Namespaces
Variants
Actions

Difference between revisions of "Non-precise data"

From Encyclopedia of Mathematics
Jump to: navigation, search
(asy code to subpage)
m (Automatically changed introduction)
 
(4 intermediate revisions by 2 users not shown)
Line 1: Line 1:
 +
<!--This article has been texified automatically. Since there was no Nroff source code for this article,
 +
the semi-automatic procedure described at https://encyclopediaofmath.org/wiki/User:Maximilian_Janisch/latexlist
 +
was used.
 +
If the TeX and formula formatting is correct and if all png images have been replaced by TeX code, please remove this message and the {{TEX|semi-auto}} category.
 +
 +
Out of 80 formulas, 78 were replaced by TEX code.-->
 +
 +
{{TEX|semi-auto}}{{TEX|part}}
 
Real data obtained from measurement processes are not precise numbers or vectors but are more or less non-precise. This uncertainty is different from measurement errors and has to be described formally in order to obtain realistic results by mathematical models.
 
Real data obtained from measurement processes are not precise numbers or vectors but are more or less non-precise. This uncertainty is different from measurement errors and has to be described formally in order to obtain realistic results by mathematical models.
  
Line 5: Line 13:
 
Further examples of non-precise data are readings on digital measurement equipments, readings of pointers on scales, colour intensity pictures and light points on screens.
 
Further examples of non-precise data are readings on digital measurement equipments, readings of pointers on scales, colour intensity pictures and light points on screens.
  
Non-precise data are different from measurement errors because in error models the observed values <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n1201101.png" /> are considered to be numbers, i.e. <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n1201102.png" />, where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n1201103.png" /> denotes the error of the <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n1201104.png" />th observation (cf. also [[Numerical analysis|Numerical analysis]]).
+
Non-precise data are different from measurement errors because in error models the observed values $y _ { i }$ are considered to be numbers, i.e. $y _ { i } = x _ { i } + \epsilon _ { i }$, where $\epsilon_{i}$ denotes the error of the $i$th observation (cf. also [[Numerical analysis|Numerical analysis]]).
  
 
Historically, non-precise data were not studied sufficiently. Some earlier work was done in interval arithmetic. In general, non-precise sets in the form of fuzzy sets were considered by L.A. Zadeh. Some publications combining fuzzy imprecision and stochastic uncertainty came up in the 1980s, see [[#References|[a2]]]. Some of these approaches are more theoretically oriented. An applicable approach for statistical analysis of non-precise data is given in [[#References|[a4]]].
 
Historically, non-precise data were not studied sufficiently. Some earlier work was done in interval arithmetic. In general, non-precise sets in the form of fuzzy sets were considered by L.A. Zadeh. Some publications combining fuzzy imprecision and stochastic uncertainty came up in the 1980s, see [[#References|[a2]]]. Some of these approaches are more theoretically oriented. An applicable approach for statistical analysis of non-precise data is given in [[#References|[a4]]].
  
 
==Characterizing functions of non-precise numbers.==
 
==Characterizing functions of non-precise numbers.==
In the case of measurements of one-dimensional quantities, non-precise observations can be reasonably described by so-called fuzzy numbers <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n1201105.png" />. Fuzzy numbers are generalizations of real numbers in the following sense.
+
In the case of measurements of one-dimensional quantities, non-precise observations can be reasonably described by so-called fuzzy numbers $x ^ { * }$. Fuzzy numbers are generalizations of real numbers in the following sense.
  
Each real number <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n1201106.png" /> is characterized by its indicator function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n1201107.png" />. Specializing the membership functions from fuzzy set theory, a fuzzy number is characterized by its so-called characterizing function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n1201108.png" />, which is a generalization of an indicator function. A characterizing function is a real-valued function of a real variable obeying the following:
+
Each real number $X \in \mathbf R$ is characterized by its indicator function $I_{ \{ x \} } ( \cdot )$. Specializing the membership functions from fuzzy set theory, a fuzzy number is characterized by its so-called characterizing function $\xi ( . )$, which is a generalization of an indicator function. A characterizing function is a real-valued function of a real variable obeying the following:
  
1) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n1201109.png" />;
+
1) $\xi : \mathbf{R} \rightarrow [ 0,1 ]$;
  
2) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011010.png" /> such that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011011.png" />;
+
2) $\exists x \in \mathbf R $ such that $\xi ( x ) = 1$;
  
3) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011012.png" />, the so-called <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011014.png" />-cut <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011015.png" /> is a closed finite interval. Characterizing functions describe the imprecision of a single observation. They should not be confused with probability densities, which describe the stochastic variation of a random quantity <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011016.png" /> (cf. also [[Probability distribution|Probability distribution]]). Examples of characterizing functions are depicted in Fig.a1.
+
3) $\forall \alpha \in ( 0,1 ]$, the so-called $\alpha$-cut $B _ { \alpha } = \{ x \in \mathbf{R} : \xi ( x ) \geq \alpha \}$ is a closed finite interval. Characterizing functions describe the imprecision of a single observation. They should not be confused with probability densities, which describe the stochastic variation of a random quantity $X$ (cf. also [[Probability distribution|Probability distribution]]). Examples of characterizing functions are depicted in Fig.a1.
  
 
{{:Non-precise data/Fig1}}
 
{{:Non-precise data/Fig1}}
Line 24: Line 32:
 
A fundamental problem is how to obtain the characterizing function of a non-precise observation. This depends on the area of application. An example is as follows.
 
A fundamental problem is how to obtain the characterizing function of a non-precise observation. This depends on the area of application. An example is as follows.
  
For data in the form of gray intensities in one dimension, as boundaries of regions the gray intensity <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011017.png" /> as an increasing function of one real variable <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011018.png" /> can be used to obtain the characterizing function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011019.png" /> in the following way. Take the derivative <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011020.png" /> and divide it by its maximum. The resulting function, or its convex hull, can be used as characterizing function of the non-precise observation.
+
For data in the form of gray intensities in one dimension, as boundaries of regions the gray intensity $g ( x )$ as an increasing function of one real variable $x$ can be used to obtain the characterizing function $\xi ( x )$ in the following way. Take the derivative $( d / d x ) g ( x )$ and divide it by its maximum. The resulting function, or its convex hull, can be used as characterizing function of the non-precise observation.
  
 
In Fig.a2 the construction of the characterizing function from a gray intensity is explained.
 
In Fig.a2 the construction of the characterizing function from a gray intensity is explained.
<center><asy>
+
{{:Non-precise data/Fig2}}
import gsl;
 
 
 
int N=30;
 
 
 
picture whole;
 
picture common;
 
 
 
draw ((-0.3,0)--(1.45,0),Arrow);
 
draw ((0,0)--(0,1.2));
 
draw ((1,0)--(1,1.2));
 
 
 
label("$x$",(1.45,0),E);
 
 
 
add ( common, currentpicture );
 
erase();
 
 
 
 
 
add ( currentpicture, common );
 
 
 
guide g;
 
for (int k=floor(-0.2N); k<floor(1.2N); ++k) {
 
  real x = k/N;
 
  real y = cdf_gaussian_P(7*(x-0.5));
 
  y =0.85 y+0.15;
 
  g=g..(x,y);
 
}
 
draw(g,defaultpen+1.3);
 
 
 
label("$\scriptstyle g(x)$",(0.33,0.8));
 
 
 
add ( whole, shift(-60,0)*scale(40,36)*currentpicture );
 
erase();
 
 
 
 
 
add ( currentpicture, common );
 
 
 
draw((-0.2,1)--(1.2,1));
 
label("$1$",(-0.2,1),W);
 
 
 
guide g1;
 
guide g2;
 
for (int k=floor(-0.2N); k<floor(1.2N); ++k) {
 
  real x = k/N;
 
  real y = sqrt(2pi) * pdf_gaussian(7*(x-0.5));
 
  g1=g1..(x,y);
 
  g2=g2..(x,1.3y);
 
}
 
draw(g1,defaultpen+1.3);
 
draw(g2,defaultpen+1.3);
 
 
 
label("$\scriptstyle \xi(x)$",(0.5,0.4));
 
label("$\scriptstyle g'(x)$",(0.8,1.3));
 
 
 
add ( whole, shift(60,0)*scale(40)*currentpicture );
 
erase();
 
 
 
label(whole,"\small Fig. a2: Characterizing function obtained from a gray intensity ",(30,-20));
 
 
 
shipout(scale(1.3)*whole);
 
 
 
</asy></center>
 
  
 
