CMPSCI 240: Reasoning About Uncertainty: Lecture 4: Sequential Experiments

Recap Sequential Experiments Total Probability and Bayes Theorem
CMPSCI 240: Reasoning about Uncertainty

Lecture 4: Sequential experiments
Andrew McGregor
University of Massachusetts
Last Compiled: February 2, 2017

Outline
1 Recap
2 Sequential Experiments
3 Total Probability and Bayes Theorem

Conditional Probability Laws
Conditional probability laws specify the conditional probability

P(A | B) of any event A given that we know with certainty that the
true outcome is contained in event B.
Conditional probability of event A given event B where P(B) > 0:
P(A | B) = P(A ∩ B)/P(B)
and so P(A ∩ B) = P(B)P(A|B).

Basic Properties:
1 P(A|B) ≥ 0
2 P(Ω|B) = 1
3 P(AC |B) = 1 − P(A|B)
4 If A1 , ..., AN are mutually disjoint events
P(A1 ∪ . . . ∪ AN |B) = P(A1 |B) + . . . + P(AN |B)

Outline
1 Recap

Sequential Experiments
A sequential experiment is an experiment that involves several

different steps that happen in sequence.
Each step is an event that narrows down the possible outcomes until
the final step, which assigns a unique outcome.
Probability laws for sequential experiments are often much more

intuitive to specify using conditional probability.
Example: Radar Detection
Problem: We’re in charge of air traffic control. We know that if a

plane is near the airport, the radar will correctly detect it with
probability 0.99. If a plane is not present, the radar still generates a
detection with probability 0.1. We know that an aircraft is near the
airport with probability 0.05. What is the probability that no plane
is present and the radar generates a detection?
Steps in the sequential experiment:
1 Is an aircraft present? Define event A=“an aircraft is present”
2 Does the radar generate a detection? Define event D=“radar
generates detection”
Outcomes of the sequential experiment:
1 A ∩ D: Aircraft and detection
2 A ∩ D C : Aircraft and no detection
3 AC ∩ D: No aircraft and detection
4 AC ∩ D C : No aircraft and no detection
Problem: If a plane is near the airport, the radar will correctly

detect it with probability 0.99. If a plane is not present, the radar
still generates a detection with probability 0.1. We know that an
aircraft is near the airport with probability 0.05.
Question: What probabilities can we deduce?

Answer:
1 P(D|A) = 0.99: “If a plane is near the airport, the radar will
correctly detect it with probability 0.99”
2 P(D|AC ) = 0.1: “If a plane is not present, the radar still generates a
detection with probability 0.1”
3 P(A) = 0.05: “An aircraft is near the airport with probability 0.05”
Question: If P(D|A) = 0.99, P(D|AC ) = 0.1 and P(A) = 0.05,

what is the probability that no plane is present and the radar
generates a detection? In other words, what is P(AC ∩ D)?
Answer:
P(AC ∩ D) = P(D|AC )P(AC ) = P(D|AC )(1 − P(A))

= 0.1 · 0.95 = 0.095
Question: If P(D|A) = 0.99, P(D|AC ) = 0.1 and P(A) = 0.05,

what is the probability that a plane is present and the radar does not
generate a detection?
Answer: The probability we want to compute is P(A ∩ D C ). Using

the multiplication and complement rules, we find:
P(A ∩ D C ) = P(D C |A)P(A) = (1 − P(D|A))P(A)

= 0.01 · 0.05 = 0.0005
Conditional Probability Trees for Sequential Experiments
One convenient way to describe these types of sequential

experiments is to use a conditional probability tree.
1 The steps in the experiment correspond to edges in the tree.
2 A path from the root to a leaf is an outcome.
3 The probability of any outcome can be computed by multiplying the
edge probabilities on the path from the outcome to the root.
Given the conditional probabilities P(D|A) = 0.99, P(D|AC ) = 0.1 and

P(A) = 0.05, we construct the corresponding tree as shown below.
P(A ∩ D)
D
|A)
P (D
DC
A
P (D C
) |A )
P (A
P(A ∩ D C )
Ω
P(AC ∩ D)
AC
D
P (A C
) C)
|A
P (D
DC
P (D C
|A C
)
P(AC ∩ D C )
We then substitute the numerical values derived from the probabilities

P(D|A) = 0.99, P(D|AC ) = 0.1 and P(A) = 0.05 into the tree.
0.05 · 0.99
D
0.99
DC
A
0.01
5
0 .0
0.05 · 0.01
Ω
0.95 · 0.1
AC
D
0 .9
5
0.1
DC
0.90
0.95 · 0.9
To get the probability of an outcome, just multiply the probabilities along

the corresponding path.
Extending to more steps: N-Step Multiplication Rule
Now suppose we have N events A1 , ..., AN and we’re interested in

the probability of the conjunction of these events
P(∩N
n=1 An ) = P(A1 ∩ · · · ∩ AN ) .
The N-step multiplication rule states that:

N
Y
P(∩N
n=1 An ) = P(An | ∩n−1
i=1 Ai )
n=1
= P(A1 )P(A2 |A1 )P(A3 |A1 ∩ A2 ) · · · P(AN |A1 ∩ · · · ∩ AN−1 )
N-Step Multiplication Rule
We can prove the result using repeated application of the multiplication

rule for conditional probabilities.
P(A1 )P(A2 |A1 )P(A3 |A1 ∩ A2 ) · · · P(AN |A1 ∩ · · · ∩ AN−1 )

