introduction to probabilistic reasoningdagos/stellenbosch/bayes2013b.pdfintroduction to...
TRANSCRIPT
![Page 1: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/1.jpg)
Introduction toProbabilistic Reasoning
– inference, forecasting, decision –
Giulio D’Agostini
Dipartimento di Fisica
Universita di Roma La Sapienza
– Part 2 –
“Probability is good sense reduced to a calculus” (Laplace)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 1
![Page 2: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/2.jpg)
Probability
What is probability?
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 2
![Page 3: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/3.jpg)
Standard textbook definitions
p =# favorable cases
#possible equiprobable cases
p =# times the event has occurred
# independent trials under same conditions
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 3
![Page 4: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/4.jpg)
Standard textbook definitions
It is easy to check that ‘scientific’ definitions suffer ofcircularity
p =# favorable cases
#possible equiprobable cases
p =# times the event has occurred
# independent trials under same conditions
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 3
![Page 5: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/5.jpg)
Standard textbook definitions
It is easy to check that ‘scientific’ definitions suffer ofcircularity
p =# favorable cases
#possible equally possible cases
p =# times the event has occurred
# independent trials under same conditions
Laplace: “lorsque rien ne porte à croire que l’un de ces cas doitarriver plutot que les autres”
Pretending that replacing ‘equi-probable’ by‘equi-possible’ is just cheating students (as I did in myfirst lecture on the subject. . . ).
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 3
![Page 6: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/6.jpg)
Standard textbook definitions
It is easy to check that ‘scientific’ definitions suffer ofcircularity, plus other problems
p =# favorable cases
#possible equiprobable cases
p = limn→∞# times the event has occurred
# independent trials under same condition
Future⇔ Past (believedso)
n→∞: → “usque tandem?”→ “in the long run we are all dead”→ It limits the range of applicationsG. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 3
![Page 7: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/7.jpg)
Definitions→ evaluation rules
Very useful evaluation rules
A) p =# favorable cases
#possible equiprobable cases
B) p =# times the event has occurred
#independent trials under same condition
If the implicit beliefs are well suited for each case ofapplication.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 4
![Page 8: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/8.jpg)
Definitions→ evaluation rules
Very useful evaluation rules
A) p =# favorable cases
#possible equiprobable cases
B) p =# times the event has occurred
#independent trials under same condition
If the implicit beliefs are well suited for each case ofapplication.
BUT they cannot define the concept of probability!
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 4
![Page 9: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/9.jpg)
Definitions→ evaluation rules
Very useful evaluation rules
A) p =# favorable cases
#possible equiprobable cases
B) p =# times the event has occurred
#independent trials under same condition
If the implicit beliefs are well suited for each case ofapplication.
In the probabilistic approach we are going to see
Rule A will be recovered immediately (under theassumption of equiprobability, when it applies).
Rule B will result from a theorem (under well definedassumptions). G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 4
![Page 10: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/10.jpg)
Probability
What is probability?
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 5
![Page 11: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/11.jpg)
Probability
What is probability?
It is what everybody knows what it isbefore going at school
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 5
![Page 12: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/12.jpg)
Probability
What is probability?
It is what everybody knows what it isbefore going at school
→ how much we are confident thatsomething is true
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 5
![Page 13: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/13.jpg)
Probability
What is probability?
It is what everybody knows what it isbefore going at school
→ how much we are confident thatsomething is true
→ how much we believe something
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 5
![Page 14: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/14.jpg)
Probability
What is probability?
It is what everybody knows what it isbefore going at school
→ how much we are confident thatsomething is true
→ how much we believe something
→ “A measure of the degree of belief thatan event will occur”
[Remark: ‘will’ does not imply future, but onlyuncertainty.]
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 5
![Page 15: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/15.jpg)
An helpful diagram
The previous diagram seems to help the understanding ofthe concept of probability
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 6
![Page 16: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/16.jpg)
An helpful diagram
(. . . but NASA guys are afraid of ‘subjective’, or ‘psychological’)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 7
![Page 17: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/17.jpg)
Beliefs and ‘coherent’ bets
Remarks:
Subjective does not mean arbitrary!
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 8
![Page 18: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/18.jpg)
Beliefs and ‘coherent’ bets
Remarks:
Subjective does not mean arbitrary!
How to force people to assess how much they areconfident on something?
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 8
![Page 19: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/19.jpg)
Beliefs and ‘coherent’ bets
Remarks:
Subjective does not mean arbitrary!
How to force people to assess how much they areconfident on something?
