The invalidation of pysr under some constraint conditions

[jackeuylov]

Hi Miles @MilesCranmer
In my work, I need to lock the equation y1 = f( k1, k2,... ) * x1 + g( k1, k2,... ) * x2, (x, y, k are all variables) , and need to balance the dimension, and limit the appearance of constants. Under these constraints, pysr unexpectedly fails : it cannot grow a reliable expression tree, but stays at the root node.
I have tried to increase randomness and increase the number of populations, but all failed. As shown in the picture, the final result has only a very simple expression.

But miraculously, when I reverse the problem, that is, x1 = f ( k1, k2,... ) * y1 + g ( k1, k2,... ) * y2, pysr can grow normally and evolve a binary tree.

Do you have any clue about this ? I guess it is because the amount of information contained in x and y is different ?
Thanks！

[MilesCranmer]

This is going to be so so so easy once 1.0 drops. This kind of feature is built-in! 1.0.0 should drop this or next week.

but until then, what is your current approach? Is it a custom loss function?

[jackeuylov]

Yes, I implemented the built-in function according to Example 9, which returns the maximum value for a binary tree that does not satisfy the specified equation form. I have been trying to denoise the data for the past few days, but it still doesn't seem to work. I suspect that it may be because there is no solution in the solution space of this form? This is the only possibility I can think of, so I am planning to abandon this equation form.
However, the built-in functions I customized for controlling other equation forms have been working very well.

[MilesCranmer]

Check out https://astroautomata.com/SymbolicRegression.jl/dev/examples/template_expression/