Meta Dynamic Pricing: Learning Across Experiments

Bastani, Hamsa; Simchi‐Levi, David; Zhu, Ruihao

doi:10.2139/ssrn.3334629

Cited by 11 publications

(7 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our work has a close conceptual connection to work on meta-learning (see e.g., Finn et al, 2017;Baxter, 2000). As is nicely articulated by Bastani et al (2019), many companies face a large sequence of experimentation tasks, raising the question of how to effectively share information across these tasks. Consider a web company who may run thousands of A/B tests per year, giving them strong prior knowledge of the distribution of effect sizes and click through rates.…”

Section: Introductionmentioning

confidence: 60%

“…Each day can be viewed as its own instance of a bandit problem and the platform's goal is to do well on average across a large number of days. Bastani et al (2019) suggest an empirical Bayesian approach, where the prior of Thompson sampling is statistically estimated from data on previous tasks. This view of meta-learning as learning a prior distribution has long been recognized.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Policy Gradient Optimization of Thompson Sampling Policies

Min¹,

Moallemi²,

Russo³

2020

Preprint

View full text Add to dashboard Cite

We study the use of policy gradient algorithms to optimize over a class of generalized Thompson sampling policies. Our central insight is to view the posterior parameter sampled by Thompson sampling as a kind of pseudo-action. Policy gradient methods can then be tractably applied to search over a class of sampling policies, which determine a probability distribution over pseudo-actions (i.e., sampled parameters) as a function of observed data. We also propose and compare policy gradient estimators that are specialized to Bayesian bandit problems. Numerical experiments demonstrate that direct policy search on top of Thompson sampling automatically corrects for some of the algorithm's known shortcomings and offers meaningful improvements even in long horizon problems where standard Thompson sampling is extremely effective.

show abstract

Section: Introductionmentioning

confidence: 60%

Section: Introductionmentioning

confidence: 99%

Policy Gradient Optimization of Thompson Sampling Policies

Min¹,

Moallemi²,

Russo³

2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Several bandit approaches have also been explored recently. Bastani et al (2019) introduced an approach that uses Thompson sampling and a linear demand model to learn a pricing policy across multiple related products through pricing experiments. They defined a discrete set of auto loan products whereas in our approach the number of products is not fixed and new products may be defined as part of the continuous state.…”

Section: Related Workmentioning

confidence: 99%

Offline Deep Reinforcement Learning for Dynamic Pricing of Consumer Credit

Khraishi¹,

Okhrati²

2022

Preprint

View full text Add to dashboard Cite

We introduce a method for pricing consumer credit using recent advances in offline deep reinforcement learning. This approach relies on a static dataset and requires no assumptions on the functional form of demand. Using both real and synthetic data on consumer credit applications, we demonstrate that our approach using the conservative Q-Learning algorithm is capable of learning an effective personalized pricing policy without any online interaction or price experimentation.

show abstract

“…• Side Information: Thanks to historical data from past experiments, prior information about r is often available in the form of probabilistic prior (Bastani et al 2021b, Simchowitz et al 2021 or confidence intervals (Zhang et al 2020). In this case, we assume that side information about r is given as confidence intervals (as this can also be constructed with a given prior), i.e., ∀a ∈ A, r(a) ∈ [L(a), U (a)] (⊂ [0, 1]).…”

Section: Problem Formulationmentioning

confidence: 99%

Safe Data Collection for Offline and Online Policy Learning

Zhu¹,

Kveton²

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Motivated by practical needs in online experimentation and off-policy learning, we study the problem of safe optimal design, where we develop a data logging policy that efficiently explores while achieving competitive rewards with a baseline production policy. We first show, perhaps surprisingly, that a common practice of mixing the production policy with uniform exploration, despite being safe, is sub-optimal in maximizing information gain. Then we propose a safe optimal logging policy for the case when no side information about the actions' expected rewards is available. We improve upon this design by considering side information and also extend both approaches to a large number of actions with a linear reward model. We analyze how our data logging policies impact errors in off-policy learning. Finally, we empirically validate the benefit of our designs by conducting extensive experiments.

show abstract

Meta Dynamic Pricing: Learning Across Experiments

Cited by 11 publications

References 41 publications

Policy Gradient Optimization of Thompson Sampling Policies

Policy Gradient Optimization of Thompson Sampling Policies

Offline Deep Reinforcement Learning for Dynamic Pricing of Consumer Credit

Safe Data Collection for Offline and Online Policy Learning

Contact Info

Product

Resources

About