“The usual touchstone, whether that which someone asserts
is merely his persuasion – or at least his subjective
conviction, that is, his firm belief – is betting. It often happens
that someone propounds his views with such positive and
uncompromising assurance that he seems to have entirely
set aside all thought of possible error. A bet disconcerts him.
Sometimes it turns out that he has a conviction which can be
estimated at a value of one ducat, but not of ten. For he is
very willing to venture one ducat, but when it is a question of
ten he becomes aware, as he had not previously been, that it
may very well be that he is in error.” (Kant)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 8
![Page 20: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/20.jpg)
Beliefs and ‘coherent’ bets
Remarks:
Subjective does not mean arbitrary!
How to force people to assess how much they areconfident on something?
1 X 2G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 8
![Page 21: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/21.jpg)
Beliefs and ‘coherent’ bets
Remarks:
Subjective does not mean arbitrary!
How to force people to assess how much they areconfident on something?
Coherent bet:
you state the odds according on your beliefs;
somebody else will choose the direction of the bet.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 8
![Page 22: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/22.jpg)
Beliefs and ‘coherent’ bets
Remarks:
Subjective does not mean arbitrary!
How to force people to assess how much they areconfident on something?
Coherent bet:
you state the odds according on your beliefs;
somebody else will choose the direction of the bet.
“His [Bouvard] calculations give him the mass ofSaturn as 3,512th part of that of the sun. Applyingmy probabilistic formulae to these observations, Ifind that the odds are 11,000 to 1 that the error inthis result is not a hundredth of its value.” (Laplace)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 8
![Page 23: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/23.jpg)
Beliefs and ‘coherent’ bets
Remarks:
Subjective does not mean arbitrary!
How to force people to assess how much they areconfident on something?
Coherent bet:
you state the odds according on your beliefs;
somebody else will choose the direction of the bet.
“His [Bouvard] calculations give him the mass ofSaturn as 3,512th part of that of the sun. Applyingmy probabilistic formulae to these observations, Ifind that the odds are 11,000 to 1 that the error inthis result is not a hundredth of its value.” (Laplace)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 8
![Page 24: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/24.jpg)
Beliefs and ‘coherent’ bets
Remarks:
Subjective does not mean arbitrary!
How to force people to assess how much they areconfident on something?
Coherent bet:
you state the odds according on your beliefs;
somebody else will choose the direction of the bet.
“His [Bouvard] calculations give him the mass ofSaturn as 3,512th part of that of the sun. Applyingmy probabilistic formulae to these observations, Ifind that the odds are 11,000 to 1 that the error inthis result is not a hundredth of its value.” (Laplace)
→ P (3477 ≤MSun/MSat ≤ 3547 | I(Laplace)) = 99.99%
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 8
![Page 25: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/25.jpg)
‘C.L.’ Vs Degree of Confidence
Is a ‘conventional’ 95% C.L. lower/upperbound a 19 to 1 bet?
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 9
![Page 26: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/26.jpg)
‘C.L.’ Vs Degree of Confidence
Is a ‘conventional’ 95% C.L. lower/upperbound a 19 to 1 bet?
NO!
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 9
![Page 27: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/27.jpg)
‘C.L.’ Vs Degree of Confidence
Is a ‘conventional’ 95% C.L. lower/upperbound a 19 to 1 bet?
NO!
It does not imply one has to be 95% confident onsomething!If you do so you are going to make a bad bet!
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 9
![Page 28: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/28.jpg)
‘C.L.’ Vs Degree of Confidence
Is a ‘conventional’ 95% C.L. lower/upperbound a 19 to 1 bet?
NO!
It does not imply one has to be 95% confident onsomething!If you do so you are going to make a bad bet!
For more on the subject:http://arxiv.org/abs/1112.3620
http://www.roma1.infn.it/~dagos/badmath/#added)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 9
![Page 29: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/29.jpg)
Mathematics of beliefs
The good news:
The basic laws of degrees of beliefare the same we get from theinventory of favorable and possiblecases, or from events occurred in thepast.[ Details skipped. . . ]
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 10
![Page 30: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/30.jpg)
Basic rules of probability
1. 0 ≤ P (A | I) ≤ 1
2. P (Ω | I) = 1
3. P (A ∪ B | I) = P (A | I) + P (B | I) [ if P (A ∩ B | I) = ∅ ]
4. P (A ∩ B | I) = P (A |B, I) · P (B | I) = P (B |A, I) · P (A | I)
Remember that probability is always conditional probability!
