Adjusted mutual information¶

By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is.

Version: v1 · 2026-09-28 · History
Domain-specific #: 7884
Domain group: Interdisciplinary & Synthetic
Origin domain: Data Science & Analytics
Subdomains: Clustering Evaluation, Machine Learning → Data Science & Analytics

Core Idea¶

Adjusted mutual information is treated here as the recurring computer science and information systems identity summarized by this source-grounded definition: By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is.

In probability theory and information theory, adjusted mutual information, a variation of mutual information may be used for comparing clusterings. It corrects the effect of agreement solely due to chance between clusterings, similar to the way the adjusted rand index corrects the Rand index. It is closely related to variation of information: when a similar adjustment is made to the VI index, it becomes equivalent to the AMI.

The adjusted measure however is no longer metrical. It is presumed here that the partitions are so-called hard clusters; the partitions are pairwise disjoint. The mutual information of cluster overlap between U and V can be summarized in the form of an RxC contingency table M=[n_{ij}]^{i=1 \ldots R}{j=1 \ldots C} , where n denotes the number of objects that are common to clusters U_i and V_j .

For Adjusted mutual information, the abstraction is narrower than the article's general subject matter: a positive case must preserve By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is. Retaining only the name, a familiar example, or a downstream effect is insufficient. The specialist roles and tests remain anchored in computer science and information systems, which is why this identity is domain-specific rather than prime.

How would you explain it like I'm…

Matching Minus Luck

Two kids each sort the same pile of toys into groups. You want to know how much their groupings match. But some matching would happen just by luck, so adjusted mutual information takes away the lucky part and only counts the real agreement.

Luck-Corrected Group Matching

Imagine two people sorting the same set of photos into piles, each person in their own way. Mutual information is a score for how much knowing one person's piles tells you about the other's. The trouble is that even random piles would match a little by chance. Adjusted mutual information subtracts the amount of matching you would expect from random sorting, so a high score really means the two sortings agree. Each photo goes in exactly one pile.

Chance-Corrected Clustering Agreement

Adjusted mutual information (AMI) is a way to compare two clusterings, meaning two ways of dividing the same items into non-overlapping groups. It starts from mutual information, which measures how much one grouping tells you about the other, computed from a table counting how many items land in each pair of groups. Raw mutual information is biased upward because two random clusterings still share some information by chance. AMI corrects this by subtracting the expected mutual information between random clusterings, calculated under a hypergeometric model of randomness, similar to how the adjusted Rand index corrects the Rand index. After this adjustment the score is no longer a true distance metric.

Adjusted mutual information is a chance-corrected measure for comparing two partitions U and V of the same objects. Cluster overlap is summarized in an R×C contingency table whose entry n_ij counts objects shared by clusters U_i and V_j, from which mutual information is computed. Under a hypergeometric model of randomness, the expected mutual information between two random clusterings can be derived, and AMI adjusts the observed mutual information by removing this expected chance agreement. This parallels the way the adjusted Rand index corrects the Rand index. It is closely related to variation of information: applying the analogous adjustment to VI makes it equivalent to AMI. The adjusted measure is no longer a metric, and the construction assumes hard (pairwise disjoint) clusters.

Structural Signature¶

Sig role-phrases:

Defining carrier — MI is a non-negative quantity upper bounded by the entropies H(U) and H(V).
Constitutive relation — It quantifies the information shared by the two clusterings and thus can be employed as a clustering similarity measure.
Operating condition — Given a set S of N elements S={s_1, s_2,\ldots s_N} , consider two partitions of S, namely U={U_1, U_2,\ldots, U_R} with R clusters, and V={V_1, V_2,\ldots, V_C} with C clusters.
Recognition evidence — It is presumed here that the partitions are so-called hard clusters; the partitions are pairwise disjoint.
Admissible variation — Suppose an object is picked at random from S; the probability that the object falls into cluster U_i is.
Characteristic consequence — H(U) is non-negative and takes the value 0 only when there is no uncertainty determining an object's cluster membership, i.e., when there is only one cluster.
Failure boundary — MI(U,V)=\sum_{i=1}^R \sum_{j=1}^C P_{UV}(i,j)\log \frac{P_{UV}(i,j)}{P_U(i)P_V(j)}.

