A Formal Context for Closures of Acyclic Hypergraphs ...icfca2019/... · Two Views of RDBM...
Transcript of A Formal Context for Closures of Acyclic Hypergraphs ...icfca2019/... · Two Views of RDBM...
A Formal Context for Closures of AcyclicHypergraphs
ICFCA’19
Jaume Baixeries
Departament de Ciencies de la Computacio.Universitat Politecnica de Catalunya.
26/06/2019
Layout
1 Two complementary views on RDBM dependencies.
2 Acyclic Hypergraphs (AHG).
3 Main Result: AHG’s and FCA.
Two Views of RDBM Dependencies
Dependencies are constraints that (must) apply on databases(ex: functional dependencies).
We can see dependencies from two different points of view:
Semantics and Syntax
Two Views of RDBM Dependencies: Semantics
Semantics explains a dependency w.r.t. a dataset:
Example
A functional dependency X → Y holds if and only if:
∀ti , tj : ti [X ] = tj [X ]⇒ ti [Y ] = tj [Y ]
Two Views of RDBM Dependencies: Semantics
According to its Semantics we have:
Implications.
Functional dependencies.
Multivalued dependencies.
Join dependencies.
Acyclic Join dependencies.
...
Two Views of RDBM Dependencies: Syntax
Syntax explains a dependency w.r.t. a set of dependencies (usuallyof the same kind).
Example: Transitivity
If two functional dependencies X → Y and Y → Z hold,then X → Z holds as well.
Two Views of RDBM Dependencies: Syntax
Syntax answers the following question:
Given a set of dependendencies Σ that hold in a dataset, whatother dependencies necessarily hold as well?
(regardless of the dataset)
This set is called Σ+.
Syntax induces a set of axioms for each kind of dependency.
Two Views of RDBM Dependencies: Syntax
According to its Syntax we have:
Armstrong dependencies.
Symmetric dependencies.
Join dependencies.
...
Two Views of RDBM Dependencies
Semantics Syntax
Horn Clauses / ImplicationsFunctional Dependencies Armstrong DependenciesSimilarity Dependencies
MVD ClausesDegenerated MVD Symmetric DependenciesMultivalued Dependencies
Acyclic Join Dependencies Acyclic Hypergraphs
Table: Table of references (non exhaustive) of FCA/lattice-basedcharacterization of database constraints.
Notation
What is an Acyclic Hypergraph?
Acyclic Hypergraphs
Let U be a set of attributes.
A Hypergraph is a set of subsets of U :
H = [A1,A2, . . . ,An ] Ai ⊆ U
All attributes in U need to be in one of those subsets.
Join Dependencies
With these conditions, we just have a Join Dependency.
Acyclic Hypergraphs
Let U be a set of attributes.
A Hypergraph is a set of subsets of U :
H = [A1,A2, . . . ,An ]
All attributes in U need to be in one of those subsets.
The hypergraphs is α-acyclic.
α-Acyclic Hypergraphs
With these conditions, we now have a Acyclic Hypergraph.
Acyclic Hypergraphs
Hypergraph
U = { a, b, c , d , e }
H = [ abc, bcd , cde, ae ]
α-Acyclic Hypergraph
U = { a, b, c , d , e }
H = [ abc, bcd , cde ]
Characterization of α-acyclicity
Every AHG has (at least) one join tree
bcd
abc
bc
cde
cd
Figure: Join tree for [ abc, bcd , cde ].
From the Semantical point of view
Reminder!!
This is the definition of a (Acyclic) Join Dependency.
[ ab, bc, cd ] holds in R because:
a b
1 2
3 4
×
b c
2 3
2 1
4 3
×
c d
3 5
1 2
1 5
=
R a b c d
t1 1 2 3 5
t2 1 2 1 2
t3 3 4 3 5
t4 1 2 1 5
Acyclic Hypergraphs
However, we said that we only focus on the syntax of AJD’s.
Acyclic Join Dependencies → Acyclic Hypergraphs
Join Dependencies → Hypergraphs
Acyclic Hypergraphs vs Symmetric Dependencies
A Symmetric Dependency (SD) is an AHG of size 2:
[ ab, bcd ] ≡ a⇒ b | cd
AHG’s and FCA
The closure of a set of Symmetric Dependencies has already beencharacterized in terms of FCA:
KSD = (SDU ,℘(U), I )
←− SOMETHING (with attributes) −→↑
DEPENDENCIES↓
Σ′′ = Σ+
Acyclic Hypergraphs
Symmetric Dependencies ⊂ Acyclic Hypergraphs ⊂ Hypergraphs
Acyclic Join Dependencies - Acyclic Hypergraphs
Join dependencies can’t be axiomatized.
Acyclic Join Dependencies can be axiomatized.
Acyclic Join Dependencies are equivalent to a set ofSymmetric dependencies.
Acyclic Join Dependencies - Acyclic Hypergraphs
Join dependencies can’t be axiomatized.
Acyclic Join Dependencies can be axiomatized.
Acyclic Join Dependencies are equivalent to a set ofSymmetric dependencies.
Symmetric Dependencies Axioms
(B1) Let [A,B] be a SD, X ⊆ U
[A,B]⇒ [A ∪ X ,B ∪ X ]
(B2) Let [A,E ] and [X ,Y ] be SD’s, such that Y = A ∩ E .
[A,E ], [X ,Y ]⇒ [A,E ∩ X ]
(Trivial axioms are ommited)
Acyclic Hypergraphs Axioms
(B1) Let [A,B] be an AHG, X ⊆ U
[A,B]⇒ [A ∪ X ,B ∪ X ]
(B2) Let [A,E ] and [X ,Y ] be AHG’s, such that Y = A ∩ E .
[A,E ], [X ,Y ]⇒ [A,E ∩ X ]
(A3) Let [E ,H1, . . . ,Hn] and [X ,Y ] be AHG such that X ∩ Y ⊆ Eand ∀Hi : Hi ∩ E ⊆ X Hi ∩ E ⊆ Y . LetH ′ = reduction([E ∩ X ,E ∩ Y ,H1, . . . ,Hn]).
[E ,H1, . . . ,Hn], [X ,Y ]⇒ [H ′]
(Trivial axioms are ommited)
AHG’s and FCA: Main Result
Key idea: An AHG is equivalent to a set of SD’s.
Therefore: we need to test all SD’s in the previously definedformal context.
AHG’s and FCA: Main Result
The main result in this paper is the corollary of two results:
1 The characterization of SD’s with a formal context.
2 The fact that an AHG is equivalent to a set of SD’s.
AHG’s and FCA: Main Result
Let ΣAHG be a set of AHG.
ΣAHG ⇒ ΣSD ⇒ Σ+SD ⇒ Σ+
AHGeq. SD-AHG (SDU ,℘(U), I ) eq. SD-AHG
Th. 1 & 2 Papers [*]Cor. 1 & Prop. 2
Papers [*]:
1 J. Baixeries. A Formal Context for Symmetric Dependencies,pages 90–105. Springer Berlin Heidelberg, Berlin, Heidelberg,2008.
2 J. Baixeries. A New Formal Context for SymmetricDependencies. Concept Lattices and Applications 2011.
Future Work
1 Can we (still) simplify this formal context?
2 Can we have a native formal context in terms of AHG?
3 FCA still has a long way to go...
Thanks
Vielen Dank!
Questions??