pimp.import {LogicForest} | R Documentation |
Predictor Importance - Variables and Interactions
Description
Measures the predictor importance for each predictor and interaction importance for each iteration.
Usage
pimp.import(fit, data, testdata, BSpred, pred, Xs, mtype)
Arguments
fit |
Fit information including outcome/model type, input data, logic tree, etc. |
data |
In-bag sample (i.e., training data). |
testdata |
Out-of-bag sample (i.e., test data). |
BSpred |
Number of Xs in the interactions (includes not-ed variables). |
pred |
Matrix of predicted values. |
Xs |
Matrix or data frame of zeros and ones for all predictor variables. |
mtype |
Model type: |
Details
This function is called to calculate importance measures for each bootstrapped sample. Importance measures are calculated as differences between the original out-of-bag sample and a permuted out-of-bag sample. Model fit for both samples is evaluated using:
Concordance for classification,
Mean squared error for linear regression,
Harrell's C-index for survival regression.
Value
A list with:
- single.vimp
Vector of predictor importance estimates. One estimate per predictor used in the tree of that sample.
- pimp.vimp
Vector of interaction importance estimates. One estimate per interaction detected in that sample.
- Ipimat
Matrix updated for each sample. Contains all predictors (and their NOT-ed versions if used) for each interaction.
- vec.Xvars
Vector of predictors used in the tree of that sample.
- Xids
Vector of predictor IDs used in the tree of that sample.
Author(s)
Bethany J. Wolf wolfb@musc.edu
J. Madison Hyer madison.hyer@osumc.edu
References
Wolf BJ, Hill EG, Slate EH. Logic Forest: an ensemble classifier for discovering logical combinations of binary markers. Bioinformatics. 2010;26(17):2183-2189. doi:10.1093/bioinformatics/btq354