I. EVALUACIÓN DE LAS POLÍTICAS PÚBLICAS CUESTIONES
I.4. CUANTIFICACIÓN DEL PROGRAMA DE EVALUACIÓN
I.4.1. El Sistema de Seguimiento
I.4.1.1. Los indicadores de Seguimiento
the presence of records, and in the presence of records and lists.
Let
Tbe any non-empty subset of {lists, sets, multisets} apart from {lists} . The gen
eralised A rmstrong axioms, i. e.,
X -+ Y y � X' -+
{Y} y � X,
{X, Y}
-+ UNX, Y reconcilable,
X -+ Y, Y -+ Z X -+ Z
form a minimal, sound and complete set of inference rules for the implication of FDs in
the presence of records and
T. 0In terms of Figure 1 . 1 , Theorem 5.23 extends the knowledge on the class of FDs and the problem of axiomatisability along the data type dimension covering all combinations of record-, list-, set- and multiset-valued attributes. We would like to do the same extension for the implication problem of FDs.
5 .4 Implication Problem
In view of Theorem 5. 15,
E
f= X -+ Y holds if and only ifE
f-9\ X-+
Y holds where9'\ are the generalised Armstrong axioms from Definition 5.4. Given some set
E
one can enumerate all FDs derivable from it. However, the enumeration algorithm is time consuming and therefore impractical. We will now present a provably-correct membership algorithm and prove that it works efficiently, i.e., in polynomial time in the number of subattributes of the underlying nested attribute and the number of FDs given.5.4.1 The Closure
Similar to the RDM [29] and similar to the case of list-valued attributes in Chapter 3 we introduce the notion of a closure for a set of nested attributes with respect to a given set of FDs. Please note that this notion already played an important role in proving Theorem 5. 15. The closure is defined with respect to the set 9'\ of the generalised Armstrong axioms from Definition 5.4.
5.4. IMPLICATION PROBLEM Sebastian Link Definition 5.24. Let
N
E N A, X� Sub(N)
a set of subattributes ofN,
andE
a set of FDs onN.
Theclosure x+ � Sub(N)
of X with respect toE
isx+ =
{ Z : X -t { Z} EE+
} . oAccording to Theorem 5.15 the closure
x+
of X is therefore the set of all nested attributes which are functionally determined by X with respect to a given setE
of FDs. The computation ofx+
is sufficient for deciding whetherE f=
X -tY
holds.Lemma 5.25.
Let N E
N A,and E a set of FDs on N. Then
X -t
Y
EE+
if and only if Y
�x+
Proof.
If X ---+Y
EE+,
then X -t { Y } EE+
for all Y EY
by the subset rule. This meansall Y E
Y
are elements ofx+,
i.e. ,Y � x+.
Assuming that every Y E
Y
also satisfies Y Ex+
implies that X -t {Y} EJ.;+
for allY E
Y.
We infer that X -tY
EE+
by the derivability of the union rule. 0Let X, Y �
Sub(N) .
We call X ageneralised subset
ofY,
denoted by X�gen Y,
if and only if for every X E X there is some Y EY
with X ::; Y (Hoare-ordering) . Note that� gen
is a pre-order (reflexive, transitive) on the powersetP(Sub(N))
ofSub(N) .
The distinct sets X = {L[A] , L [-\] } andY =
{L[A] } are generalised subsets of one another, i.e.,� gen
is not symmetric.The projection of any tuple on a superattribute always determines the projection of this tuple on each of its subattributes. It is therefore sufficient to consider only maximal subattributes with respect to ::; . The set of all maximal elements of some :S-ideal X
�
Sub(N)
is formally defined as Xmax = { X E X : VZ E X.X ::; Z implies X = Z} . It is an immediate consequence of Lemma 5 .26 thatY � x+
if and only if Y �gen X�ax·Lemma 5.26.
Let N
E N A,and
X �Sub(N) an ideal with respect to ::; . For all Y �
Sub(N) we have
Y
� Xif and only if Y �gen
XmaxProof.
We show theonly if
part first. Let Y EY
be arbitrary. FromY �
X follows Y E X.Consequently there is some Z E Xmax with Y ::; Z. This shows that Y
�gen
Xmax·It remains to consider the
if
part. Let Y EY
be arbitrary again. SinceY �gen
Xmax, there is some Z E Xmax with Y ::; Z. Since Xmax�
X we haveZ
E X. However, X is an ideal with respect to ::; , i.e., Y E X holds as well. This showsY
� X. 05.4.2 Units of Nested Attributes
In order to solve the implication problem for FDs on some nested attribute
N
we will splitN
into mutually reconcilable subattributesNi
and solve the projected implication problems on theNi
simultaneously. The idea is to choose the unitsNi
ofN
such that for all subattributesU, V
ESub(N)
we have thatU
andV
are reconcilable if and only if for all unitsNi
ofN
the subattributesU n Ni
andV n Ni
are comparable with respect to ::; . This motivates the following definition.Definition 5.27. Let
N
E N A. A nested attributeNi
E N A is aunit of N
if and only if1 .
Ni
:SN,
2.
VX, Y ::; Ni,
ifX
andY
are reconcilable, thenX ::; Y
orY ::; X,
3.Ni is ::;-maximal with properties 1 . and 2.
The set of all units of
N
is denoted byU(N) .
DThe property that two subattributes
U, V
ESub(N)
are reconcilable is not tran sitive: ifN = L(K { M(A, B)}, C)
andU = L(K { M (A, .\) } , A) ,
W= L(A, C)
andV = L(K{M(A, B) } , A),
thenU
and W are reconcilable, W andV
are reconcilable, butU
andV
are not reconcilable. In fact,U, V
ESub(L(K{M(A, B) } , A)) ,
but they are incom parable with respect to ::; .E xAMPLE 5 . 5 . Let
N = L1 (L2(L3 (A, B)) , L4 [L5(C, L6 (D) )], L7(E, L8{ L9 (F, G, H)})).
The units ofN
are- Ll (L2 (L3(A, B)) , A, L7(.\, .\)) ,
- L1 (.\, £4[£5( C, .\)] , £7(.\, .\)),
- L1 (.\, L4[L5(A, L6(D) )] , L7 (A, .\)),
- L1 (.\, A, L7(E, A))
and- L1 (.\, A, L1(.\, L8 {L9(F, G, H) })).
Clearly
L1 (A, A, L7(A, L8 {L9 (.\, G, H)}))
also has properties 1 . and 2. of Definition 5.27,but is not maximal with respect
::;
. DNext we give an inductive characterisation of units.
k
Lemma 5 .28.