What It Is Not¶

Not the whole field of computer science and information systems. The node requires the specific identity stated by By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is.
Not an over-broad reading. Like the Rand index, the baseline value of mutual information between two random clusterings does not take on a constant value, and tends to be larger when the two partitions have a larger number of clusters (with a fixed number of set elements N).
Not an over-broad reading. The adjusted measure however is no longer metrical.
Not an over-broad reading. Given a set S of N elements S={s_1, s_2,\ldots s_N} , consider two partitions of S, namely U={U_1, U_2,\ldots, U_R} with R clusters, and V={V_1, V_2,\ldots, V_C} with C clusters.
Not automatically Rand index. Retrieval proximity does not establish equivalence; the two identities must be compared by carrier, operation, and failure boundary.

Scope of Application¶

Adjusted mutual information applies literally inside computer science and information systems wherever the source-defined carrier and relation can be established. Its documented habitats include:

Documented setting. In probability theory and information theory, adjusted mutual information, a variation of mutual information may be used for comparing clusterings.
Mutual information of two partitions. Given a set S of N elements S={s_1, s_2,\ldots s_N} , consider two partitions of S, namely U={U_1, U_2,\ldots, U_R} with R clusters, and V={V_1, V_2,\ldots, V_C} with C clusters.
Mutual information of two partitions. It is presumed here that the partitions are so-called hard clusters; the partitions are pairwise disjoint.
Mutual information of two partitions. Suppose an object is picked at random from S; the probability that the object falls into cluster U_i is.
Mutual information of two partitions. H(U) is non-negative and takes the value 0 only when there is no uncertainty determining an object's cluster membership, i.e., when there is only one cluster.
Mutual information of two partitions. MI(U,V)=\sum_{i=1}^R \sum_{j=1}^C P_{UV}(i,j)\log \frac{P_{UV}(i,j)}{P_U(i)P_V(j)}.

Outside computer science and information systems, the name should be retained only when these same operational conditions survive; otherwise the comparison belongs to the broader parent Theory or should be marked as analogy.

Clarity¶

A clear use of Adjusted mutual information names the carrier, the operative relation, and the conditions under which the source treats the identity as present. The minimal definition is By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is. The strongest recognition evidence in the frozen account is: It is presumed here that the partitions are so-called hard clusters; the partitions are pairwise disjoint. A report should distinguish that evidence from a proxy, consequence, or common implementation. It should also state the qualification Like the Rand index, the baseline value of mutual information between two random clusterings does not take on a constant value, and tends to be larger when the two partitions have a larger number of clusters (with a fixed number of set elements N). so that a reader can reproduce the classification rather than infer it from topical resemblance.

Manages Complexity¶

Adjusted mutual information compresses multiple computer science and information systems details into a stable diagnostic relation. The source shows both the central mechanism—it quantifies the information shared by the two clusterings and thus can be employed as a clustering similarity measure.—and the practical consequence—h(U) is non-negative and takes the value 0 only when there is no uncertainty determining an object's cluster membership, i.e., when there is only one cluster. This compression makes cases comparable while leaving parameters, conventions, exceptions, and evidential quality explicit. It is lossy by design: local history and implementation details may be omitted only when they do not alter the defining relation.

Abstract Reasoning¶

Type the carrier. Identify the computer science and information systems entities to which the claim applies.
State the relation. Use the source-grounded identity: By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is.
Check operation and conditions. Given a set S of N elements S={s_1, s_2,\ldots s_N} , consider two partitions of S, namely U={U_1, U_2,\ldots, U_R} with R clusters, and V={V_1, V_2,\ldots, V_C} with C clusters.
Demand recognition evidence. It is presumed here that the partitions are so-called hard clusters; the partitions are pairwise disjoint.
Test variation. Change an implementation or setting while preserving suppose an object is picked at random from S; the probability that the object falls into cluster U_i is.
Run the collapse test. Remove the defining operation; if the label still seems equally apt, only a topic or correlate was retained.
Reduce cautiously. When the specialist conditions cannot be carried, route the residual comparison to Theory.

