Variation of information

In probability theory and information theory, the variation of information or shared information distance is a measure of the distance between two clusterings (partitions of elements). It is closely related to mutual information; indeed, it is a simple linear expression involving the mutual information. Unlike the mutual information, however, the variation of information is a true metric, in that it obeys the triangle inequality.

Information diagram illustrating the relation between information entropies, mutual information and variation of information.

Definition

Suppose we have two partitions $X$ and $Y$ of a set $A$ into disjoint subsets, namely $X=\{X_{1},X_{2},\ldots ,X_{k}\}$ and $Y=\{Y_{1},Y_{2},\ldots ,Y_{l}\}$.

Let:

$n=\sum _{i}|X_{i}|=\sum _{j}|Y_{j}|=|A|$

$p_{i}=|X_{i}|/n$ and $q_{j}=|Y_{j}|/n$

$r_{ij}=|X_{i}\cap Y_{j}|/n$

Then the variation of information between the two partitions is:

$\mathrm {VI} (X;Y)=-\sum _{i,j}r_{ij}\left[\log(r_{ij}/p_{i})+\log(r_{ij}/q_{j})\right]$.

This is equivalent to the shared information distance between the random variables i and j with respect to the uniform probability measure on $A$ defined by $\mu (B):=|B|/n$ for $B\subseteq A$.

Explicit information content

We can rewrite this definition in terms that explicitly highlight the information content of this metric.

The set of all partitions of a set form a compact lattice where the partial order induces two operations, the meet $\wedge$ and the join $\vee$, where the maximum ${\overline {\mathrm {1} }}$ is the partition with only one block, i.e., all elements grouped together, and the minimum is ${\overline {\mathrm {0} }}$, the partition consisting of all elements as singletons. The meet of two partitions $X$ and $Y$ is easy to understand as that partition formed by all pair intersections of one block of, $X_{i}$, of $X$ and one, $Y_{i}$, of $Y$. It then follows that $X\wedge Y\subseteq X$ and $X\wedge Y\subseteq Y$.

Let's define the entropy of a partition $X$ as

$H\left(X\right)\,=\,-\sum _{i}\,p_{i}\log p_{i}$,

where $p_{i}=|X_{i}|/n$. Clearly, $H({\overline {\mathrm {1} }})=0$ and $H({\overline {\mathrm {0} }})=\log \,n$. The entropy of a partition is a monotonous function on the lattice of partitions in the sense that $X\subseteq Y\Rightarrow H(X)\geq H(Y)$.

Then the VI distance between $X$ and $Y$ is given by

$\mathrm {VI} (X,Y)\,=\,2H(X\wedge Y)\,-\,H(X)\,-\,H(Y)$.

The difference $d(X,Y)\equiv |H\left(X\right)-H\left(Y\right)|$ is a pseudo-metric as $d(X,Y)=0$ doesn't necessarily imply that $X=Y$. From the definition of ${\overline {\mathrm {1} }}$, it is $\mathrm {VI} (X,\mathrm {1} )\,=\,H\left(X\right)$.

If in the Hasse diagram we draw an edge from every partition to the maximum ${\overline {\mathrm {1} }}$ and assign it a weight equal to the VI distance between the given partition and ${\overline {\mathrm {1} }}$, we can interpret the VI distance as basically an average of differences of edge weights to the maximum

$\mathrm {VI} (X,Y)\,=\,|\mathrm {VI} (X,{\overline {\mathrm {1} }})\,-\,\mathrm {VI} (X\wedge Y,{\overline {\mathrm {1} }})|\,+\,|\mathrm {VI} (Y,{\overline {\mathrm {1} }})\,-\,\mathrm {VI} (X\wedge Y,{\overline {\mathrm {1} }})|\,=\,d(X,X\wedge Y)\,+\,d(Y,X\wedge Y)$.

For $H(X)$ as defined above, it holds that the joint information of two partitions coincides with the entropy of the meet

$H(X,Y)\,=\,H(X\wedge Y)$

and we also have that $d(X,X\wedge Y)\,=\,H(X\wedge Y|X)$ coincides with the conditional entropy of the meet (intersection) $X\wedge Y$ relative to $X$.

Identities

The variation of information satisfies

$\mathrm {VI} (X;Y)=H(X)+H(Y)-2I(X,Y)$,

where $H(X)$ is the entropy of $X$, and $I(X,Y)$ is mutual information between $X$ and $Y$ with respect to the uniform probability measure on $A$. This can be rewritten as

$\mathrm {VI} (X;Y)=H(X,Y)-I(X,Y)$,

where $H(X,Y)$ is the joint entropy of $X$ and $Y$, or

$\mathrm {VI} (X;Y)=H(X|Y)+H(Y|X)$,

where $H(X|Y)$ and $H(Y|X)$ are the respective conditional entropies.

The variation of information can also be bounded, either in terms of the number of elements:

$\mathrm {VI} (X;Y)\leq \log(n)$,

Or with respect to a maximum number of clusters, $K^{*}$:

$\mathrm {VI} (X;Y)\leq 2\log(K^{*})$

Triangle inequality

To verify the triangle inequality $\mathrm {VI} (X;Z)\leq \mathrm {VI} (X;Y)+\mathrm {VI} (Y;Z)$, expand using the identity $\mathrm {VI} (X;Y)=H(X|Y)+H(Y|X)$. It suffices to prove $$H(X|Z)\leq H(X|Y)+H(Y|Z)$$The right side has a lower bound $$H(X|Y)+H(Y|Z)\geq H(X|Y,Z)+H(Y|Z)=H(X,Y|Z)$$ which is no less than the left side.

Further reading

• Meila, M. (2007). "Comparing clusterings—an information based distance". Journal of Multivariate Analysis. 98 (5): 873–895. doi:10.1016/j.jmva.2006.11.013.
• Kingsford, Carl (2009). "Information Theory Notes" (PDF). Retrieved 22 September 2009.
• Kraskov, Alexander; Harald Stögbauer; Ralph G. Andrzejak; Peter Grassberger (2003). "Hierarchical Clustering Based on Mutual Information". arXiv:q-bio/0311039.

External links

• Partanalyzer includes a C++ implementation of VI and other metrics and indices for analyzing partitions and clusterings
• C++ implementation with MATLAB mex files