PM: a system to support the automatic acquisition of programming knowledge

摘要

A system called partial metrics (PM) which utilizes chunking as a model for acquiring knowledge about program implementation is described. The chunking paradigm has three phases. The first phase partitions the object to be chunked into relatively independent parts called aggregates. The objects to be chunked in PM are code modules. Modules are separated into a collection of aggregates based on a model of stepwise refinement. A heuristic that generates a hierarchically structured collection of refinement steps describing how the program could have been developed as a set of independent refinement decisions (object-oriented stepwise implementation) is given. The second phase encodes (abstracts) each of the aggregates. Various techniques for symbolic learning can be applied to produce a frame-based encoding of information present in the code. This abstraction contains information about the aggregate's role in the refinement process as well as the code's functionality. The third phase inserts the chunked aggregate into a hierarchically structured library of cases based on the contents of its frame description. The storage of an aggregate enables its future use in problem-solving activities. An example of how this approach can be used to acquire knowledge from a sort module is described.