Shattering

From Machinelearning
Revision as of 20:02, 15 July 2019 by IssaRice (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Let X be a set of objects one wants to label as 0 or 1, and let H be a class of functions from X to {0,1}, called a hypothesis class. In other words, H{0,1}X. Let S={p1,,pn}X be a set of points in X. The hypothesis class H shatters S iff for every label assignment f:S{0,1} there exists some hH such that h(p)=f(p) for all pS.

We can use the idea of function restriction to restate this idea:

The hypothesis class H shatters S iff for every label assignment f:S{0,1} there exists some hH such that h|S=f.

The hypothesis class H shatters S iff {h|S:hH}={0,1}S.

Since functions going into {0,1} can be seen as subsets of the domain, we can also state the definition using just sets. Let H'={h1({1}):hH} be the set of possible regions that the hypothesis class can label as positive.

The hypothesis class H shatters S iff for each subset E of S there exists some set HH such that E=HS.

The hypothesis class H shatters S iff P(S)={HS:HH}.

References

https://en.wikipedia.org/wiki/Shattered_set

Understanding Machine Learning, chapter on VC dimension.