I is the background condition (related to information ‘I ′s)
→ usually implicit (we only care on ‘re-conditioning’)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 11
![Page 31: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/31.jpg)
Basic rules of probability
1. 0 ≤ P (A | I) ≤ 1
2. P (Ω | I) = 1
3. P (A ∪ B | I) = P (A | I) + P (B | I) [ if P (A ∩ B | I) = ∅ ]
4. P (A ∩ B | I) = P (A |B, I) · P (B | I) = P (B |A, I) · P (A | I)
Remember that probability is always conditional probability!
I is the background condition (related to information ‘I ′s)
→ usually implicit (we only care on ‘re-conditioning’)
Note: 4. does not define conditional probability.(Probability is always conditional probability!)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 11
![Page 32: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/32.jpg)
Mathematics of beliefs
An even better news:
The fourth basic rulecan be fully exploided!
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 12
![Page 33: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/33.jpg)
Mathematics of beliefs
An even better news:
The fourth basic rulecan be fully exploided!
(Liberated by a curious ideology that forbits its use)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 12
![Page 34: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/34.jpg)
A simple, powerful formula
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 13
![Page 35: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/35.jpg)
A simple, powerful formula
P (A |B | I)P (B | I) = P (B |A, I)P (A | I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 13
![Page 36: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/36.jpg)
A simple, powerful formula
Take the courage to use it!G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 13
![Page 37: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/37.jpg)
A simple, powerful formula
It’s easy if you try. . . !G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 13
![Page 38: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/38.jpg)
Telling it with Gauss’ words
A quote from the Princeps Mathematicorum (Prince ofMathematicians) is a must in this town and in this place.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 14
![Page 39: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/39.jpg)
Telling it with Gauss’ words
A quote from the Princeps Mathematicorum (Prince ofMathematicians) is a must in this town and in this place.
P (Ci |data) =P (data |Ci)
P (data)P0(Ci)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 14
![Page 40: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/40.jpg)
Telling it with Gauss’ words
A quote from the Princeps Mathematicorum (Prince ofMathematicians) is a must in this town and in this place.
P (Ci |data) =P (data |Ci)
P (data)P0(Ci)
”post illa observationes” “ante illa observationes”
(Gauss)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 14
![Page 41: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/41.jpg)
Bayes formulae
The essence is all contained in the fourth basic rule ofprobability theory:
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 15
![Page 42: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/42.jpg)
Bayes formulae
The essence is all contained in the fourth basic rule ofprobability theory:
P (Ci |E, I)
P (Ci | I)=
P (E |Ci, I)
P (E | I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 15
![Page 43: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/43.jpg)
Bayes formulae
The essence is all contained in the fourth basic rule ofprobability theory:
P (Ci |E, I)
P (Ci | I)=
P (E |Ci, I)
P (E | I)
P (Ci |E, I) =P (E |Cj , I)
P (E | I) P (Ci | I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 15
![Page 44: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/44.jpg)
Bayes formulae
The essence is all contained in the fourth basic rule ofprobability theory:
P (Ci |E, I)
P (Ci | I)=
P (E |Ci, I)
P (E | I)
P (Ci |E, I) =P (E |Cj , I)
P (E | I) P (Ci | I)
P (Ci |E | I) =P (E |Ci | I) · P (Ci | I)
∑
k P (E |Ck, I) · P (Ck | I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 15
![Page 45: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/45.jpg)
Bayes formulae
The essence is all contained in the fourth basic rule ofprobability theory:
P (Ci |E, I)
P (Ci | I)=
P (E |Ci, I)
P (E | I)
P (Ci |E, I) =P (E |Cj , I)
P (E | I) P (Ci | I)
P (Ci |E | I) =P (E |Ci | I) · P (Ci | I)
∑
k P (E |Ck, I) · P (Ck | I)P (Ci |E, I) ∝ P (E |Ci, I) · P (Ci | I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 15
![Page 46: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/46.jpg)
Bayes formulae
The essence is all contained in the fourth basic rule ofprobability theory:
P (Ci |E, I)
P (Ci | I)=
P (E |Ci, I)
P (E | I)
P (Ci |E, I) =P (E |Cj , I)
P (E | I) P (Ci | I)
P (Ci |E | I) =P (E |Ci | I) · P (Ci | I)
