Besides classification and regression type of problems (predictive modeling), one could be interested in finding mathematical expressions that best fit the relationship between inputs and output.
In other words, I want to reframe the problem from something like y = f(x1, x2, ...) (where
y is the predicted output) to an implicit equation of the form f(x1, x2, .. y) = 0. In this setup, symbolic regression would search for an analytical expression that satisfies this equation for all data points.
Could you please confirm:
1 If the current implementation doesn’t care that the target is a constant (0), and it will still attempt to find a meaningful expression connecting all variables. In this case there is no variability in the target variable.
2 If it is not supported currently, is there a work-around to force the search work properly in this case?
A quick research shows that these 4 metrics could make sense in this context:
Mean Error
RMS Error
Maximum Error
Residual Sum of Squares (RSS)
Thanks.
Tim
