flipping the dicejeremy orlo , jonathan bloom (mit math) flipping the dice june 23, 2014 11 / 19...
TRANSCRIPT
Flipping the Dice
Jeremy Orloff and Jonathan Bloom
Mathematics Department, MIT
[email protected] [email protected]
June 23, 2014
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 1 / 19
Bayesian dice demo
We’ve placed the five Platonic dice shown in a cup.
We need a volunteer to pick one at random.
What if the cup contained one thousand 20-sided dice?
Goal: make our intuition/learning mathematically precise.
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 2 / 19
Bayesian dice demo
We’ve placed the five Platonic dice shown in a cup.
We need a volunteer to pick one at random.
What if the cup contained one thousand 20-sided dice?
Goal: make our intuition/learning mathematically precise.
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 2 / 19
Bayesian dice demo
We’ve placed the five Platonic dice shown in a cup.
We need a volunteer to pick one at random.
What if the cup contained one thousand 20-sided dice?
Goal: make our intuition/learning mathematically precise.
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 2 / 19
Review of probability
Find the probability of each of the following events, assumingstandard (6-sided) dice are used.
Roll one die:1 Roll a 12 Roll a 43 Roll a 7
Roll two dice:1 Sum of two rolls is 42 Sum of two rolls is 7
We write p(roll 1) for the “probability of rolling a 1”.
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 3 / 19
Conditional probability
If we have many types of dice and we want to make it clear which onewe picked to roll, we write
p(roll 1 | pick 6-sided)
to mean “probability of rolling a 1 given that I picked a 6-sided die.”
Find:1 p(roll 1 | pick 4-sided)2 p(roll 5 | pick 4-sided)3 p(roll 5 | pick 6-sided)4 p(roll 5 | pick 20-sided)
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 4 / 19
Multiplication Rule
p(pick 6-sided and roll 1) = p(pick 6-sided) p(roll 1 | pick 6-sided)
P(A and B) = P(B)P(A|B)
Example. Suppose you have 2 four-sided dice and 3 six-sided dice.
Pick 6-sided die
Roll a 1 with 6-sided die
p(pick 6) = 3/5
(fraction of time you pick 6)
p(roll 1 |pick 6) = 1/6
(fraction of time you roll a 1given a 6-sided die)
Stage 1: pick a die
Stage 2: roll the die
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 5 / 19
The Law of Total Probability
Simplified notation:
P6 = pick a 6-sided die
R1 = roll a 1
p(R1 |P6) = probability we roll 1 given that we picked 6-sided die
Example. Suppose you have two 4-sided dice and three 6-sided dice.If I randomly pick a 4- or 6-sided die, then
p(R1) = p(P4 and R1) + p(P6 and R1)
= p(P4)p(R1 |P4) + p(P6)p(R1 |P6)
=2
5· 1
4+
3
5· 1
6= 0.2
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 6 / 19
Probability trees
Example. Have two 4-sided, three 6-sided. Find P(R1) and P(R5).
P4 P6
R1 R2 R3 R4 R1 R2 R3 R4 R5 R6
2/5 3/5
14
14
14
14
16
16
16
16
16
16
Pick die
Roll die
P(R1) = p(P4)p(R1 |P4) + p(P6)p(R1 |P6) =2
5· 1
4+
3
5· 1
6= 0.2
P(R5) = p(P4)p(R5 |P4) + p(P6)p(R5 |P6) =2
5· 0 +
3
5· 1
6= 0.1
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 7 / 19
Board question: probability trees
Before: cup contained two 4-sided dice and three 6-sided dice.
Now: cup contains three 4-sided dice and two 6-sided dice.
1 Is p(roll 1) now bigger, smaller, or the same as before?2 Make a tree.3 Find p(pick 6-sided and roll 1).4 Find p(roll 1).
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 8 / 19
Bayes theorem
The cup still contains three 4-sided dice and two 6-sided dice.