Knowledge Transfer¶

Within the home domain. Knowledge about Adjusted mutual information transfers literally when a new case preserves the same carrier type, relation, and recognition test. In probability theory and information theory, adjusted mutual information, a variation of mutual information may be used for comparing clusterings. Given a set S of N elements S={s_1, s_2,\ldots s_N} , consider two partitions of S, namely U={U_1, U_2,\ldots, U_R} with R clusters, and V={V_1, V_2,\ldots, V_C} with C clusters.

Beyond the home domain. No canonical parent is asserted for Adjusted mutual information. An outside case receives the specialist name only when the same typed roles and rejection conditions can be filled literally; otherwise the comparison remains an analogy pending later graph densification.

Examples¶

Canonical¶

Given a set S of N elements S={s_1, s_2,\ldots s_N} , consider two partitions of S, namely U={U_1, U_2,\ldots, U_R} with R clusters, and V={V_1, V_2,\ldots, V_C} with C clusters. This case is canonical because it supplies a concrete carrier and lets the defining relation be checked rather than merely named.

Mapped back: carrier → the entities in the documented case; operation → By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is; recognition evidence → It is presumed here that the partitions are so-called hard clusters; the partitions are pairwise disjoint

Applied / In Practice¶

It is presumed here that the partitions are so-called hard clusters; the partitions are pairwise disjoint. The applied case shows how the identity is used under a second setting or qualification while keeping the same operative relation.

Mapped back: changed setting → Mutual information of two partitions; invariant → By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is; boundary → the case exits the class when like the Rand index, the baseline value of mutual information between two random clusterings does not take on a constant value, and tends to be larger when the two partitions have a larger number of clusters (with a fixed number of set elements N)

Structural Tensions¶

T1 — Stable identity versus admissible variation. Like the Rand index, the baseline value of mutual information between two random clusterings does not take on a constant value, and tends to be larger when the two partitions have a larger number of clusters (with a fixed number of set elements N). The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Which changes preserve the defining relation, and which replace it?

T2 — Recognition versus proxy. The adjusted measure however is no longer metrical. The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Does the cited evidence establish the identity or only a correlated sign?

T3 — Definition versus implementation. Given a set S of N elements S={s_1, s_2,\ldots s_N} , consider two partitions of S, namely U={U_1, U_2,\ldots, U_R} with R clusters, and V={V_1, V_2,\ldots, V_C} with C clusters. The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Is the observed implementation constitutive, optional, or merely common?

T4 — Scope versus overextension. It is presumed here that the partitions are so-called hard clusters; the partitions are pairwise disjoint. The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Can every claimed application fill the same typed roles without metaphor?

T5 — Transfer versus domain accent. MI is a non-negative quantity upper bounded by the entropies H(U) and H(V). The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: Does the receiving case instantiate Adjusted mutual information literally, co-instantiate Theory, or only resemble it?

T6 — Autonomy versus reduction. It quantifies the information shared by the two clusterings and thus can be employed as a clustering similarity measure. The tension matters because emphasizing only one side either dissolves the identity or overstates what the evidence and domain conventions warrant.

Diagnostic: What does Adjusted mutual information distinguish that the broader parent Theory leaves together?

Structural–Framed Character¶

Adjusted mutual information is structural-leaning. Its structural side is the repeatable organization summarized by By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is. Its framed side is the computer science and information systems vocabulary that fixes the carrier, evidence, exceptions, and admissible transformations.

Evaluative weight: the identity can be stated descriptively even when applications carry practical stakes. Human-practice dependence: the source-grounded carrier determines whether the relation exists independently or is constituted by a practice. Institutional origin: disciplinary conventions stabilize the name and test. Vocabulary portability: Given a set S of N elements S={s_1, s_2,\ldots s_N} , consider two partitions of S, namely U={U_1, U_2,\ldots, U_R} with R clusters, and V={V_1, V_2,\ldots, V_C} with C clusters. Import versus recognition: literal transfer requires the same mechanism; shape alone is analogy.

