In many classifier systems, the classifier strength parameter serves as a predictor of future payoff and as the classifier's fitness for the genetic algorithm. We investigate a classifier system, XCS, in which each classifier maintains a prediction of expected payoff, but the classifier's fitness is given by a measure of the prediction's accuracy. The system executes the genetic algorithm in niches defined by the match sets, instead of panmictically. These aspects of XCS result in its population tending to form a complete and accurate mapping X × A → P from inputs and actions to payoff predictions. Further, XCS tends to evolve classifiers that are maximally general, subject to an accuracy criterion. Besides introducing a new direction for classifier system research, these properties of XCS make it suitable for a wide range of reinforcement learning situations where generalization over states is desirable.

This longitudinal study examined change in adolescents' daily range of emotional states between early and late adolescence. A sample of 220 youth provided reports on their daily emotions at random times during two 1-week periods. At Time 1 they were in the fifth through eighth grades; 4 years later, at Time 2, they were in the ninth through twelfth grades. Results showed that average emotional states became less positive across early adolescence, but that this downward change in average emotions ceased in grade 10. The results also showed greatest relative instability between youth in the early adolescent years--correlations over time were lower--with stability increasing in late adolescence. Lastly, the study found that adolescents' average emotions had relatively stable relations to life stress and psychological adjustment between early and late adolescence. As a whole, the findings suggest that late adolescence is associated with a slowing of the emotional changes of early adolescence.

A basic classifier system, ZCS, is presented that keeps much of Holland's original framework but simplifies it to increase understandability and performance. ZCS's relation to Q-learning is brought out, and their performances compared in environments of two difficulty levels. Extensions to ZCS are proposed for temporary memory, better action selection, more efficient use of the genetic algorithm, and more general classifier representation.

A concise description of the XCS classi®er system's parameters, structures, and algorithms is presented as an aid to research. The algorithms are written in modularly structured pseudo code with accompanying explanations. IntroductionXCS is a recently developed learning classi®er system (LCS) that differs in several ways from more traditional LCSs. In XCS, classi®er ®tness is based on the accuracy of a classi®er's payoff prediction instead of the prediction itself. Second, the genetic algorithm (GA) takes place in the action sets instead of the population as a whole. Finally, unlike the traditional LCS, XCS has no message list and so is only suitable for learning in Markov environments (XCS extensions using an internal-state register have shown promise in non-Markov environments).XCS's ®tness de®nition and GA locus together result in a strong tendency for the system to evolve accurate, maximally general classi®ers that ef®ciently cover the state-actions pace of the problem and allow the system's knowledge' to be readily seen. As a result of these properties, there has been considerable interest in further investigation and potential extension of XCS and its principles. We therefore thought it would be useful to provide a basic algorithmic description of XCS, both as a core de®nition of the system and as a common framework from which new variants and research directions could spring.We ®rst present XCS's relation to the problem environment, followed by the system's structures and parameters. The rest of the paper consists of a top±down modular description of the XCS algorithm, written in pseudo-code accompanied by explanatory notes. We hope the result will be useful, and we encourage researchers to give us feedback regarding potential problems and clari®cations. This document should de®nitely be read in conjunction with some of the basic XCS literature, for example [Wil95], [Kov97], and [Wil98]. Additional papers on XCS and other LCSs, together with a complete LCS bibliography, are found in [LSW00]. 2 Environment interaction, structures, and parameters 2.1 Interaction with the environmentIn keeping with the typical LCS model, the environment provides as input to the system a series of sensory situations rt P f0; 1g L , where L is the number of bits in each situation. In response, the system executes actions at P fa 1 ; . . . ; a n g upon the environment. Each action results in a scalar reward qt (possibly zero). The interaction is divided into problems, which may be either single-step or multi-step. A¯ag eop indicates the end of a problem. While rt and at are interactions with the environment itself, the reward qt and the¯ag are normally provided by another component which, following [DC98], we term the reinforcement program rp. The reinforcement program determines the reward according to the current environmental input and the action that was executed. The separation of environment and reinforcement components is useful and natural because reinforcement is often not an inherent aspect of the environment, but may ...

In 1995, Beck and Katz (B&K) instructed the profession on ''What to do (and not to do) with time-series, cross-section data,'' and almost instantly their prescriptions became the new orthodoxy for practitioners. Our assessment of the intellectual aftermath of this paper, however, does not inspire confidence in the conclusions reached during the past decade. The 195 papers we reviewed show a widespread failure to diagnose and treat common problems of time-series, cross-section (TSCS) data analysis. To show the importance of the consequences of the B&K assumptions, we replicate eight papers in prominent journals and find that simple alternative specifications often lead to drastically different conclusions. Finally, we summarize many of the statistical issues relative to TSCS data and show that there is a lot more to do with TSCS data than many researchers have apparently assumed.Authors' note: We greatly benefited from the comments of