A Field Guide for Pacing Budget and ROS Constraints

Authors: Santiago R. Balseiro, Kshipra Bhawalkar, Zhe Feng, Haihao Lu, Vahab Mirrokni, Balasubramanian Sivan, Di Wang

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our theoretical findings empirically by showing that the min-pacing algorithm performs almost as well as the canonical dual-based algorithm on a semisynthetic dataset that was generated from a large online advertising platform s auction data. Empirical evaluation. Section 4 explains in detail our evaluation methodology, including how we construct our semi-synthetic dataset, how we obtain the different quantities in our optimization formulation (1) based on real auction data.
Researcher Affiliation Collaboration 1Google Research, USA 2Columbia University, NYC, USA. Correspondence to: Di Wang <wadi@google.com>.
Pseudocode Yes Algorithm 1 Dual-Optimal Pacing
Open Source Code No The paper does not provide an explicit statement or link to open-source code for the described methodology.
Open Datasets No For confidentiality and advertiser privacy reasons, we use a semi-synthetic dataset based on actual advertising auctions from an online platform... Our dataset includes 105 randomly selected campaigns... The paper does not provide a link or specific access information for this semi-synthetic dataset.
Dataset Splits No The paper mentions dividing the day into 10-minute periods (T=144) and simulating algorithms 10 times, but it does not specify explicit training, validation, or test dataset splits.
Hardware Specification No The paper does not provide any specific hardware details (e.g., CPU, GPU models, memory) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies or version numbers (e.g., Python 3.8, PyTorch 1.9) needed to replicate the experiment.
Experiment Setup Yes For each campaign, we set the budget constraint (i.e. ρT in (1)) using its actual daily budget B. We divide the day into 10-minute periods and use T = 144. For each campaign, we simulate an algorithm 10 times... For each algorithm, we do a grid search over the step-sizes used in the dual variables updates.