Portfolio Methods for Optimal Planning: An Empirical Analysis

Abstract

Abstract-Combining the complementary strengths of several algorithms through portfolio approaches has been demonstrated to be effective in solving a wide range of AI problems. Notably, portfolio techniques have been prominently applied to suboptimal (satisficing) AI planning.Here, we consider the construction of sequential planner portfolios for (domain-independent) optimal planning. Specifically, we introduce four techniques (three of which are dynamic) for per-instance planner schedule generation using problem instance features, and investigate the usefulness of a range of static and dynamic techniques for combining planners. Our extensive experimental analysis demonstrates the benefits of using static and dynamic sequential portfolios for optimal planning, and provides insights on the most suitable conditions for their fruitful exploitation.

What you always wanted to know about the deterministic part of the International Planning Competition (IPC) 2014 (but were too afraid to ask)

“…Evidence points to there being a set of reasons for this break in trend. First, it has been empirically shown that algorithm selection and combination approaches for optimal planning do not generalize well on previously unseen domains (Rizzini et al ., 2015). Second, all the portfolio approaches that took part in the optimal track of IPC 2014 exploited a very similar set of basic solvers, which is a subset of the participants of IPC 2011.…”

“…(2015), Rizzini et al . (2015), Vallati et al . (2015b); a thorough discussion about mining IPC 2011 results can be found in Cenamor et al .…”

Static and Dynamic Portfolio Methods for Optimal Planning: An Empirical Analysis

“…In particular, we find that (i) our new model-based and similarity-based approaches are more robust in that they generalise better to new domains of planning problems than the static portfolios and Planzilla; (ii) when the training set is representative of testing problems, our model-based approaches consistently outperform static portfolios. This paper is a continuation of, and expands upon, an earlier version of our work on portfolio selection [17]. Specifically, in this paper we give much more detail on our four new portfolio selection techniques and dynamic portfolio selection in general, present additional experimental results, and discuss our previous results in more detail.…”

Automated Algorithm Selection: Survey and Perspectives

“…A second per-instance algorithm selection approach for AI planning has been considered by Rizzini et al (2015Rizzini et al ( , 2017. Their PLANZILLA system can be seen as an application of the previously outlined * ZILLA approach (Cameron et al, 2017) to AI planning.…”

“…Using all planners that participated in the optimal track of the 2014 International Planning Competition (IPC-14), PLANZILLA was found to substantially outperform these individual planners and achieve performance close to that of the VBS (Rizzini et al, 2015(Rizzini et al, , 2017. However, when evaluated on a set of testing instances dissimilar from those used for training, it was found that dynamic algorithm scheduling approaches performed better than PLANZILLA; these approaches dynamically construct an algorithm schedule by performing multiple stages of per-instance algorithm selection, using not only features of the planning instance i to be solved, but also taking into account which component planners have already been run on i, without success, in earlier stages of the schedule.…”