use gradient boosting with regression trees when learnrate > 0? That is, use ctree instead of glmtree as in Friedman (2001) with a second order Taylor expansion instead of first order as in Chen and Guestrin (2016).
tree.control	ctree_control with options for the ctree function.
winsfrac	Quantile to winsorize linear terms. The value should be in [0,0.5)
normalize	TRUE. Should value be scaled by .4 times the inverse standard deviation? If TRUE, gives linear terms the same influence as a typical rule.
degree	Maximum degree of interactions in earth model.
nk	Maximum number of basis functions in earth model.
ntrain	Number of models to fit.
cor_thresh	A threshold on the pairwise correlation for removal of basis functions. This is similar to remove_duplicates_complements. One of the basis functions in pairs where the correlation exceeds the threshold is excluded. NULL implies no exclusion. Setting a value closer to zero will decrease the time needed to fit the final model.

Details

gpe_trees provides learners for tree method. Either ctree or glmtree from the partykit package will be used.

gpe_linear provides linear terms for the gpe.

gpe_earth provides basis functions where each factor is a hinge function. The model is estimated with earth.

Value

A function that has formal arguments formula, data, weights, sample_func, verbose, family, .... The function returns a vector with character where each element is a term for the final formula in the call to cv.glmnet

References

Hothorn, T., & Zeileis, A. (2015). partykit: A modular toolkit for recursive partytioning in R. Journal of Machine Learning Research, 16, 3905-3909.

Friedman, J. H. (1991). Multivariate adaptive regression splines. The Annals Statistics, 19(1), 1-67.

Friedman, J. H. (2001). Greedy function approximation: a gradient boosting machine. The Annals of Applied Statistics, 29(5), 1189-1232.

Friedman, J. H. (1993). Fast MARS. Dept. of Statistics Technical Report No. 110, Stanford University.

Friedman, J. H., & Popescu, B. E. (2008). Predictive learning via rule ensembles. The Annals of Applied Statistics, 2(3), 916-954.

Chen T., & Guestrin C. (2016). Xgboost: A scalable tree boosting system. Proceedings of the 22Nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2016.

See Also

gpe, rTerm, lTerm, eTerm

pre documentation built on May 29, 2024, 5:10 a.m.