==Non-precise samples.==
 
==Non-precise samples.==
Taking observations of a one-dimensional continuous quantity <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011021.png" /> in order to estimate the distribution of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011022.png" />, usually a finite sequence <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011023.png" /> of non-precise numbers is obtained. These non-precise data are given in form of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011024.png" /> characterizing functions <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011025.png" /> corresponding to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011026.png" />. Facing this kind of samples, even a most simple concept like the histogram has to be modified. This is necessary because, for a given class <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011027.png" /> in a histogram, for a non-precise observation <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011028.png" /> with characterizing function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011029.png" /> obeying <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011030.png" /> for an element <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011031.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011032.png" /> for an element <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011033.png" />, it is not possible to decide whether or not <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011034.png" /> is an element of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011035.png" />. The situation is explained in Fig.a3.
+
Taking observations of a one-dimensional continuous quantity $X$ in order to estimate the distribution of $X$, usually a finite sequence $x _ { 1 } ^ { * } , \ldots , x _ { n } ^ { * }$ of non-precise numbers is obtained. These non-precise data are given in form of $n$ characterizing functions $\xi _ { 1 } ( . ) , \ldots , \xi _ { n } ( . )$ corresponding to $x _ { 1 } ^ { * } , \ldots , x _ { n } ^ { * }$. Facing this kind of samples, even a most simple concept like the histogram has to be modified. This is necessary because, for a given class $K_j$ in a histogram, for a non-precise observation $x _ { i } ^ {\color{blue} *}$ with characterizing function $\xi _ { i }(.)$ obeying $\xi _ { i } ( x ) &gt; 0$ for an element $x \in K_j $ and $\xi _ { i } ( y ) &gt; 0$ for an element $y \in K _ { j } ^ { c }$, it is not possible to decide whether or not $x _ { i } ^ {\color{blue} *}$ is an element of $K_j$. The situation is explained in Fig.a3.
  
<center><asy>
+
{{:Non-precise data/Fig3}}
import gsl;
 
  
int M=30;
+
A generalization of the concept of histogram is possible by defining so-called fuzzy histograms. For such histograms the height of the histogram over a fixed class $K_j$ is a fuzzy number $h _ { j } ^ { * }$. For the definition of the characterizing function of $h _ { j } ^ { * }$, see [[#References|[a3]]]. For other concepts of statistics in the case of non-precise data, see [[#References|[a4]]].
 
 
picture whole;
 
 
 
draw ((-0.05,0)--(1.05,0),Arrow);
 
draw ((0,0)--(0,1.2),Arrow);
 
 
 
label("$x$",(1.05,0),E);
 
label("$\scriptstyle \xi_i(x)$",(0,1.2),N);
 
 
 
draw((-0.05,1)--(1.05,1));
 
label("$1$",(-0.05,1),W);
 
 
 
draw ((0.3,0)--(0.7,0),defaultpen+2);
 
draw ((0.3,0)--(0.3,1),dashed);
 
draw ((0.7,0)--(0.7,1),dashed);
 
 
 
label("$K_j$",(0.5,0),S);
 
 
 
guide g;
 
for (int k=floor(-3M); k<floor(3M); ++k) {
 
  real x = k/M;
 
  real y = sqrt(2pi) * pdf_gaussian(x);
 
  g=g..(x,y);
 
}
 
draw(shift(0.33,0)*scale(0.03,1)*g,defaultpen+1);
 
draw(shift(0.5,0)*scale(0.045,1)*g,defaultpen+1);
 
draw(shift(0.72,0)*scale(0.04,1)*g,defaultpen+1);
 
 
 
label("\small Fig. a3: Non-precise observations and a class of a histogram.",(0.5,-0.4));
 
label("\small $K_j$ is a class of a histogram and $\xi_i(\cdot)$ is a characterizing function",(0.5,-0.7));
 
 
 
shipout(scale(170,50)*currentpicture);
 
</asy></center>
 
 
 
A generalization of the concept of histogram is possible by defining so-called fuzzy histograms. For such histograms the height of the histogram over a fixed class <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011038.png" /> is a fuzzy number <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011039.png" />. For the definition of the characterizing function of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011040.png" />, see [[#References|[a3]]]. For other concepts of statistics in the case of non-precise data, see [[#References|[a4]]].
 
