|A suggestion and a question on answer evaluators
|topic started 2/7/2002; 12:03:26 PM
last post 2/7/2002; 3:55:47 PM
|Michael Gage - Re: A suggestion and a question on answer evaluators
2/7/2002; 3:55:47 PM (reads: 1210, responses: 0)
In response to your suggestion: I think the idea of callbooks or hooks is a very good idea. First steps towards implementing this have already been taken (see the code in the last section of AnswerHash.pm which defines the AnswerEvaluator objects). NUM_CMP and FUN_CMP have been modified to use this structure, but some of the other answer evaluators should probably be rewritten to use these hooks as well. You will need to look at the code itself. The documentation notes, even the pod documentation in the code, is not yet completed.
The construction is entirely around filters, and the idea is that you can construct varients of these by plugging in additional filters. Or you could simply recombine existing filters to create new answer evaluators. I think the basic structure is pretty sound, but there is a ways to go to add the right API's to make it easy to construct answer evaluators. And (of course) documentation is lacking. Take a look at AnswerHash.pm and let us know what you think.
In answer to your second question, there have been some answer evaluators written to take a string of numbers, parse them and compare (for example to evaluate a vector). Gavin La Rose at University of Michigan has written some of these. I'll see if I can find the specific examples.
Hope this helps.