This is a work in the tradition of de Finetti. Rather than starting with probability assessments over a field of events, it considers probability assessments, first, over a finite number of events that may exhibit logical relations, and, second, over a finite number of "conditional events'', again that may exhibit logical relations. A natural question that arises about a probability assessment over a finite set of events is: Is it coherent? A second question is: Can it be extended to include some new event? Similar questions can be asked concerning a finite set of assessments of conditional events. In addition, given an assessment over events and an assessment over conditional events, the question of the coherence of the two assessments may be raised. Linear programming algorithms are provided to answer these questions; in some cases the program will yield upper and lower limits—upper and lower probabilities—for the extension of an original assessment to new events or conditional events.
What is new in this approach is the emphasis on partial assessments, to begin with, and on the insistence that conditional events are central and irreducible. The conditional event E|H has one of three truth values: 0, 1, and p, where p is the subjective conditional probability of E|H. This allows the introduction of conditional probabilities P(E|H) where the condition has probability 0, and that in turn is used in the discussion of fuzzy set theory, possibility theory, and default logic. These discussions are limited in scope and persuasiveness, but the algorithms introduced in the main part of the book may be of interest to a number of scholars.
The English in which the book is written is not standard idiomatic English, but it has a certain charm that is unlikely to interfere with technical understanding.
proba etre entre 0 et 0,5 sachant que x= 0,22 est 1/2 et aussi 1/2 d'etre dans 0,5 à 1
GaBuZoMeu a écrit:Beagle, n'as-tu pas un peu l'impression de tourner en rond ?
Un petit détail :proba etre entre 0 et 0,5 sachant que x= 0,22 est 1/2 et aussi 1/2 d'etre dans 0,5 à 1
D'où sors-tu cela ? Si on prend la mesure de probabilité uniforme sur [0,1], alors les événements x=0,22 et x<1/2 sont indépendants. Mais ça ne veut absolument pas dire que la probabilité de x<1/2 sachant que x= 0,22 est 1/2 ; cette phrase sort de ton imagination, elle n'a absolument rien de mathématique : la probabilité de x<1/2 sachant que x= 0,22 n'existe pas vu que l'événement x=0.22 est de probabilité nulle.
Changeons maintenant de mesure de probabilité sur [0,1] : on prend la mesure qui charge chaque point k/100 avec 0<=k<=100 avec la probabilité 1/101. Alors les événements x=0.22 et X<1/2 ne sont plus indépendants, et la probabilité de x<1/2 sachant que x= 0,22 est maintenant 1
GaBuZoMeu a écrit:Tu t'imagines peut-être que quelqu'un va te dire que la probabilité de x=0.22 pour la distribution uniforme sur [0,1] n'est pas nulle ?
Peut-on espérer un discours un peu cohérent de ta part ? Ça serait sympa, merci !
Utilisateurs parcourant ce forum : Aucun utilisateur enregistré et 8 invités
Tu pars déja ?
Identification
Pas encore inscrit ?
Ou identifiez-vous :