Skip to main content
12 events
when toggle format what by license comment
Apr 16 at 6:32 comment added domotorp I'm again getting a bit lost with the definitions. I think it would be best if someone else answered your question.
Apr 15 at 16:36 comment added Neophyte In general, if we take a set of $d$ independent vectors and apply an affine transformation to transform them into $e_1,\cdots, e_d$, it need not be the case that the hypothesis that realize the labeling vector $l \in \{-1,1\}^d$ lie in $C$ itself. For $\mathbb R^d$ cone, this was trivial but need not be the same for a non-trivial full dimensional polyhedral cones like pointed ones.
Apr 15 at 16:30 comment added Neophyte I think the issue is that the cone $C$ can be restricted/pointed and need not contain all hypotheses that would shatter the d independent points. For example, in 2D, we can clearly see that no two points can be shattered by a pointed full dimensional cone - example. So, the VC of that 2D cone seems to align with $VC = d-1 = 1$.
Apr 13 at 4:42 comment added domotorp I agree that the cited answer by @domotorp is quite hard to read. But why can't you take an affine transformation that converts any independent $x_1\ldots,x_d$ to $e_1,\ldots,e_d$ to get an orthant? Now it seems to me that the VC-dim will be even $d$, I don't get why the other answer claims only $d-1$.
Apr 10 at 20:04 history edited Neophyte CC BY-SA 4.0
added 47 characters in body
Apr 10 at 20:03 history edited Neophyte CC BY-SA 4.0
deleted 47 characters in body
Apr 10 at 19:28 history edited Neophyte CC BY-SA 4.0
deleted 6 characters in body
Apr 10 at 19:27 history edited Neophyte CC BY-SA 4.0
added 349 characters in body
Apr 10 at 19:18 history edited Neophyte CC BY-SA 4.0
added 1 character in body
Apr 10 at 19:18 history edited Neophyte CC BY-SA 4.0
added 30 characters in body
S Apr 10 at 19:16 review First questions
Apr 10 at 23:30
S Apr 10 at 19:16 history asked Neophyte CC BY-SA 4.0