  
 
==Non-precise vectors.==
 
==Non-precise vectors.==
In the case of multi-variate data <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011041.png" />, for example the position of an object on a radar screen, the observations are non-precise vectors <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011042.png" />. Such non-precise vectors are characterized by so-called vector-characterizing functions <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011043.png" />. These vector-characterizing functions <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011044.png" /> are real-valued functions of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011045.png" /> real variables <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011046.png" /> obeying the following:
+
In the case of multi-variate data $\underline { x } = ( x _ { 1 } , \dots , x _ { n } )$, for example the position of an object on a radar screen, the observations are non-precise vectors $\underline{x} ^ { * }$. Such non-precise vectors are characterized by so-called vector-characterizing functions $\xi _ { \underline{x}^{*}}  ( . , \dots , . )$. These vector-characterizing functions $\xi _ { \underline{x}^{*}}  ( . , \dots , . )$ are real-valued functions of $n$ real variables $x _ { 1 } , \ldots , x _ { n }$ obeying the following:
  
a) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011047.png" />;
+
a) $\xi _ { \underline{x}^*} : \mathbf{R} ^ { n } \rightarrow [ 0,1 ]$;
  
b) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011048.png" /> such that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011049.png" />;
+
b) $\exists \underline{x} = ( x _ { 1 } , \dots , x _ { n } ) \in \mathbf{R} ^ { n }$ such that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011049.png"/>;
  
c) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011050.png" />, the <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011051.png" />-cut <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011052.png" /> is a closed and star-shaped subset with finite <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011053.png" />-dimensional content.
+
c) $\forall \alpha \in ( 0,1 ]$, the $\alpha$-cut $B _ { \alpha } ( \underline{x} ^ { * } ) = \{ \underline{x} \in \mathbf{R} ^ { n } : \xi _ { \underline{x} ^ { * } } ( \underline{x} ) \geq \alpha \}$ is a closed and star-shaped subset with finite $n$-dimensional content.
  
 
==Functions of non-precise arguments.==
 
==Functions of non-precise arguments.==
 
For the generalization of functions of real variables, the so-called extension principle from fuzzy set theory is used. This principle generalizes a function to the situation when the value of the argument variable is non-precise.
 
For the generalization of functions of real variables, the so-called extension principle from fuzzy set theory is used. This principle generalizes a function to the situation when the value of the argument variable is non-precise.
  
Let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011054.png" /> be a classical real-valued function of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011055.png" /> variables. If the argument <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011056.png" /> is precise, then the value <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011057.png" /> of the function is also a precise real number. For a non-precise argument <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011058.png" /> it is natural that the value <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011059.png" /> becomes non-precise also. The quantitative description of this imprecision is done by using the characterizing function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011060.png" /> of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011061.png" />. Let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011062.png" /> be the vector-characterizing function of the fuzzy vector <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011063.png" />. Then, for all real numbers <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011064.png" />, the values <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011065.png" />, are given by the extension principle in the following way:
+
Let $\psi : \mathbf{R} ^ { n } \rightarrow \mathbf{R}$ be a classical real-valued function of $n$ variables. If the argument $\underline { x } = ( x _ { 1 } , \dots , x _ { n } )$ is precise, then the value $\psi ( \underline{x} )$ of the function is also a precise real number. For a non-precise argument $\underline{x} ^ { * }$ it is natural that the value $\psi ( \underline{x} ^ { * } )$ becomes non-precise also. The quantitative description of this imprecision is done by using the characterizing function $\eta ( . )$ of $\psi ( \underline{x} ^ { * } )$. Let $\xi ( ., \dots , . )$ be the vector-characterizing function of the fuzzy vector $\underline{x} ^ { * }$. Then, for all real numbers $y$, the values $\eta ( y )$, are given by the extension principle in the following way:
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011066.png" /></td> </tr></table>
+
\begin{equation*} \eta ( y ) = \left\{ \begin{array} { l l } { \operatorname { sup } \{ \xi ( \underline{x} ) : \underline{x} \in {\bf R} ^ { n } , \psi ( \underline{x} ) = y \} , } &amp; { \psi ^ { - 1 } ( y ) \neq \emptyset ,} \\ { 0 , } &amp; { \psi ^ { - 1 } ( y ) = \emptyset .} \end{array} \right. \end{equation*}
  
For continuous functions <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011067.png" /> it can be proved that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011068.png" /> is a characterizing function, see [[#References|[a4]]].
+
For continuous functions $\psi ( . )$ it can be proved that $\eta ( . )$ is a characterizing function, see [[#References|[a4]]].
  
 
==Non-precise functions.==
 
==Non-precise functions.==
Consider the monitoring of quantities in continuous time for real measurements. The results are functions of time with non-precise values. Therefore the concept of a non-precise function is necessary. These are functions whose values are fuzzy numbers. Let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011069.png" /> be the domain of the non-precise function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011070.png" /> and let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011071.png" /> be the set of all fuzzy numbers; then
+
Consider the monitoring of quantities in continuous time for real measurements. The results are functions of time with non-precise values. Therefore the concept of a non-precise function is necessary. These are functions whose values are fuzzy numbers. Let $M$ be the domain of the non-precise function $f ^ { * } ( . )$ and let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011071.png"/> be the set of all fuzzy numbers; then
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011072.png" /></td> </tr></table>
 
 
 
A non-precise function can also be considered as a family of fuzzy numbers with characterizing functions <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011073.png" />, i.e.
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011074.png" /></td> </tr></table>
 
 
 
A graphical description of non-precise functions can be given using the <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011075.png" />-cuts <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011076.png" /> of the fuzzy numbers <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011077.png" />.
 
