int_conformal_quantile {probably} | R Documentation |
Prediction intervals via conformal inference and quantile regression
Description
Nonparametric prediction intervals can be computed for fitted regression workflow objects using the split conformal inference method described by Romano et al (2019). To compute quantiles, this function uses Quantile Random Forests instead of classic quantile regression.
Usage
int_conformal_quantile(object, ...)
## S3 method for class 'workflow'
int_conformal_quantile(object, train_data, cal_data, level = 0.95, ...)
Arguments
object |
A fitted |
... |
Options to pass to |
train_data , cal_data |
Data frames with the predictor and outcome data.
|
level |
The confidence level for the intervals. |
Details
Note that the significance level should be specified in this function
(instead of the predict()
method).
cal_data
should be large enough to get a good estimates of a extreme
quantile (e.g., the 95th for 95% interval) and should not include rows that
were in the original training set.
Note that the because of the method used to construct the interval, it is possible that the prediction intervals will not include the predicted value.
Value
An object of class "int_conformal_quantile"
containing the
information to create intervals (which includes object
).
The predict()
method is used to produce the intervals.
References
Romano, Yaniv, Evan Patterson, and Emmanuel Candes. "Conformalized quantile regression." Advances in neural information processing systems 32 (2019).
See Also
predict.int_conformal_quantile()
Examples
library(workflows)
library(dplyr)
library(parsnip)
library(rsample)
library(tune)
library(modeldata)
set.seed(2)
sim_train <- sim_regression(500)
sim_cal <- sim_regression(200)
sim_new <- sim_regression(5) %>% select(-outcome)
# We'll use a neural network model
mlp_spec <-
mlp(hidden_units = 5, penalty = 0.01) %>%
set_mode("regression")
mlp_wflow <-
workflow() %>%
add_model(mlp_spec) %>%
add_formula(outcome ~ .)
mlp_fit <- fit(mlp_wflow, data = sim_train)
mlp_int <- int_conformal_quantile(mlp_fit, sim_train, sim_cal,
level = 0.90
)
mlp_int
predict(mlp_int, sim_new)