∑
k P (E |Ck, I) · P (Ck | I)P (Ci |E, I) ∝ P (E |Ci, I) · P (Ci | I)
or even (my preferred form to grasp its meaning):
P (Ci |E | I)P (Cj |E | I)
=P (E |Ci | I)P (E |Cj | I)
· P (Ci | I)P (Cj | I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 15
![Page 47: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/47.jpg)
Bayesian parametric inference
If we want to infer a continuous parameter p from a set ofdata
→ straghtforwad extension to probability density functions(pdf)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 16
![Page 48: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/48.jpg)
Bayesian parametric inference
If we want to infer a continuous parameter p from a set ofdata
→ straghtforwad extension to probability density functions(pdf)
f (p | data, I) ∝ f (data | p, I) · f (p | I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 16
![Page 49: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/49.jpg)
Bayesian parametric inference
If we want to infer a continuous parameter p from a set ofdata
→ straghtforwad extension to probability density functions(pdf)
f (p | data, I) ∝ f (data | p, I) · f (p | I)
f (p | data, I) =f (data | p, I) · f (p | I)
∫
p f (data | p, I) · f (p | I) dp
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 16
![Page 50: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/50.jpg)
Application to the six box problem
H0 H1 H2 H3 H4 H5
Remind:
E1 = White
E2 = Black
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 17
![Page 51: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/51.jpg)
Collecting the pieces of information we need
Our tool:
P (Hj |Ei, I) =P (Ei |Hj , I)P (Ei | I)
P (Hj | I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 18
![Page 52: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/52.jpg)
Collecting the pieces of information we need
Our tool:
P (Hj |Ei, I) =P (Ei |Hj , I)P (Ei | I)
P (Hj | I)
P (Hj | I) = 1/6
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 18
![Page 53: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/53.jpg)
Collecting the pieces of information we need
Our tool:
P (Hj |Ei, I) =P (Ei |Hj , I)P (Ei | I)
P (Hj | I)
P (Hj | I) = 1/6
P (Ei | I) = 1/2
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 18
![Page 54: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/54.jpg)
Collecting the pieces of information we need
Our tool:
P (Hj |Ei, I) =P (Ei |Hj , I)P (Ei | I)
P (Hj | I)
P (Hj | I) = 1/6
P (Ei | I) = 1/2
P (Ei |Hj , I) :
P (E1 |Hj , I) = j/5
P (E2 |Hj , I) = (5− j)/5
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 18
![Page 55: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/55.jpg)
Collecting the pieces of information we need
Our tool:
P (Hj |Ei, I) =P (Ei |Hj , I)P (Ei | I)
P (Hj | I)
P (Hj | I) = 1/6
P (Ei | I) = 1/2
P (Ei |Hj , I) :
P (E1 |Hj , I) = j/5
P (E2 |Hj , I) = (5− j)/5
Our prior belief about Hj
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 18
![Page 56: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/56.jpg)
Collecting the pieces of information we need
Our tool:
P (Hj |Ei, I) =P (Ei |Hj , I)P (Ei | I)
P (Hj | I)
P (Hj | I) = 1/6
P (Ei | I) = 1/2
P (Ei |Hj , I) :
P (E1 |Hj , I) = j/5
P (E2 |Hj , I) = (5− j)/5
Probability of Ei under a well defined hypothesis Hj
It corresponds to the ‘response of the apparatus inmeasurements.→ likelihood (traditional, rather confusing name!)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 18
![Page 57: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/57.jpg)
Collecting the pieces of information we need
Our tool:
P (Hj |Ei, I) =P (Ei |Hj , I)P (Ei | I)
P (Hj | I)
P (Hj | I) = 1/6
P (Ei | I) = 1/2
P (Ei |Hj , I) :
P (E1 |Hj , I) = j/5
P (E2 |Hj , I) = (5− j)/5
Probability of Ei taking account all possible Hj
→ How much we are confident that Ei will occur.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 18
![Page 58: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/58.jpg)
Collecting the pieces of information we need
Our tool:
P (Hj |Ei, I) =P (Ei |Hj , I)P (Ei | I)
P (Hj | I)
P (Hj | I) = 1/6
P (Ei | I) = 1/2
P (Ei |Hj , I) :
P (E1 |Hj , I) = j/5
P (E2 |Hj , I) = (5− j)/5
Probability of Ei taking account all possible Hj
→ How much we are confident that Ei will occur.We can rewrite it asP (Ei | I) =
∑
j P (Ei |Hj , I) · P (Hj | I)G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 18
![Page 59: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/59.jpg)
We are ready
Now that we have set up our formalism, let’s play a little
analyse real data
some simulations
Then
Hj ←→ j ←→ pj
extending p to a continuum:
⇒ Bayes’ billiard
(prototype for all questions related to efficiencies,branching ratios)
On the meaning of p
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 19
![Page 60: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/60.jpg)
Which box? Which ball?
