# Is feasible learning using just one algorithm and just one class of functions?

Suppose that in an enterprise there is a section specialized in data prediction problems, and to make easier the software maintainance, the next decision is taken: it will be used only one algorithm and only one class of functions to approach solutions for any future problems. Will this decision benefit the enterprise?

I know that, for example, for non-linear models, it is possible to transform datasets using a transformation of the form $x \mapsto (x,f(x))$, in order to obtain a model that can be approximated by linear functions (as a particular example, we can transform quadratic datasets in $\mathbb{R}^2$ in linear ones using the mapping $(x,y) \mapsto (1,x,y,xy,x^2,y^2)$) but, is this enough to ensure that using only one algorithm and one class of functions will benefit the enterprise?

• It's never a good idea to limit oneself. The intended answer, though, is possibly the kernel trick. – Yuval Filmus Mar 26 '17 at 13:17
• "Will this decision benefit the enterprise?" sounds unanswerable to me, based on the information you have provided, as in real life benefits to an enterprise are a complex mix of factors. If this is an exercise, I suggest trying to reframe your question as a specific, answerable technical question. Do you want to know whether the kernel trick is enough to learn every possible binary classification situation? – D.W. Mar 26 '17 at 23:10