Merge pull request #3 from TRI-ML/joss-reupload

Joss reupload

joss/paper.bib

@@ -0,0 +1,35 @@
+@phdthesis{eudey1949,
+  title={\href{https://search.library.berkeley.edu/permalink/01UCS_BER/1thfj9n/alma991076184199706532}{On the Treatment of Discontinuous Random Variables}},
+  author={Eudey, Mark W},
+  year={1949},
+  school={University of California - Berkeley}
+}
+
+
+@book{lehmann_textbook,
+  title={\href{https://link.springer.com/book/10.1007/978-3-030-70578-7}{Testing Statistical Hypotheses}},
+  author={Lehmann, Erich Leo and Romano, Joseph P},
+  volume={4},
+  year={2022},
+  publisher={Springer}
+}
+
+
+@article{clopper_pearson,
+  title={\href{https://www.jstor.org/stable/2331986}{The Use of Confidence or Fiducial Limits Illustrated in the Case of the Binomial}},
+  author={Clopper, Charles J and Pearson, Egon S},
+  journal={Biometrika},
+  volume={26},
+  number={4},
+  pages={404--413},
+  year={1934},
+  publisher={JSTOR}
+}
+
+
+@article{vincent2024generalizable,
+  title={\href{https://arxiv.org/abs/2405.05439}{How Generalizable Is My Behavior Cloning Policy? A Statistical Approach to Trustworthy Performance Evaluation}},
+  author={Vincent, Joseph A and Nishimura, Haruki and Itkina, Masha and Shah, Paarth and Schwager, Mac and Kollar, Thomas},
+  journal={arXiv preprint arXiv:2405.05439},
+  year={2024}
+}

joss/paper.md

@@ -0,0 +1,58 @@
+---
+title: 'binomial_cis: A Python Package for Optimal Binomial Confidence Intervals'
+tags:
+  - Python
+  - statistics
+  - confidence intervals
+  - binomial
+authors:
+  - name: Joseph A. Vincent
+    orcid: 0000-0002-2270-7395
+    affiliation: 1
+affiliations:
+ - name: Department of Aeronautics and Astronautics, Stanford University
+   index: 1
+date: 11 June 2024
+bibliography: paper.bib
+---
+
+
+# Summary
+[binomial_cis](https://github.com/TRI-ML/binomial_cis) is a Python package for computing confidence intervals for the probability of success parameter, $p$, of a binomial distribution. The binomial distribution represents the likelihood of observing $k$ successes in $n$ trials where the probability of success for each trial is $p$. For example, $p$ may be the probability of a coin flip landing on heads, and $k$ the number of heads we observe after $n$ flips. One often does not know the value of $p$ and wishes to estimate it. A confidence interval is a set, constructed based on $k, n$, that covers the unknown parameter $p$ with some user-specified probability. The binomial_cis package computes confidence intervals that lower and/or upper bound $p$ with a user-specified probability. 
+
+
+
+# Statement of Need
+
+Constructing confidence intervals for an unknown probability success given samples of successes and failures is one of the most fundamental problems in statistical inference.
+Research into this question dates back at least to the 1930s with the work of Clopper and Pearson [@clopper_pearson].
+A foundational result for constructing binomial confidence intervals of minimal width was given by [@eudey1949] and is formalized in [@lehmann_textbook].
+We refer to these intervals as *optimal binomial confidence intervals* and they have the property of being uniformly most accurate (UMA) and uniformly most accurate unbiased (UMAU).
+Practically, these intervals can provide better inference of $p$ at small sample sizes ($n \le 50$).
+The binomial_cis package is the first open-source implementation of these optimal binomial confidence intervals.
+In addition, this package provides worst-case analysis of the tightness for the confidence intervals, a feature that is not present in other software for binomial confidence intervals.
+Practically, this feature assists the user in understanding how many samples an experiment should have in order to meet a desired level of accuracy in inferring the value of $p$.
+
+
+# Comparison to Existing Software
+There are many existing software packages for computing binomial confidence intervals.
+binomial_cis differs from the existing software by providing:
+
+1. Open-source implementations for the optimal binomial confidence intervals given by [@eudey1949] and formalized in [@lehmann_textbook].
+
+2. Functionality for worst-case analysis of the tightness for the confidence intervals, which helps guide users on selecting the sample size for their experiments.
+
+
+# Research Usage
+
+binomial_cis has been used to compute confidence intervals for the success rate of robots in simulated and real-world tasks [@vincent2024generalizable].
+
+
+
+# Acknowledgements
+
+Financial support was provided by Toyota Research Institute.
+
+
+
+# References