Inferential/forecasting history:
1. k = 0P0(Hj) = P (Hj | I0) (priors)
2. begin loop:
k = k + 1⇒ E(k) (k-th extraction)
3. Pk(Hj | Ik) ∝ P (E(k) |Hj)× Pk−1(Hj | Ik)
Pk(Ei | Ik) =∑
j P (Ei |Hj) · Pk(Hj | Ik)4. → go to 2
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 20
![Page 61: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/61.jpg)
Which box? Which ball?
Inferential/forecasting history:
1. k = 0P0(Hj) = P (Hj | I0) (priors)
2. begin loop:
k = k + 1⇒ E(k) (k-th extraction)
3. Pk(Hj | Ik) ∝ P (E(k) |Hj)× Pk−1(Hj | Ik)
Pk(Ei | Ik) =∑
j P (Ei |Hj) · Pk(Hj | Ik)4. → go to 2
Let’s play!G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 20
![Page 62: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/62.jpg)
Bayes’ billiard
This is the original problem in the theory of chances solvedby Thomas Bayes in late ’700:
imagine you roll a ball at random on a billiard;
you mark the relative position of the ball along thebilliard’s length (l/L) and remove the ball
then you roll at random other balls
write down if it stopped left or right of the first ball;
remove it and go on with n balls.
Somebody has to guess the position of the first ballknowing only how mane balls stopped left and howmany stoppe right
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 21
![Page 63: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/63.jpg)
Bayes’ billiard and Bernoulli trials
It is easy to recongnize the analogy:
Left/Right→ Success/Failure
if Left↔ Success:
l/L↔ p of binomial (Bernoulli trials)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 22
![Page 64: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/64.jpg)
Bayes’ billiard and Bernoulli trials
It is easy to recongnize the analogy:
Left/Right→ Success/Failure
if Left↔ Success:
l/L↔ p of binomial (Bernoulli trials)
Solution with modern notation:Imagine a sequence S, S, F, S, . . . [f0 is uniform]:
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 22
![Page 65: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/65.jpg)
Bayes’ billiard and Bernoulli trials
It is easy to recongnize the analogy:
Left/Right→ Success/Failure
if Left↔ Success:
l/L↔ p of binomial (Bernoulli trials)
Solution with modern notation:Imagine a sequence S, S, F, S, . . . [f0 is uniform]:
f(p |S) ∝ f(S | p) = p
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 22
![Page 66: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/66.jpg)
Bayes’ billiard and Bernoulli trials
It is easy to recongnize the analogy:
Left/Right→ Success/Failure
if Left↔ Success:
l/L↔ p of binomial (Bernoulli trials)
Solution with modern notation:Imagine a sequence S, S, F, S, . . . [f0 is uniform]:
f(p |S) ∝ f(S | p) = p
f(p |S, S) ∝ f(S | p) · f(p |S) = p2
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 22
![Page 67: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/67.jpg)
Bayes’ billiard and Bernoulli trials
It is easy to recongnize the analogy:
Left/Right→ Success/Failure
if Left↔ Success:
l/L↔ p of binomial (Bernoulli trials)
Solution with modern notation:Imagine a sequence S, S, F, S, . . . [f0 is uniform]:
f(p |S) ∝ f(S | p) = p
f(p |S, S) ∝ f(S | p) · f(p |S) = p2
f(p |S, S, F ) ∝ f(F | p) · f(p |S, S) = p2(1− p)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 22
![Page 68: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/68.jpg)
Bayes’ billiard and Bernoulli trials
It is easy to recongnize the analogy:
Left/Right→ Success/Failure
if Left↔ Success:
l/L↔ p of binomial (Bernoulli trials)
Solution with modern notation:Imagine a sequence S, S, F, S, . . . [f0 is uniform]:
f(p |S) ∝ f(S | p) = p
f(p |S, S) ∝ f(S | p) · f(p |S) = p2
f(p |S, S, F ) ∝ f(F | p) · f(p |S, S) = p2(1− p)
. . . . . .
f(p |#S,#F ) ∝ p#S(1− p)#F = p#S(1− p)(1−#s)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 22
![Page 69: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/69.jpg)
Bayes’ billiard and Bernoulli trials
It is easy to recongnize the analogy:
Left/Right→ Success/Failure
if Left↔ Success:
l/L↔ p of binomial (Bernoulli trials)
Solution with modern notation:Imagine a sequence S, S, F, S, . . . [f0 is uniform]:
f(p |S) ∝ f(S | p) = p
f(p |S, S) ∝ f(S | p) · f(p |S) = p2
f(p |S, S, F ) ∝ f(F | p) · f(p |S, S) = p2(1− p)
. . . . . .
