Given a hypothesis class $H=\{h:X\to\{0,1\}\}$.

Let $c\in H$ be the correct predictor.

Denote $H^c = \{c\Delta h:h\in H\}$, where $c\Delta h=(h\backslash c)\cup (c\backslash h)$.

Please prove that VCdim($H^c$) = VCdim($H$)


If some $A = (a_1,\dots,a_d\}\subset X$ is shattered by $H$, then, for any vector $b$ = $\{0,1\}^d$,$\exists$ $h \in H s.t. (h(a_1),\dots,h(a_d)) = b$.

It suffices to prove that for any vector $(h(a_1),\dots,h(a_d)) = b$, $\exists$ $h' \in H s.t. (c\Delta h'(a_1),\dots,c\Delta h'(a_d)) = (h(a_1),\dots,h(a_d))$.

Proof of my claim:

$\forall x_i \in A$, if $c(a_i) = 0$,let $h'(a_i) = h(a_i)$, then $c\Delta h' (a_i) = h(a_i)$.

If $c(a_i) = 1$, let $h'(a_i) = -h(a_i)$, then $c\Delta h'(a_i) = h(a_i)$.

Then $(c\Delta h'(a_1),\dots,c\Delta h'(a_d)) = (h(a_1),\dots,h(a_d))$.

For the same reason, the inverse is also true.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.