Its portable skeleton is Theory. Its character: a recurring specialist identity whose thin organization can be abstracted, while its operational meaning remains domain-bound.

Structural Core vs. Domain Accent¶

What is skeletal. By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is. The stable skeleton is the typed relation expressed in that definition and the entry's recognition and collapse tests. The source identifies these operative conditions: MI is a non-negative quantity upper bounded by the entropies H(U) and H(V). It quantifies the information shared by the two clusterings and thus can be employed as a clustering similarity measure. It further constrains recognition and variation through: Given a set S of N elements S={s1, s2,\ldots sN} , consider two partitions of S, namely U={U1, U2,\ldots, UR} with R clusters, and V={V1, V2,\ldots, VC} with C clusters. It is presumed here that the partitions are so-called hard clusters; the partitions are pairwise disjoint.

What is domain-bound. computer science and information systems supplies the operative entities, technical vocabulary, warrants, and exceptions that make Adjusted mutual information literal. Its documented scope includes the condition that In probability theory and information theory, adjusted mutual information, a variation of mutual information may be used for comparing clusterings. Another bounded application condition is that Given a set S of N elements S={s1, s2,\ldots sN} , consider two partitions of S, namely U={U1, U2,\ldots, UR} with R clusters, and V={V1, V2,\ldots, VC} with C clusters. These are not decorative examples; they determine which carrier and evidence can fill the abstraction's roles.

Why no parent is asserted. Removing those specialist details does not currently yield one live catalog node that is a necessary genus for every instance. The entry is therefore approved as unparented rather than attached by topical resemblance. Its collapse evidence remains specific—Suppose an object is picked at random from S; the probability that the object falls into cluster Ui is.—and future graph densification may discover a defensible relation only if it preserves that boundary.

Approved unparented node. No current live node supplies a defensible necessary genus or structural prerequisite for Adjusted mutual information. The reviewed identity is: By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is. The accelerated suggestion was declined because topical or lexical similarity does not establish hierarchy; the node is admitted without a parent pending later graph densification.
Related reasoning operations. Evidence, representation, comparison, classification, transformation, or evaluation may participate in particular cases, but participation does not make any one of them a necessary parent of every instance.

Neighborhood in Abstraction Space¶

Adjusted mutual information sits in a sparse region of the domain-specific corpus (64^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Clustering Methods & Validity Measures (14 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

Theory. The parent omits the specialist differentia. Tell: Can the case establish By adopting a hypergeometric model of randomness, it can be shown that the expected mutual information between two random clusterings is?
Rand index. A pair-counting similarity measure for two partitions that counts element pairs on which both clusterings agree. Tell: Which entry's carrier, operation, and failure condition are satisfied?
Modifiable Areal Unit Problem. Statistics computed on aggregated data change, sometimes reversing sign, when the boundaries used to aggregate are redrawn — the partition is a non-neutral analytical input. Tell: Which entry's carrier, operation, and failure condition are satisfied?
Partition Dependence of Aggregates. Any statistic computed on partition-aggregated data is a function of the partition itself, not solely of the underlying data. Tell: Which entry's carrier, operation, and failure condition are satisfied?
A measurement, proxy, or consequence. Those may provide evidence without being the identity. Tell: Would Adjusted mutual information remain present if the detector or downstream effect changed?
A metaphorical analogue. A similar shape outside computer science and information systems lacks the specialist mechanism. Tell: Do the native roles transfer literally, or only the parent Theory?

References¶

Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Adjusted_mutual_information (revision 1211840087).
Preserved source candidate: http://jmlr.csail.mit.edu/papers/volume11/vinh10a/vinh10a.pdf
Preserved source candidate: http://sites.google.com/site/vinhnguyenx/softwares
Preserved source candidate: https://github.com/defleury/adjusted_mutual_information
Preserved source candidate: https://scikit-learn.org/stable/modules/generated/sklearn.metrics.adjusted_mutual_info_score.html

The frozen Wikipedia revision is discovery provenance. The retained source set was reviewed for identity, formal or operational relation, and scope. The encyclopedia's structural synthesis is bounded to those claims; a thin authority surface is recorded as a nonblocking source-strengthening repair rather than concealed.