 
 
The curves <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011078.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011079.png" /> are called lower and upper <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011082.png" />-level curves. Taking a finite suitable number of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011083.png" />-levels from <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011084.png" /> and depicting the corresponding <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/n/n120/n120110/n12011085.png" />-level curves in a diagram, a good graphical display of the non-precise function is obtained. An example is given in Fig.a4.
 
<center><asy>
 
int M=30;
 
real a = 0.07;
 
real a0 = 0.15;
 
real b = 0.02;
 
real c = 0.6;
 
real d = -0.2;
 
 
 
real x0 = -0.22;
 
real u = 0.2;
 
real v = 0.32;
 
 
 
draw ((-0.7,0)--(0.3,0),Arrow);
 
draw ((x0,-0.02)--(x0,1.2),Arrow);
 
 
 
label("$x$",(0.3,0),E);
 
label(rotate(90)*"$y$",(x0,1.2),N);
 
label("$x_0$",(x0,-0.02),S);
 
 
 
guide g1; guide g2; guide g3; guide g4; guide g5;
 
for (int k=floor(-0.7M); k<floor(0.3M); ++k) {
 
  real x = k/M;
 
  real z = 1+3*x^2;
 
  real y1 = 1/(z-2a-a0)+2b*(1+c*x)+d;
 
  real y2 = 1/(z-a-a0)+b*(1+c*x)+d;
 
  real y3 = 1/(z-a0)+d;
 
  real y4 = 1/(z+a-a0)-b*(1+c*x)+d;
 
  real y5 = 1/(z+2a-a0)-2b*(1+c*x)+d;
 
  g1=g1..(x,y1);
 
  g2=g2..(x,y2);
 
  g3=g3..(x,y3);
 
  g4=g4..(x,y4);
 
  g5=g5..(x,y5);
 
}
 
draw(g1,defaultpen+1);
 
draw(g2,defaultpen+1);
 
draw(g3,defaultpen+1);
 
draw(g4,defaultpen+1);
 
draw(g5,defaultpen+1);
 
 
 
real x = x0;
 
real z = 1+3*x^2;
 
real y1 = 1/(z-2a-a0)+2b*(1+c*x)+d;
 
real y2 = 1/(z-a-a0)+b*(1+c*x)+d;
 
real y3 = 1/(z-a0)+d;
 
real y4 = 1/(z+a-a0)-b*(1+c*x)+d;
 
real y5 = 1/(z+2a-a0)-2b*(1+c*x)+d;
 
path g = (x,y1)..(x-u,y2)..(x-v,y3)..(x-u,y4)..(x,y5);
 
draw( g );
 
 
 
pair w = (0.1,-0.6);
 
 
 
pair p = point(g,0.5);
 
dot ( p );
 
draw( p--p-0.5w, dashed );
 
label(rotate(90)*"$\Psi_{x_0}(y)$",p-0.5w,N);
 
 
 
draw( (x,y2)--(x-u,y2) );
 
draw( (x,y3)--(x-v,y3) );
 
draw( (x,y4)--(x-u,y4) );
 
 
 
draw( (x-u,y2+0.05)--(x-u,y4-0.05) );
 
draw( (x-v,y2+0.1)--(x-v,y4-0.1) );
 
  
real x = -0.15;
+
\begin{equation*} f ^ { * } : M \rightarrow \mathcal{F} ( \mathbf{R} ). \end{equation*}
real z = 1+3*x^2;
 
real y4 = 1/(z+a-a0)-b*(1+c*x)+d;
 
dot( (x,y4) );
 
draw( (x,y4)--(x,y4)+w, dashed );
 
label("$\scriptstyle \underline f_\alpha(x)$",(x,y4)+w,SE);
 
  
real x = -0.05;
+
A non-precise function can also be considered as a family of fuzzy numbers with characterizing functions $\psi _ { x } ( \cdot )$, i.e.
real z = 1+3*x^2;
 
real y3 = 1/(z-a0)+d;
 
dot( (x,y3) );
 