f(p |#S,#F ) ∝ p#S(1− p)#F = p#S(1− p)(1−#s)
f(p |x, n) ∝ px(1− p)(n−x) [x = #S]
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 22
![Page 70: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/70.jpg)
Inferring the Binomial p
f(p |x, n,B) = (n+1)!x! (n−x)! p
x (1− p)n−x ,
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 23
![Page 71: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/71.jpg)
Inferring the Binomial p
f(p |x, n,B) = (n+1)!x! (n−x)! p
x (1− p)n−x ,
E(p) =x+ 1
n+ 2Laplace’s rule of successions
Var(p) =(x+ 1)(n− x+ 1)
(n+ 3)(n+ 2)2
= E(p) (1− E(p))1
n+ 3.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 23
![Page 72: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/72.jpg)
Interpretation of E(p)
Think at any future event Ei>n
⇒ if we were sure of p, then our confidence on Ei>n will beexactly p, i.e.
P (Ei | p) = p .
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 24
![Page 73: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/73.jpg)
Interpretation of E(p)
Think at any future event Ei>n
⇒ if we were sure of p, then our confidence on Ei>n will beexactly p, i.e.
P (Ei | p) = p .
But we are uncertain about p.
How much should we believe Ei>n?.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 24
![Page 74: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/74.jpg)
Interpretation of E(p)
Think at any future event Ei>n
⇒ if we were sure of p, then our confidence on Ei>n will beexactly p, i.e.
P (Ei | p) = p .
But we are uncertain about p.
How much should we believe Ei>n?.
P (Ei>n |x, n,B) =
∫ 1
0
P (Ei | p) f(p |x, n,B)dp
=
∫ 1
0
p f(p |x, n,B)dp
= E(p)
=x+ 1
n+ 2(for uniform prior) .
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 24
![Page 75: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/75.jpg)
From frequencies to probabilities
E(p) =x+ 1
n+ 2Laplace’s rule of successions
Var(p) = E(p) (1− E(p))1
n+ 3.
For ‘large’ n, x and n− x: asymptotic behaviors of f(p):
E(p) ≈ pm =x
n[with pm mode of f(p)]
σp ≈√
pm (1− pm)
n−−−→n→∞
0
p ∼ N (pm, σp) .
Under these conditions the frequentistic “definition”(evaluation rule!) of probability (x/n) is recovered.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 25
![Page 76: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/76.jpg)
Special case with x = 0
f(p | 0, n,B) = (n+ 1) (1− p)n
F (p | 0, n,B) = 1− (1− p)n+1
pm = 0
E(p) =1
n+ 2−→ 1
n
σ(p) =
√
(n+ 1)
(n+ 3)(n+ 2)2−→ 1
n
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 26
![Page 77: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/77.jpg)
Special case with x = 0
f(p | 0, n,B) = (n+ 1) (1− p)n
F (p | 0, n,B) = 1− (1− p)n+1
pm = 0
E(p) =1
n+ 2−→ 1
n
σ(p) =
√
(n+ 1)
(n+ 3)(n+ 2)2−→ 1
n
P (p ≤ pu | 0, n,B) = 95%
⇒ pu = 1− n+1√0.05 :
Probabilistic upper bound
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 26
![Page 78: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/78.jpg)
Special case with x = 0
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 26
![Page 79: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/79.jpg)
Special case with x = 0
0.001 0.005 0.050 0.500
0.1
0.2
0.5
1.0
p
F(p
)
n=1n=10n=100n=1000
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 26
![Page 80: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/80.jpg)
Special case with x = 0
0.001 0.005 0.050 0.500
0.1
0.2
0.5
1.0
p
F(p
)
n=1n=10n=100n=1000
0.95
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 26
![Page 81: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/81.jpg)
Special case with x = 0
For the case x = n
(like ‘observing’ a 100% efficiency):
→ just reason on the complementary
parameter
q = 1− p
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 26
![Page 82: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/82.jpg)
Continuing the game
We have seen ho to tackle with a single idea problems thatare treated difefrently in ‘standard statistics’:
comparing hypotheses
parametric inference
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 27
![Page 83: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/83.jpg)
Continuing the game
We have seen ho to tackle with a single idea problems thatare treated difefrently in ‘standard statistics’:
comparing hypotheses
parametric inference
You can continue the game
playing with other models of f(data | p, I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 27
![Page 84: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/84.jpg)
Continuing the game
We have seen ho to tackle with a single idea problems thatare treated difefrently in ‘standard statistics’:
comparing hypotheses
parametric inference
You can continue the game
playing with other models of f(data | p, I)make a simultaneous inference on several parameters
→ f(p1, p2, . . . |data, I)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 27
![Page 85: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/85.jpg)
Continuing the game
We have seen ho to tackle with a single idea problems thatare treated difefrently in ‘standard statistics’:
comparing hypotheses
parametric inference
You can continue the game
playing with other models of f(data | p, I)make a simultaneous inference on several parameters
→ f(p1, p2, . . . |data, I)
take into account for systematics
→ “integrating over subsamples of I”
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 27
![Page 86: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/86.jpg)
Continuing the game
We have seen ho to tackle with a single idea problems thatare treated difefrently in ‘standard statistics’:
comparing hypotheses
parametric inference
You can continue the game
