The need for learning problem solving knowledge has been presented in several areas, such as in behavior cloning, acquiring teaching knowledge for intelligent tutoring systems or simply to be used as heuristics in game playing. In this paper, we focus on learning understandable and easy-to-use problem solving knowledge that can be used by human problem solvers. We suggest an algorithm for learning strategies, where a strategy is a sequence of subgoals. Each subgoal is a prerequisite for the next goal in the sequence, such that achieving one goal enables us to achieve the next goal with a limited amount of search. As the sequence of subgoals concludes with the main goal, such a strategy facilitates problem solving. The algorithm learns several strategies from a state-space representation of the domain. Each state needs to be described with a set of attributes that are used to define subgoals, hence a subgoal can be seen as a subset of states satisfying some description. We demonstrate the algorithm on three domains. In the simplest one, learning strategies for mathematical equation solving, we learn strategies from a complete state-space representation, where each state corresponds to a learning example. In the other two examples, the 8-puzzle and Prolog programming, the complete state-space is too large to be used in learning, and we introduce an extension of the algorithm that can learn from particular solution paths, can exploit implicit conditions and uses active learning to select states that are expected to have the most influence on the learned strategies.

Paper submitted to journal.

• Source code of Orange 3 add-on (new version, but without active learning)
• Source code of Orange 2 add-on (with active learning)