draw( (x,y3)--(x,y3)+w, dashed );
 
label("$\scriptstyle \underline f_1(x)=\overline f_1(x)$",(x,y3)+w,SE);
 
  
real x = 0.05;
+
\begin{equation*} \{ \psi _ { x } ( . ) \widehat{=} f ^ { * } ( x ) : x \in M \}. \end{equation*}
real z = 1+3*x^2;
 
real y2 = 1/(z-a-a0)+b*(1+c*x)+d;
 
dot( (x,y2) );
 
draw( (x,y2)--(x,y2)+w, dashed );
 
label("$\scriptstyle \overline f_\alpha(x)$",(x,y2)+w,SE);
 
  
label("\small Fig. a4: Non-precise function",(x0,-0.2));
+
A graphical description of non-precise functions can be given using the $\alpha$-cuts $\left[ \underline { f } \square _ { \alpha } ( x ) , \overline { f } _ { \alpha } ( x ) \right]$ of the fuzzy numbers $f ^ { * } ( x )$.
  
shipout(scale(250,120)*currentpicture);
+
The curves $x \rightarrow \underline { f } \square_{\alpha} ( x )$ and $x \rightarrow \overline { f } _ { \alpha } ( x )$ are called lower and upper $\alpha$-level curves. Taking a finite suitable number of $\alpha$-levels from $( 0,1 ]$ and depicting the corresponding $\alpha$-level curves in a diagram, a good graphical display of the non-precise function is obtained. An example is given in Fig.a4.
</asy></center>
+
{{:Non-precise data/Fig4}}
  
 
==Applications.==
 
==Applications.==
Line 259: Line 82:
  
 
====References====
 
====References====
<table><TR><TD valign="top">[a1]</TD> <TD valign="top">  "Modelling uncertain data"  H. Bandemer (ed.) , Akad. Berlin  (1993)</TD></TR><TR><TD valign="top">[a2]</TD> <TD valign="top">  "Combining fuzzy imprecision with probabilistic uncertainty in decision making"  J. Kacprzyk (ed.)  M. Fedrizzi (ed.) , ''Lecture Notes Economics and Math. Systems'' , '''310''' , Springer  (1988)</TD></TR><TR><TD valign="top">[a3]</TD> <TD valign="top">  R. Viertl,  "Statistics with non-precise data"  ''J. Comput. Inform. Techn.'' , '''4''' :  4  (1996)</TD></TR><TR><TD valign="top">[a4]</TD> <TD valign="top">  R. Viertl,  "Statistical methods for non-precise data" , CRC  (1996)</TD></TR></table>
+
<table><tr><td valign="top">[a1]</td> <td valign="top">  "Modelling uncertain data"  H. Bandemer (ed.) , Akad. Berlin  (1993)</td></tr><tr><td valign="top">[a2]</td> <td valign="top">  "Combining fuzzy imprecision with probabilistic uncertainty in decision making"  J. Kacprzyk (ed.)  M. Fedrizzi (ed.) , ''Lecture Notes Economics and Math. Systems'' , '''310''' , Springer  (1988)</td></tr><tr><td valign="top">[a3]</td> <td valign="top">  R. Viertl,  "Statistics with non-precise data"  ''J. Comput. Inform. Techn.'' , '''4''' :  4  (1996)</td></tr><tr><td valign="top">[a4]</td> <td valign="top">  R. Viertl,  "Statistical methods for non-precise data" , CRC  (1996)</td></tr></table>

Latest revision as of 17:45, 1 July 2020

Real data obtained from measurement processes are not precise numbers or vectors but are more or less non-precise. This uncertainty is different from measurement errors and has to be described formally in order to obtain realistic results by mathematical models.

A real-life example is the water level of a river at a fixed time. It is typically not a precise multiple of the scale unit for height measurements. This water level is a fuzzy quantity called non-precise. In the past this kind of uncertainty was neglected in describing such data. The reason for that is the idea of the existence of a "true" water level which is identified with a real number times the measurement unit. But this is not realistic. The formal description of such non-precise water levels can be given using the intensity of the wetness of the gauge to obtain the so-called characterizing functions (see below).

Further examples of non-precise data are readings on digital measurement equipments, readings of pointers on scales, colour intensity pictures and light points on screens.

Non-precise data are different from measurement errors because in error models the observed values $y _ { i }$ are considered to be numbers, i.e. $y _ { i } = x _ { i } + \epsilon _ { i }$, where $\epsilon_{i}$ denotes the error of the $i$th observation (cf. also Numerical analysis).

