Discretizing Logged Interaction Data Biases Learning for Decision-Making
Machine Learning
2018-10-09 v1 Artificial Intelligence
Machine Learning
Systems and Control
Abstract
Time series data that are not measured at regular intervals are commonly discretized as a preprocessing step. For example, data about customer arrival times might be simplified by summing the number of arrivals within hourly intervals, which produces a discrete-time time series that is easier to model. In this abstract, we show that discretization introduces a bias that affects models trained for decision-making. We refer to this phenomenon as discretization bias, and show that we can avoid it by using continuous-time models instead.
Cite
@article{arxiv.1810.03025,
title = {Discretizing Logged Interaction Data Biases Learning for Decision-Making},
author = {Peter Schulam and Suchi Saria},
journal= {arXiv preprint arXiv:1810.03025},
year = {2018}
}
Comments
This is a standalone short paper describing a new type of bias that can arise when learning from time series data for sequential decision-making problems