playing with other models of f(data | p, I)make a simultaneous inference on several parameters
→ f(p1, p2, . . . |data, I)
take into account for systematics
→ “integrating over subsamples of I”
etc.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 27
![Page 87: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/87.jpg)
Continuing the game
We have seen ho to tackle with a single idea problems thatare treated difefrently in ‘standard statistics’:
comparing hypotheses
parametric inference
You can continue the game. . . although at a certain point you need to facecomputational issues
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 27
![Page 88: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/88.jpg)
Continuing the game
We have seen ho to tackle with a single idea problems thatare treated difefrently in ‘standard statistics’:
comparing hypotheses
parametric inference
You can continue the game. . . although at a certain point you need to facecomputational issues
→ that was of the main reason why Laplace’s methodswere set apart and frequentistics methods fourished
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 27
![Page 89: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/89.jpg)
Continuing the game
We have seen ho to tackle with a single idea problems thatare treated difefrently in ‘standard statistics’:
comparing hypotheses
parametric inference
You can continue the game. . . although at a certain point you need to facecomputational issues
→ that was of the main reason why Laplace’s methodswere set apart and frequentistics methods fourished
But we can now benefit of powerful computers andimpressive improvements in computation methods
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 27
![Page 90: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/90.jpg)
Continuing the game
We have seen ho to tackle with a single idea problems thatare treated difefrently in ‘standard statistics’:
comparing hypotheses
parametric inference
You can continue the game. . . although at a certain point you need to facecomputational issues
→ that was of the main reason why Laplace’s methodswere set apart and frequentistics methods fourished
But we can now benefit of powerful computers andimpressive improvements in computation methods
We have no longer excuses!!G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 27
![Page 91: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/91.jpg)
OK, . . . but the priors?
Priors are an important ingredient of the framework:
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 28
![Page 92: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/92.jpg)
OK, . . . but the priors?
Priors are an important ingredient of the framework:They are crucial in the Bayes theorem:
there is no other way to perform a probabilisticinference without passing through priors
. . . although they can be often so vague to beignored.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 28
![Page 93: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/93.jpg)
OK, . . . but the priors?
Priors are an important ingredient of the framework:They are crucial in the Bayes theorem:
there is no other way to perform a probabilisticinference without passing through priors
. . . although they can be often so vague to beignored.
They allow us to use consistently all pieces of priorinformation. And we all have much prior information inour job!Only the perfect idiot hase no priors
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 28
![Page 94: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/94.jpg)
OK, . . . but the priors?
Priors are an important ingredient of the framework:They are crucial in the Bayes theorem:
there is no other way to perform a probabilisticinference without passing through priors
. . . although they can be often so vague to beignored.
They allow us to use consistently all pieces of priorinformation. And we all have much prior information inour job!Only the perfect idiot hase no priors
Mistrust all prior-free methods that pretend to providenumbers that should mean how you have to beconfident in something.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 28
![Page 95: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/95.jpg)
OK, . . . but the priors?
Priors are an important ingredient of the framework:They are crucial in the Bayes theorem:
there is no other way to perform a probabilisticinference without passing through priors
. . . although they can be often so vague to beignored.
They allow us to use consistently all pieces of priorinformation. And we all have much prior information inour job!Only the perfect idiot hase no priors
Mistrust all prior-free methods that pretend to providenumbers that should mean how you have to beconfident in something.(Diffidate chi vi promette di far germogliar zecchini nelCampo dei Miracoli! – Pinocchio docet)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 28
![Page 96: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/96.jpg)
Good reasoning Vs ’prohibitive calculations’
The main reasons why the so called Bayesian reasoninghab been blooming in the last decades are related to
computational power
helped from brilliant ideas on how to use it(but sterile without modern computers!)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 29
![Page 97: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/97.jpg)
Good reasoning Vs ’prohibitive calculations’
The main reasons why the so called Bayesian reasoninghab been blooming in the last decades are related to
computational power
helped from brilliant ideas on how to use it(but sterile without modern computers!)