Historically, non-precise data were not studied sufficiently. Some earlier work was done in interval arithmetic. In general, non-precise sets in the form of fuzzy sets were considered by L.A. Zadeh. Some publications combining fuzzy imprecision and stochastic uncertainty came up in the 1980s, see [a2]. Some of these approaches are more theoretically oriented. An applicable approach for statistical analysis of non-precise data is given in [a4].

Characterizing functions of non-precise numbers.

In the case of measurements of one-dimensional quantities, non-precise observations can be reasonably described by so-called fuzzy numbers $x ^ { * }$. Fuzzy numbers are generalizations of real numbers in the following sense.

Each real number $X \in \mathbf R$ is characterized by its indicator function $I_{ \{ x \} } ( \cdot )$. Specializing the membership functions from fuzzy set theory, a fuzzy number is characterized by its so-called characterizing function $\xi ( . )$, which is a generalization of an indicator function. A characterizing function is a real-valued function of a real variable obeying the following:

1) $\xi : \mathbf{R} \rightarrow [ 0,1 ]$;

2) $\exists x \in \mathbf R $ such that $\xi ( x ) = 1$;

3) $\forall \alpha \in ( 0,1 ]$, the so-called $\alpha$-cut $B _ { \alpha } = \{ x \in \mathbf{R} : \xi ( x ) \geq \alpha \}$ is a closed finite interval. Characterizing functions describe the imprecision of a single observation. They should not be confused with probability densities, which describe the stochastic variation of a random quantity $X$ (cf. also Probability distribution). Examples of characterizing functions are depicted in Fig.a1.

A fundamental problem is how to obtain the characterizing function of a non-precise observation. This depends on the area of application. An example is as follows.

For data in the form of gray intensities in one dimension, as boundaries of regions the gray intensity $g ( x )$ as an increasing function of one real variable $x$ can be used to obtain the characterizing function $\xi ( x )$ in the following way. Take the derivative $( d / d x ) g ( x )$ and divide it by its maximum. The resulting function, or its convex hull, can be used as characterizing function of the non-precise observation.

In Fig.a2 the construction of the characterizing function from a gray intensity is explained.

Non-precise samples.

Taking observations of a one-dimensional continuous quantity $X$ in order to estimate the distribution of $X$, usually a finite sequence $x _ { 1 } ^ { * } , \ldots , x _ { n } ^ { * }$ of non-precise numbers is obtained. These non-precise data are given in form of $n$ characterizing functions $\xi _ { 1 } ( . ) , \ldots , \xi _ { n } ( . )$ corresponding to $x _ { 1 } ^ { * } , \ldots , x _ { n } ^ { * }$. Facing this kind of samples, even a most simple concept like the histogram has to be modified. This is necessary because, for a given class $K_j$ in a histogram, for a non-precise observation $x _ { i } ^ {\color{blue} *}$ with characterizing function $\xi _ { i }(.)$ obeying $\xi _ { i } ( x ) > 0$ for an element $x \in K_j $ and $\xi _ { i } ( y ) > 0$ for an element $y \in K _ { j } ^ { c }$, it is not possible to decide whether or not $x _ { i } ^ {\color{blue} *}$ is an element of $K_j$. The situation is explained in Fig.a3.

A generalization of the concept of histogram is possible by defining so-called fuzzy histograms. For such histograms the height of the histogram over a fixed class $K_j$ is a fuzzy number $h _ { j } ^ { * }$. For the definition of the characterizing function of $h _ { j } ^ { * }$, see [a3]. For other concepts of statistics in the case of non-precise data, see [a4].

Non-precise vectors.

In the case of multi-variate data $\underline { x } = ( x _ { 1 } , \dots , x _ { n } )$, for example the position of an object on a radar screen, the observations are non-precise vectors $\underline{x} ^ { * }$. Such non-precise vectors are characterized by so-called vector-characterizing functions $\xi _ { \underline{x}^{*}} ( . , \dots , . )$. These vector-characterizing functions $\xi _ { \underline{x}^{*}} ( . , \dots , . )$ are real-valued functions of $n$ real variables $x _ { 1 } , \ldots , x _ { n }$ obeying the following:

a) $\xi _ { \underline{x}^*} : \mathbf{R} ^ { n } \rightarrow [ 0,1 ]$;

b) $\exists \underline{x} = ( x _ { 1 } , \dots , x _ { n } ) \in \mathbf{R} ^ { n }$ such that ;

c) $\forall \alpha \in ( 0,1 ]$, the $\alpha$-cut $B _ { \alpha } ( \underline{x} ^ { * } ) = \{ \underline{x} \in \mathbf{R} ^ { n } : \xi _ { \underline{x} ^ { * } } ( \underline{x} ) \geq \alpha \}$ is a closed and star-shaped subset with finite $n$-dimensional content.

