SJ001/AI-Feynman

Would it be hard to support ranges of result values (a min and max value) and/or integer results?

Opened this issue · 0 comments

Say I have an equation with N input symbols and the result is an integer. So in the results right now I'd have a value like 4. What I'd want instead is 3.50001 to 4.49999. (If the computed equation outputs anything in that range then it would be correct). This would be the equivalent of wrapping the computed expression with a round function to get the result. Thinking out loud, but it could still try to weight toward the middle value between the min/max, but it wouldn't try to make a super complex expression if it the results are in their respective range.

The change to the input format for this use case would be the last two columns would have a min and max result value. Not sure if this kind of change is a research paper in itself to figure out or an obvious change to the existing code to support. I'd picture this as a boolean parameter so the default behavior would be a single result column.

An alternative idea would be having a boolean flag to support wrapping the resultant expression in a round function. Not sure if that works with the algorithm.

edit: I believe the integer solution wouldn't work since the data has to be normalized more or less or not use a wide range of numbers. The min/max value solution would work better as it could be normalized just like the rest of the columns.