Skip to contents

Joke

Source: vignettes/joke.Rmd
joke.Rmd

The Jester joke-ratings dataset is encoded as a contextual bandit with 10 actions (joke IDs), 100-dimensional random-projection contexts, and a rating outcome normalized to [0, 5.0].

The code assumes that you have created a KPE_DATA_DIR env variable which points to the directory where you are storing RCT_joke_data.csv (n = 48 445).

The RCT_joke_data.csv is a preprocessed RCT form of the Jester dataset [https://goldberg.berkeley.edu/jester-data/]. See details here: https://datadryad.org/dataset/doi:10.5061/dryad.bg79cnpp7

Prerequisites

This vignette assumes kpe is installed with the learner backends its call uses. It calls reward_model = "lasso", which requires the glmnet package (the "linear" contextual policy needs no extra package). See vignette("getting-started", package = "kpe") for installation and the full learner/backend table.

It also requires RCT_joke_data.csv and the KPE_DATA_DIR environment variable pointing to the directory that contains it (see above); the code chunks below are skipped when KPE_DATA_DIR is unset.

This dataset can be accessed by cloning the repository and looking in the “vignettes” directory. git clone https://github.com/StanfordAI4HI/kpe-r.git

Setup 

library(kpe)
options(future.globals.maxSize = 1000 * 1024^2)

# Processed Jester arrays (columns X0..X99, A, Y), exported once from
# RCT_joke_data.pkl to a CSV that both the R and Python vignettes read.
df <- read.csv(file.path(Sys.getenv("KPE_DATA_DIR"), "RCT_joke_data.csv"))
X <- as.matrix(df[, grep("^X", names(df))])
A <- as.integer(df$A)
Y <- as.numeric(df$Y)

pa_x <- prop.table(table(A))[as.character(A)]

Run 

result <- kpe(
  X = X, A = A, Y = Y, propensity = pa_x,
  n_shuffles        = 100,
  n_folds           = 6,
  contextual_policy = "linear",
  best_arm          = "best_arm_erm",
  reward_model      = "lasso",
  n_cores           = 5,
  seed              = 12314
)
summary(result)

The research code uses a custom learner named joker_lasso_flat; this package implements a per-arm linear regression (contextual_policy = "linear") combined with a Lasso reward model (reward_model = "lasso") as the closest in-package equivalent. The joker_* learners are not exported by this package.

Results

Running the above call on the authoritative pickle under the default leave-one-fold-out cross-fitting (n_nuisance_folds = n_folds - 1, so 5/6 of the data fits every nuisance) yields:

quantity value
psi 0.1692
std_error 0.0224
95% confidence interval [0.125, 0.213]
z-statistic 7.56
one-sided p-value 2.1e-14
policy_stability 0.273

For reference, this two-fold shared-nuisance scheme used here slightly improves over the original results report in Li and Brunskill (Science, 2026). The sign of psi and the reject-at-0.05 decision are unchanged.

Multi-arm learner choice

The Joke dataset has 10 actions. Among the built-in contextual-policy strings:

  • "linear" fits a separate OLS regression per arm and returns argmax predictions; supports any number of actions.
  • "policy_tree" wraps policytree::policy_tree; supports any number of actions.
  • "policy_forest" wraps grf::causal_forest; in this release it supports binary treatment only and raises an informative error when K > 2.

Notes

  • The rating Y is normalized to the [0, 5.0] range prior to this call; psi is on that normalized outcome scale.