→ many frequentistic ideas had their raison d’être in thecomputational barrier (and many simplified – oftensimplistic – methods were ingeniously worked out)
→ no longer an excuse!
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 29
![Page 98: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/98.jpg)
Good reasoning Vs ’prohibitive calculations’
The main reasons why the so called Bayesian reasoninghab been blooming in the last decades are related to
computational power
helped from brilliant ideas on how to use it(but sterile without modern computers!)
→ many frequentistic ideas had their raison d’être in thecomputational barrier (and many simplified – oftensimplistic – methods were ingeniously worked out)
instead of methods that require integrals, methods havebeen ’invented’ that only require derivatives⇒ Maximum Likelihood
→ no longer an excuse!
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 29
![Page 99: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/99.jpg)
Good reasoning Vs ’prohibitive calculations’
The main reasons why the so called Bayesian reasoninghab been blooming in the last decades are related to
computational power
helped from brilliant ideas on how to use it(but sterile without modern computers!)
→ many frequentistic ideas had their raison d’être in thecomputational barrier (and many simplified – oftensimplistic – methods were ingeniously worked out)
instead of methods that require integrals, methods havebeen ’invented’ that only require derivatives⇒ Maximum Likelihood . . . often a good approximatedsolution under some assumptions usually unkown topractitioners.
→ no longer an excuse!
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 29
![Page 100: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/100.jpg)
Conclusions from Part - 2
The probabilistic framework basically set up by Laplacein his monumental work is healthy and grows up well(browse e.g. Amazon.com)
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 30
![Page 101: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/101.jpg)
Conclusions from Part - 2
The probabilistic framework basically set up by Laplacein his monumental work is healthy and grows up well(browse e.g. Amazon.com)
It is very close to the natural way of reasoning ofphysicists (as of everybody else).
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 30
![Page 102: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/102.jpg)
Conclusions from Part - 2
The probabilistic framework basically set up by Laplacein his monumental work is healthy and grows up well(browse e.g. Amazon.com)
It is very close to the natural way of reasoning ofphysicists (as of everybody else).
Its consistent application in small-complex problemswas prohibitive many years ago.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 30
![Page 103: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/103.jpg)
Conclusions from Part - 2
The probabilistic framework basically set up by Laplacein his monumental work is healthy and grows up well(browse e.g. Amazon.com)
It is very close to the natural way of reasoning ofphysicists (as of everybody else).
Its consistent application in small-complex problemswas prohibitive many years ago.
But it is now possible thank to progresses in appliedmathematics and computation.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 30
![Page 104: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/104.jpg)
Conclusions from Part - 2
The probabilistic framework basically set up by Laplacein his monumental work is healthy and grows up well(browse e.g. Amazon.com)
It is very close to the natural way of reasoning ofphysicists (as of everybody else).
Its consistent application in small-complex problemswas prohibitive many years ago.
But it is now possible thank to progresses in appliedmathematics and computation.
It makes little sense to stick to old ‘ah hoc’ methods thathad their raison d’être in the computational barrier.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 30
![Page 105: Introduction to Probabilistic Reasoningdagos/stellenbosch/bayes2013b.pdfIntroduction to Probabilistic Reasoning – inference, forecasting, decision – Giulio D’Agostini giulio.dagostini@roma1.infn.it](https://reader030.vdocuments.site/reader030/viewer/2022040810/5e5118a18b5bc136512c755d/html5/thumbnails/105.jpg)
Conclusions from Part - 2
The probabilistic framework basically set up by Laplacein his monumental work is healthy and grows up well(browse e.g. Amazon.com)
It is very close to the natural way of reasoning ofphysicists (as of everybody else).
Its consistent application in small-complex problemswas prohibitive many years ago.
But it is now possible thank to progresses in appliedmathematics and computation.
It makes little sense to stick to old ‘ah hoc’ methods thathad their raison d’être in the computational barrier.
Mistrust all results that sound as ‘confidence’,’probability’ etc about physics quantities, if they areobtained by methods that do not contemplate ’beliefs’.
G. D’Agostini, Probabilistic Reasoning [2] (Stellenbosch, 23 November 2013) – c© G. D’Agostini – p. 30