policytree

policytree: Policy learning via doubly robust empirical welfare maximization over trees - Published in JOSS (2020)

https://github.com/grf-labs/policytree

Science Score: 93.0%

This score indicates how likely this project is to be science-related based on various indicators:

○
CITATION.cff file
✓
codemeta.json file
Found codemeta.json file
✓
.zenodo.json file
Found .zenodo.json file
✓
DOI references
Found 6 DOI reference(s) in README and JOSS metadata
✓
Academic publication links
Links to: arxiv.org, wiley.com, joss.theoj.org
○
Committers with academic emails
○
Institutional organization owner
✓
JOSS paper metadata
Published in Journal of Open Source Software

Keywords

causal-inference policy-learning

Last synced: 6 months ago · JSON representation

Repository

Policy learning via doubly robust empirical welfare maximization over trees

Basic Info

Host: GitHub
Owner: grf-labs
License: mit
Language: R
Default Branch: master
Homepage: https://grf-labs.github.io/policytree/
Size: 1.52 MB

Statistics

Stars: 83
Watchers: 9
Forks: 15
Open Issues: 14
Releases: 0

Topics

causal-inference policy-learning

Created about 6 years ago · Last pushed 7 months ago

Metadata Files

Readme License

policytree

A package for learning simple rule-based policies, where the rule takes the form of a shallow decision tree. Applications include settings which require interpretable predictions, such as for example a medical treatment prescription. This package uses doubly robust reward estimates from grf to find a shallow, but globally optimal decision tree.

Some helpful links for getting started:

The R package documentation contains usage examples and method references.
For community questions and answers around usage, see the GitHub issues page.
The packages fastpolicytree and sparsepolicytree implement modified solvers that may offer improved performance on larger datasets.

Installation

The latest release of the package can be installed through CRAN:

R install.packages("policytree")

To install the latest development version from source:

R devtools::install_github("grf-labs/policytree", subdir = "r-package/policytree")

Installing from source requires a C++ 11 compiler (on Windows Rtools is required as well) together with the R packages Rcpp and BH.

Multi-action policy learning example

```r library(policytree) n <- 250 p <- 10 X <- matrix(rnorm(n * p), n, p) W <- as.factor(sample(c("A", "B", "C"), n, replace = TRUE)) Y <- X[, 1] + X[, 2] * (W == "B") + X[, 3] * (W == "C") + runif(n) multi.forest <- grf::multiarmcausal_forest(X, Y, W)

Compute doubly robust reward estimates.

Gamma.matrix <- doublerobustscores(multi.forest) head(Gamma.matrix)

A B C

1 -0.002612209 -0.1438422 -0.04243015

2 0.417066177 0.4212708 1.04000173

3 2.020414370 0.3963890 1.33038496

4 1.193587749 1.7862142 -0.05668051

5 0.808323778 0.5017521 1.52094053

6 -0.045844471 -0.1460745 -1.56055025

Fit a depth 2 tree on a random training subset.

train <- sample(1:n, 200) opt.tree <- policy_tree(X[train, ], Gamma.matrix[train, ], depth = 2) opt.tree

policy_tree object

Tree depth: 2

Actions: 1: A 2: B 3: C

Variable splits:

(1) splitvariable: X3 splitvalue: 0.368037

(2) splitvariable: X2 splitvalue: -0.098143

(4) * action: 1

(5) * action: 2

(3) splitvariable: X2 splitvalue: 1.25697

(6) * action: 3

(7) * action: 2

Predict treatment on held out data

head(predict(opt.tree, X[-train, ]))

> [1] 2 3 1 2 3 3

```

Details

policy_tree(): fits a depth k tree by exhaustive search (Nxp features on Nxd actions). The optimal tree maximizes the sum of rewards: let $\Gammai \in \mathbb R^d$ be a vector of unit-specific rewards for each action 1 to $d$ and $\pi(Xi) \in \{1, ..., d\}$ a mapping from covariates $Xi$ to action. `policytree` solves the following:

$$ \pi^* = argmax{\pi \in \Pi} \left[\frac{1}{n} \sum{i=1}^{n} \Gammai(\pi(Xi)) \right], $$

where $\Pi$ is the class of depth-k decision trees. (hybrid_policy_tree() employs a mix between a optimal/greedy approach and can be used to fit deeper trees).

double_robust_scores(): computes doubly robust reward estimates for a subset of grf forest types.

Contributing

Contributions are welcome, please consult the development guide for details.

Funding

Development of policytree is supported by the National Science Foundation, the Sloan Foundation, the Office of Naval Research (Grant N00014-17-1-2131) and Schmidt Futures.

References

Susan Athey and Stefan Wager. Policy Learning With Observational Data. Econometrica 89.1 (2021): 133-161. [paper, arxiv]

Toru Kitagawa and Aleksey Tetenov. Who Should be Treated? Empirical Welfare Maximization Methods for Treatment Choice. Econometrica 86.2 (2018): 591-616. [paper]

Erik Sverdrup, Ayush Kanodia, Zhengyuan Zhou, Susan Athey, and Stefan Wager. policytree: Policy learning via doubly robust empirical welfare maximization over trees. Journal of Open Source Software, 5(50), 2020. [paper]

Zhengyuan Zhou, Susan Athey, and Stefan Wager. Offline Multi-Action Policy Learning: Generalization and Optimization. Operations Research 71.1 (2023). [paper, arxiv]

Owner

Name: GRF Labs
Login: grf-labs
Kind: organization
Location: Stanford University

Website: https://grf-labs.github.io/grf/
Repositories: 4
Profile: https://github.com/grf-labs

JOSS Publication

policytree: Policy learning via doubly robust empirical welfare maximization over trees