P(A1 ∩ A2 ) P(A1 ∩ A2 ∩ A3 ) P(A1 ∩ A2 ∩ . . . AN )
= P(A1 ) · · ...
P(A1 ) P(A1 ∩ A2 ) P(A1 ∩ A2 ∩ . . . ∩ AN−1 )
= P(A1 ∩ A2 ∩ . . . AN )
Example: Pick Three Cards
Consider a card game where you draw three consecutive cards from
a deck without replacement and win if none of the cards is a heart.
Question: What is the probability that you win the game?
1 Let H1 , H2 , H3 be the events that the first, second, and third cards
that you draw are hearts.
2 The probability that we want to compute is thus P(H1C ∩ H2C ∩ H3C ).
3 Using the multiplication rule, we can compute it as
39 38 37
P(H1C )P(H2C |H1C )P(H3C |H1C ∩ H2C ) = · · ≈ 0.41
52 51 50
Example: Pick Three Cards
Consider a simple card game where you win if you draw three
consecutive cards from a deck without replacement and none of the
cards is a heart. Question: What does the conditional probability
tree look like?
H1
5 2
13/
Ω HC
1 H2
39/ 5 1
52 13/
HC
2 H3
38/ 50
51 13/
HC
3
37/
50
P(H3C ∩ H1C ∩ H2C )
Clicker Questions
Suppose we draw two cards at random without replacement from a

standard deck of cards. Let H1 and H2 be the events that the first and
second cards that you draw are hearts.
Q1: What is the probability P(H2C |H1 )?

A: 1/2 B: 38/52 C: 39/52 D: 38/51 E: 39/51
Answer: E
Q2: What is the probability P(H1 ∩ H2 )?

13 12 13 13 13 12 13 13
A: 52 × 52 B: 52 × 52 C: 52 × 51 D: 52 × 51
Answer: C
Outline
1 Recap

Total Probability and Bayes Theorem
Total Probability If A1 , . . . , An partition Ω then for any event B:

n
X
P(B) = P(B|A1 )P(A1 ) + . . . + P(B|An )P(An ) = P(B | Ai )P(Ai )
i=1
Bayes Theorem If A1 , . . . , An partition Ω then for any event B:
P(Ai ∩ B) P(B|Ai )P(Ai )

P(Ai |B) = = Pn
P(B) j=1 P(B | Aj )P(Aj )
Proof of Total Probability Theorem

Since A1 , . . . , An partition Ω then (B ∩ A1 ), . . . , (B ∩ An ) partition B.
For any Ai , the multiplication rule gives us

P(B ∩ Ai ) = P(B|Ai )P(Ai ) .
Putting these results together we have the following proof:
P(B) = P((B ∩ A1 ) ∪ · · · ∪ (B ∩ An ))
= P(B ∩ A1 ) + · · · + P(B ∩ An )
= P(B|A1 )P(A1 ) + · · · + P(B|An )P(An )

CMPSCI 240: Reasoning About Uncertainty: Lecture 4: Sequential Experiments

Uploaded by

Copyright:

Available Formats

CMPSCI 240: Reasoning About Uncertainty: Lecture 4: Sequential Experiments

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CMPSCI 240: Reasoning About Uncertainty: Lecture 4: Sequential Experiments

Uploaded by

Copyright:

Available Formats

Recap Sequential Experiments Total Probability and Bayes Theorem

CMPSCI 240: Reasoning about Uncertainty

Last Compiled: February 2, 2017

3 Total Probability and Bayes Theorem

Conditional Probability Laws

Conditional probability laws specify the conditional probability

P(A | B) = P(A ∩ B)/P(B)

and so P(A ∩ B) = P(B)P(A|B).

P(A1 ∪ . . . ∪ AN |B) = P(A1 |B) + . . . + P(AN |B)

3 Total Probability and Bayes Theorem

A sequential experiment is an experiment that involves several

Probability laws for sequential experiments are often much more

Example: Radar Detection

Problem: We’re in charge of air traffic control. We know that if a

Example: Radar Detection

Problem: If a plane is near the airport, the radar will correctly

Question: What probabilities can we deduce?

Example: Radar Detection

Question: If P(D|A) = 0.99, P(D|AC ) = 0.1 and P(A) = 0.05,

P(AC ∩ D) = P(D|AC )P(AC ) = P(D|AC )(1 − P(A))

Example: Radar Detection

Question: If P(D|A) = 0.99, P(D|AC ) = 0.1 and P(A) = 0.05,

Answer: The probability we want to compute is P(A ∩ D C ). Using

P(A ∩ D C ) = P(D C |A)P(A) = (1 − P(D|A))P(A)

Conditional Probability Trees for Sequential Experiments

One convenient way to describe these types of sequential

Example: Radar Detection

Given the conditional probabilities P(D|A) = 0.99, P(D|AC ) = 0.1 and

Example: Radar Detection

We then substitute the numerical values derived from the probabilities

To get the probability of an outcome, just multiply the probabilities along

Extending to more steps: N-Step Multiplication Rule

Now suppose we have N events A1 , ..., AN and we’re interested in

The N-step multiplication rule states that:

N-Step Multiplication Rule

We can prove the result using repeated application of the multiplication

P(A1 )P(A2 |A1 )P(A3 |A1 ∩ A2 ) · · · P(AN |A1 ∩ · · · ∩ AN−1 )

Example: Pick Three Cards

Example: Pick Three Cards

Suppose we draw two cards at random without replacement from a

Q1: What is the probability P(H2C |H1 )?

Q2: What is the probability P(H1 ∩ H2 )?

3 Total Probability and Bayes Theorem

Total Probability and Bayes Theorem

Total Probability If A1 , . . . , An partition Ω then for any event B:

Bayes Theorem If A1 , . . . , An partition Ω then for any event B:

P(Ai ∩ B) P(B|Ai )P(Ai )

Proof of Total Probability Theorem

For any Ai , the multiplication rule gives us

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.