Suppose I pick a die and roll a 1.
What is the probability that I picked a 6-sided die?
We can find the answer using Bayes Theorem:
p(pick 6 | roll 1) =p(roll 1 | pick 6) p(pick 6)
p(roll 1)
or
p(P6 |R1) =p(R1 |P6) p(P6)
p(R1).
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 9 / 19
Proof of Bayes theorem
The multiplication rule tells us that
p(P6 and R1) = p(P6) |R1) p(R1)
p(R6 and P1) = p(R1) |P6) p(P6)
Since these are the same we have
p(P6 |R1) =p(R1 |P6)p(P6)
p(R1).
In general Bayes Theorem says:
P(A|B) =P(B |A)P(A)
P(B)
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 10 / 19
Bayesian updating
The cup still contains three 4-sided dice and two 6-sided dice.
I pick one at random and roll a 1.
What is the probability I picked a 4-sided die? a 6-sided die?That is, compute p(P4 |R1) and p(P6 |R1).
unnormalizedhypothesis prior likelihood posterior posteriorH p(H) p(R1|H) p(R1|H)p(H) p(H|R1)P4 3/5 1/4 3/20 9/13P6 2/5 1/6 2/30 4/13
total 1 p(R1) = 13/60 1
P4 P6
R1 R2 R3 R4 R1 R2 R3 R4 R5 R6
3/5 2/5
14
14
14
14
16
16
16
16
16
16
Pick die
Roll die
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 11 / 19
Board question: Bayesian updating
Now the cup contains one each of the 4, 6, 8, 12, and 20-sided dice.
We choose a die at random.
Question 1. Suppose we roll once and get a 7. What is the posteriorprobability of each die?
Question 2. Suppose we roll the same die a second time and get a9. Now what is the posterior probability of each die?
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 12 / 19
Applications
1 Medical testing2 Trial evidence3 Machine learning
e.g., classify e-mail as work, home or spam:
p(spam |word) =p(word | spam) p(spam)
p(word)
4 Genetic sequencinge.g., determine A, C, G, or T using sequencer data:
p(nucleotide | data) =p(data | nucleotide) p(nucleotide)
p(data)
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 13 / 19
18.05: Introduction to Probability and Statistics
Non-math majors, mostly life science and pre-med.For many, first and last course in statistics.
New curriculum
New pedagogy
New classroom
New technology
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 14 / 19
Active learning, flipped classroom
Meet 3 x 80min in TEAL room
60 students, 2 teachers, 3 assistants
Reading / reading questions on MITx
Minimal lecturing
Group problem solving at boards
Whole class and table discussions
Clicker questions
Computer-based studio using R
Traditional psets and pset checker
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 15 / 19
Active learning versus traditional lecture
Standing up is beneficial
Physical space is critical
Peer and teacher instruction
Student self-assessment
Teacher formative assessment
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 16 / 19
Technology and flipped classroom
Reading questions
Clickers and attendance
Pset checker
How much work was all this?
A tremendous amount because we changed so many things atonce.
How much are you able to cover?
More material with greater understanding.
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 17 / 19
Other observations
Active learning is more fun
Co-teaching is more fun
Students like getting to know their teachers
Students like targeted reading more than lecture video
Students love the pset checker
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 18 / 19
Course Arc
Pure probability
I Counting, random variables, distributions, quantiles, mean, varianceI Conditional probability, Bayes theorem, base rate fallacyI Joint distributions, covariance, correlation, independence
Pure applied probability (Statistics I)
I Bayesian inference with known priorsI Conjugate priors, probability intervals
Applied probability (Statistics II)
I Bayesian inference with unknown priorsI Frequentist significance tests and confidence intervalsI Linear regression, bootstrappingI Discussion of scientific papers
Computation, simulation and visualization using R and applets.
Jeremy Orloff, Jonathan Bloom (MIT Math) Flipping the Dice June 23, 2014 19 / 19