Functions of non-precise arguments.

For the generalization of functions of real variables, the so-called extension principle from fuzzy set theory is used. This principle generalizes a function to the situation when the value of the argument variable is non-precise.

Let $\psi : \mathbf{R} ^ { n } \rightarrow \mathbf{R}$ be a classical real-valued function of $n$ variables. If the argument $\underline { x } = ( x _ { 1 } , \dots , x _ { n } )$ is precise, then the value $\psi ( \underline{x} )$ of the function is also a precise real number. For a non-precise argument $\underline{x} ^ { * }$ it is natural that the value $\psi ( \underline{x} ^ { * } )$ becomes non-precise also. The quantitative description of this imprecision is done by using the characterizing function $\eta ( . )$ of $\psi ( \underline{x} ^ { * } )$. Let $\xi ( ., \dots , . )$ be the vector-characterizing function of the fuzzy vector $\underline{x} ^ { * }$. Then, for all real numbers $y$, the values $\eta ( y )$, are given by the extension principle in the following way:

\begin{equation*} \eta ( y ) = \left\{ \begin{array} { l l } { \operatorname { sup } \{ \xi ( \underline{x} ) : \underline{x} \in {\bf R} ^ { n } , \psi ( \underline{x} ) = y \} , } & { \psi ^ { - 1 } ( y ) \neq \emptyset ,} \\ { 0 , } & { \psi ^ { - 1 } ( y ) = \emptyset .} \end{array} \right. \end{equation*}

For continuous functions $\psi ( . )$ it can be proved that $\eta ( . )$ is a characterizing function, see [a4].

Non-precise functions.

Consider the monitoring of quantities in continuous time for real measurements. The results are functions of time with non-precise values. Therefore the concept of a non-precise function is necessary. These are functions whose values are fuzzy numbers. Let $M$ be the domain of the non-precise function $f ^ { * } ( . )$ and let be the set of all fuzzy numbers; then

\begin{equation*} f ^ { * } : M \rightarrow \mathcal{F} ( \mathbf{R} ). \end{equation*}

A non-precise function can also be considered as a family of fuzzy numbers with characterizing functions $\psi _ { x } ( \cdot )$, i.e.

\begin{equation*} \{ \psi _ { x } ( . ) \widehat{=} f ^ { * } ( x ) : x \in M \}. \end{equation*}

A graphical description of non-precise functions can be given using the $\alpha$-cuts $\left[ \underline { f } \square _ { \alpha } ( x ) , \overline { f } _ { \alpha } ( x ) \right]$ of the fuzzy numbers $f ^ { * } ( x )$.

The curves $x \rightarrow \underline { f } \square_{\alpha} ( x )$ and $x \rightarrow \overline { f } _ { \alpha } ( x )$ are called lower and upper $\alpha$-level curves. Taking a finite suitable number of $\alpha$-levels from $( 0,1 ]$ and depicting the corresponding $\alpha$-level curves in a diagram, a good graphical display of the non-precise function is obtained. An example is given in Fig.a4.

Applications.

Whenever measurements have to be modeled, non-precise numbers appear. This occurs in the initial conditions for differential equations, in the time-dependent description of quantities, as well as in statistical inference.

Mathematical methods for the analysis of non-precise data exist and should be used in order to obtain more realistic results from mathematical modeling.

References

[a1] "Modelling uncertain data" H. Bandemer (ed.) , Akad. Berlin (1993)
[a2] "Combining fuzzy imprecision with probabilistic uncertainty in decision making" J. Kacprzyk (ed.) M. Fedrizzi (ed.) , Lecture Notes Economics and Math. Systems , 310 , Springer (1988)
[a3] R. Viertl, "Statistics with non-precise data" J. Comput. Inform. Techn. , 4 : 4 (1996)
[a4] R. Viertl, "Statistical methods for non-precise data" , CRC (1996)
How to Cite This Entry:
Non-precise data. Encyclopedia of Mathematics. URL: http://encyclopediaofmath.org/index.php?title=Non-precise_data&oldid=35606
This article was adapted from an original article by Reinhard Viertl (originator), which appeared in Encyclopedia of Mathematics - ISBN 1402006098. See original article