modelProfile: Profile a Fitted Uplift Model

In uplift: Uplift Modeling

Description

This function can be used to profile a fitted uplift model. Given a vector of scores (uplift predictions), it computes basic summary statistics for each predictor by score quantile.

Usage

modelProfile(formula, data, groups = 10, 
group_label = c("I", "D"), digits_numeric = 1, digits_factor = 4,
exclude_na = FALSE, LaTex = FALSE)

Arguments

formula	a formula expression of the form score ~ predictors, where the LHS of the model formula should include the predictions from a fitted model.
data	a data.frame in which to interpret the variables named in the formula.
groups	number of groups of equal observations in which to partition the data set to show results. The default value is 10 (deciles). Other possible values are 5 and 20.
group_label	possible values are "I" or "D", for group number labels which are increasing or decreasing with the model score, respectively.
digits_numeric	number of digits to show for numeric predictors.
digits_factor	number of digits to show for factor predictors.
exclude_na	should the results exclude observations with missing values in any of the variables named in the formula?
LaTex	should the function output LaTex code?

Details

This function ranks the variable supplied in the LHS of the model formula and classifies it into groups of equal number of observations. It subsequently calls the function tabular from the tables package to compute the average of each numeric predictor and the distribution of each factor within each group.

Value

An object of S3 class tabular. See help("tabular") in the tables package for details.

Author(s)

Leo Guelman <[e-mail]>

Examples

library(uplift)

### Simulate data
set.seed(12345)
dd <- sim_pte(n = 1000, p = 5, rho = 0, sigma =  sqrt(2), beta.den = 4)
dd$treat <- ifelse(dd$treat == 1, 1, 0) # required coding for upliftRF

### Fit upliftRF model
fit1 <- upliftRF(y ~ X1 + X2 + X3 + X4 + X5 + trt(treat),
                 data = dd, 
                 mtry = 3,
                 ntree = 50, 
                 split_method = "KL",
                 minsplit = 100, 
                 verbose = TRUE)

### Fitted values on train data
pred <- predict(fit1, dd)

### Compute uplift predictions
uplift_pred <- pred[, 1] - pred[, 2]

### Put together data, predictions and add some dummy factors for illustration only
dd2 <- data.frame(dd, uplift_pred, F1 = gl(2, 50, labels = c("A", "B")), 
                  F2 = gl(4, 25, labels = c("a", "b", "c", "d")))
                     
### Profile data based on fitted model
modelProfile(uplift_pred ~ X1 + X2 + X3 + F1 + F2, 
             data = dd2, 
             groups = 10,
             group_label = "D", 
             digits_numeric = 2,
             digits_factor = 4,
             exclude_na = FALSE, 
             LaTex = FALSE)

uplift documentation built on May 2, 2019, 9:32 a.m.