Ordinal regression

About: Ordinal regression is a research topic. Over the lifetime, 1879 publications have been published within this topic receiving 65431 citations.

Ordinal Consistent Partial Least Squares

Abstract: In this chapter, we present a new variance-based estimator called ordinal consistent partial least squares (OrdPLSc). It is a promising combination of consistent partial least squares (PLSc) and ordinal partial least squares (OrdPLS), respectively, which is capable to deal in structural equation models with common factors, composites, and ordinal categorical indicators. Besides providing the theoretical background of OrdPLSc, we present three approaches to obtain constructs scores from OrdPLS and OrdPLSc, which can be used, e.g., in importance-performance matrix analysis. Finally, we show its behavior on an empirical example and provide a practical guidance for the assessment of SEMs with ordinal categorical indicators in the context of OrdPLSc.

Inverse Reinforcement Learning with Multiple Ranked Experts.

Abstract: We consider the problem of learning to behave optimally in a Markov Decision Process when a reward function is not specified, but instead we have access to a set of demonstrators of varying performance. We assume the demonstrators are classified into one of k ranks, and use ideas from ordinal regression to find a reward function that maximizes the margin between the different ranks. This approach is based on the idea that agents should not only learn how to behave from experts, but also how not to behave from non-experts. We show there are MDPs where important differences in the reward function would be hidden from existing algorithms by the behaviour of the expert. Our method is particularly useful for problems where we have access to a large set of agent behaviours with varying degrees of expertise (such as through GPS or cellphones). We highlight the differences between our approach and existing methods using a simple grid domain and demonstrate its efficacy on determining passenger-finding strategies for taxi drivers, using a large dataset of GPS trajectories.

Ordinal Pattern Analysis

Abstract: Our task in writing this chapter is to outline a strategy for analyzing data that we believe to be better suited to most psychological research than the most widely used statistical techniques (e.g., t, F, and chi square tests, product moment correlation, regression, covariance, discriminant, and factor analyses). We call the strategy Ordinal Pattern Analysis (OPA), and derive it from a small set of first principles that deviate somewhat from those on which most classical statistical models are based. First, we assume that the goal of statistical practice is to aid in detecting and analyzing patterns in data rather than to aid in making decisions about populations given samples of data. Second, we assume that a statistic must be useful in analyzing data generated by individual subjects as well as data based on aggregations of subjects. Third, we assume that most predictions and observations in psychological research possess no more than ordinal scale properties, and that statistics employed to assess the fit between predictions and observations must be derived on the basis of this constraint.

Cautious ordinal classification by binary decomposition

Abstract: We study the problem of performing cautious inferences for an ordinal classification (a.k.a. ordinal regression) task, that is when the possible classes are totally ordered. By cautious inference, we mean that we may produce partial predictions when available information is insufficient to provide reliable precise ones. We do so by estimating probabilistic bounds instead of precise ones. These bounds induce a (convex) set of possible probabilistic models, from which we perform inferences. As the estimates or predictions for such models are usually computationally harder to obtain than for precise ones, we study the extension of two binary decomposition strategies that remain easy to obtain and computationally efficient to manipulate when shifting from precise to bounded estimates. We demonstrate the possible usefulness of such a cautious attitude on tests performed on benchmark data sets.