Use of Custom Function

bhctuary

I have a target variable "Actual" and three potential input variables X1, X2, X3.

I want the minimum error of target variable Actual to be of the form Actual = cf(X1,X2,X3)+ (1-c)g(X1,X2,X3) where c is a constant between 0 and 1 (inclusive). How would I write this in the Custom Formula box?

Is this possible to perform in TuringBot now?

Thank You,

Bert

admin

Hi @bhctuary. You can use:

Actual = f() * f(X1,X2,X3) + (1-f()) * f(X1,X2,X3)

In the custom formula syntax, f(X1,X2,X3) is a placeholder for an unknown function of X1, X2, and X3. The two f(X1,X2,X3) in the formula above will be optimized independently.
Similarly, the two constants f() will be optimized independently.

This is the closest you can get with the current version of the program. It is not currently possible to constrain the two constants to be related to each other.

admin

Hi @bhctuary, update: The ability to have repeated unknown terms in custom search has been added in v2.23, released today. As per the documentation, to use an unknown term multiple times in your equation, write it as an indexed term by adding a number after f. For example:

y = f1() + f1()*x

Both instances of f1() will be identical, creating a line with the same intercept and slope. Similarly, in:

y = f1(x) / (1 + f1(x))

Both occurrences of f1(x) will represent the same function. So, in your case, you would write:

y = f1() * f(X1,X2,X3) + (1-f1()) * f(X1,X2,X3)

Here, the two f1() constants will be identical, but not the two f(X1,X2,X3).

bhctuary

Just tried this and it very nicely addresses my issue!

THANK YOU!!!!!

Without using abs, min, or max, I wonder if there is a way to constrain a constant c (e.g., 0<c<1)?

admin

Without using abs, min, or max, I wonder if there is a way to constrain a constant c (e.g., 0<c<1)?

Not at the moment, the best I can think of is something indeed with min/max: max(min(1, f1()), 0). Or, in your case,

y = max(min(1, f1()), 0) * f(X1,X2,X3) + (1-max(min(1, f1()), 0)) * f(X1,X2,X3)

bhctuary

Thanks... that's just what I was thinking too.....