Published

June 22, 2020

DOI

10.21105/joss.02232

Volume 5, Issue 50, Page 2232

Authors

Erik Sverdrup

Stanford Graduate School of Business

Ayush Kanodia
Stanford Graduate School of Business

Zhengyuan Zhou
NYU Stern

Susan Athey
Stanford Graduate School of Business

Stefan Wager
Stanford Graduate School of Business

Editor

Arfon Smith

GitHub Events

Total

Issues event: 4
Watch event: 7
Delete event: 1
Issue comment event: 4
Push event: 1

Last Year

Issues event: 4
Watch event: 7
Delete event: 1
Issue comment event: 4
Push event: 1

Committers

Last synced: 7 months ago

All Time

Total Commits: 169
Total Committers: 3
Avg Commits per committer: 56.333
Development Distribution Score (DDS): 0.047

Past Year

Commits: 1
Committers: 1
Avg Commits per committer: 1.0
Development Distribution Score (DDS): 0.0

Top Committers

Name	Email	Commits
Erik Sverdrup	e**p@g**m	161
Stefan Wager	s****r	4
Ayush Kanodia	k**h@g**m	4

Issues and Pull Requests

Last synced: 6 months ago

All Time

Total issues: 25
Total pull requests: 86
Average time to close issues: 4 months
Average time to close pull requests: 1 day
Total issue authors: 13
Total pull request authors: 2
Average comments per issue: 4.0
Average comments per pull request: 0.1
Merged pull requests: 80
Bot issues: 0
Bot pull requests: 0

Past Year

Issues: 3
Pull requests: 0
Average time to close issues: 6 days
Average time to close pull requests: N/A
Issue authors: 3
Pull request authors: 0
Average comments per issue: 1.67
Average comments per pull request: 0
Merged pull requests: 0
Bot issues: 0
Bot pull requests: 0

View more stats

Top Authors

Issue Authors

njawadekar (6)
erikcs (6)
hhsievertsen (3)
gabriel-ruiz (1)
jkortner (1)
aczura2024 (1)
shafayetShafee (1)
MCKnaus (1)
beniaminogreen (1)
jarkki (1)
ZhiyLiu (1)
kmrnykgbn (1)
if-sudo (1)

Pull Request Authors

erikcs (87)
kanodiaayush (2)

Top Labels

Issue Labels

question (12) enhancement (3) bug (2) breaking (1) documentation (1) code quality (1)

Pull Request Labels

bug (2) documentation (1)

Packages

Total packages: 2
Total downloads:
- cran 672 last-month
Total docker downloads: 38

Total dependent packages: 1
(may contain duplicates)
Total dependent repositories: 2
(may contain duplicates)
Total versions: 18
Total maintainers: 1

cran.r-project.org: policytree

Policy Learning via Doubly Robust Empirical Welfare Maximization over Trees

Homepage: https://github.com/grf-labs/policytree
Documentation: http://cran.r-project.org/web/packages/policytree/policytree.pdf
License: MIT + file LICENSE
Latest release: 1.2.3
published over 1 year ago

Versions: 14
Dependent Packages: 1
Dependent Repositories: 2
Downloads: 672 Last month
Docker Downloads: 38

Rankings

Forks count: 5.1%

Stargazers count: 5.1%

Average: 13.6%

Dependent packages count: 18.1%

Dependent repos count: 19.3%

Downloads: 20.3%

Maintainers (1)

erikcs@stanford.edu

Last synced: 6 months ago

conda-forge.org: r-policytree

Homepage: https://github.com/grf-labs/policytree
License: GPL-3.0-only
Latest release: 1.2.0
published almost 4 years ago

Versions: 4
Dependent Packages: 0
Dependent Repositories: 0

Rankings

Dependent repos count: 34.0%

Stargazers count: 34.8%

Average: 40.2%

Forks count: 40.9%

Dependent packages count: 51.2%

Last synced: 6 months ago

Dependencies

r-package/policytree/DESCRIPTION cran

R >= 3.5.0 depends
Rcpp * imports
grf >= 2.0.0 imports
DiagrammeR * suggests
testthat >= 3.0.4 suggests

policytree

Science Score: 93.0%

Keywords

Repository

Basic Info

Statistics

Topics

Metadata Files

README.md

policytree

Installation

Multi-action policy learning example

Compute doubly robust reward estimates.

A B C

1 -0.002612209 -0.1438422 -0.04243015

2 0.417066177 0.4212708 1.04000173

3 2.020414370 0.3963890 1.33038496

4 1.193587749 1.7862142 -0.05668051

5 0.808323778 0.5017521 1.52094053

6 -0.045844471 -0.1460745 -1.56055025

Fit a depth 2 tree on a random training subset.

policy_tree object

Tree depth: 2

Actions: 1: A 2: B 3: C

Variable splits:

(1) splitvariable: X3 splitvalue: 0.368037

(2) splitvariable: X2 splitvalue: -0.098143

(4) * action: 1

(5) * action: 2

(3) splitvariable: X2 splitvalue: 1.25697

(6) * action: 3

(7) * action: 2

Predict treatment on held out data

> [1] 2 3 1 2 3 3

Details

Contributing

Funding

References

Owner

JOSS Publication

policytree: Policy learning via doubly robust empirical welfare maximization over trees

Authors

Editor

Tags

GitHub Events

Total

Last Year

Committers

All Time

Past Year

Top Committers

Issues and Pull Requests

All Time

Past Year

Top Authors

Issue Authors

Pull Request Authors

Top Labels

Issue Labels

Pull Request Labels

Packages

cran.r-project.org: policytree

Rankings

Maintainers (1)

conda-forge.org: r-policytree

Rankings

Dependencies