Adaptive Experimental Design and Counterfactual Inference

2025-04-30 0 0 707.41KB 5 页 10玖币

TANNER FIEZ, SERGIO GAMEZ, ARICK CHEN, HOUSSAM NASSIF, and LALIT JAIN∗,Amazon

Adaptive experimental design methods are increasingly being used in industry as a tool to boost testing throughput or reduce

experimentation cost relative to traditional A/B/N testing methods. This paper shares lessons learned regarding the challenges and

pitfalls of naively using adaptive experimentation systems in industrial settings where non-stationarity is prevalent, while also

providing perspectives on the proper objectives and system specications in these settings. We developed an adaptive experimental

design framework for counterfactual inference based on these experiences, and tested it in a commercial environment.

ACM Reference Format:

Tanner Fiez, Sergio Gamez, Arick Chen, Houssam Nassif, and Lalit Jain. 2022. Adaptive Experimental Design and Counterfactual

Inference. In Proceedings of Recommender Systems Conference, CONSEQUENCES Workshop (RecSys’22 CONSEQUENCES Workshop).

ACM, New York, NY, USA, 5 pages. https://doi.org/XXXXXXX.XXXXXXX

1 INTRODUCTION

A/B/N testing is a classic and ubiquitous form of experimentation that has a proven track record of driving key

performance indicators within industry [

]. Yet, experimenters are steadily shifting toward Adaptive Experimental

Design (AED) methods with the goal of increasing testing throughput or reducing the cost of experimentation. AED

promises to use a fraction of the impressions that traditional A/B/N tests require to yield high condence inferences

or to directly drive business impact. In this paper, we share lessons learned regarding the challenges and pitfalls of

naively using adaptive experimentation systems in industrial settings where non-stationarity is the norm rather than

the exception. Moreover, we provide perspectives on the proper objectives and system specications in these settings.

This culminates in a high level presentation of an AED framework for counterfactual inference. To provide a robust and

exible tool for experimenters with performance certicates at minimal cost, our methodology combines cumulative

gain estimators, always-valid condence intervals, and an elimination algorithm.

2 A CASE STUDY

Imagine a setting where on a retailer web page, a marketer has been running a message

𝐴

for the last year and now

wants to test whether message

𝐵

beats

𝐴

. At the start of the experiment the messages are initialized with a default

prior distribution, and then at each round a Thompson sampling bandit dynamically allocates trac to each treatment,

playing each message according to the posterior probability of its mean being the highest [

]. After day 8, the algorithm

directs most trac to message

𝐴

(see Figure 1). On day 14, the experimenter needs to decide whether

𝐴

has actually

beaten

𝐵

. They conduct a paired t-test which, somewhat surprisingly, does not produce a signicant

𝑝

-value. As the

bandit shifted all trac to message

𝐴

, not enough trac was directed to message

𝐵

, diminishing the power of the test.

∗Also with University of Washington.

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not

made or distributed for prot or commercial advantage and that copies bear this notice and the full citation on the rst page. Copyrights for components

of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to

redistribute to lists, requires prior specic permission and/or a fee. Request permissions from permissions@acm.org.

Manuscript submitted to ACM

arXiv:2210.14369v1 [cs.LG] 25 Oct 2022

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

AdaptiveExperimentalDesignandCounterfactualInferenceTANNERFIEZ,SERGIOGAMEZ,ARICKCHEN,HOUSSAMNASSIF,andLALITJAIN∗,AmazonAdaptiveexperimentaldesignmethodsareincreasinglybeingusedinindustryasatooltoboosttestingthroughputorreduceexperimentationcostrelativetotraditionalA/B/Ntestingmethods.Thispapershares...

展开>> 收起<<

Adaptive Experimental Design and Counterfactual Inference.pdf

共5页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Adaptive Experimental Design and Counterfactual Inference

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: