lec-1 probabilistic models.ppt
-
Upload
clara-agustin-stephani -
Category
Documents
-
view
220 -
download
0
Transcript of lec-1 probabilistic models.ppt
-
8/10/2019 lec-1 probabilistic models.ppt
1/29
Probabilistic models
Haixu Tang
School of Informatics
-
8/10/2019 lec-1 probabilistic models.ppt
2/29
Probability
Experiment: a procedure involving
chance that leads to different results
Outcome:the result of a single trial of an
experiment;
Event:one or more outcomes of an
experiment;
Probability:the measure of how likely an
event is;
-
8/10/2019 lec-1 probabilistic models.ppt
3/29
Example: a fair 6-sided dice
Outcome: The possible outcomes of
this experiment are 1, 2, 3, 4, 5 and
6;
Events: 1; 6; even
Probability: outcomes are equal ly
l ikelyto occur. P(A) = The Number Of Ways Event A Can Occur / The Total
Number Of Possible Outcomes
P(1)=P(6)=1/6; P(even)=3/6=1/2;
-
8/10/2019 lec-1 probabilistic models.ppt
4/29
Probability distribution
Probability distribution: the assignment of a
probability P(x) to each outcome x.
A fair dice: outcomes are equally l ikelyto occur
the probability distribution over the all sixoutcomes P(x)=1/6, x=1,2,3,4,5 or 6.
A loaded dice: outcomes are unequal ly l ikelyto
occurthe probability distribution over the all
six outcomes P(x)=f(x), x=1,2,3,4,5 or 6, but
f(x)=1.
-
8/10/2019 lec-1 probabilistic models.ppt
5/29
Example: DNA sequences
Event: Observing a DNA sequence S=s1s2sn:
si{A,C,G,T};
Random sequence model (or Independent and
identically-distributed, i.i.d.model): si occurs atrandom with the probability P(si), independent
of all other residues in the sequence;
P(S)=
This model will be used as a background
model (or called null hypothesis).
n
i
isP
1
-
8/10/2019 lec-1 probabilistic models.ppt
6/29
Conditional probability
P(i|): the measure of how likely an event ihappens under the condition ; Example: two dices D1, D2
P(i|D1)probability for picking i using dicer D1
P(i|D2)probability for picking i using dicer D2
-
8/10/2019 lec-1 probabilistic models.ppt
7/29
Joint probability
Two experiments X and Y
P(X,Y)joint probability (distribution) of experiments
X and Y
P(X,Y)=P(X|Y)P(Y)=P(Y|X)P(X) P(X|Y)=P(X), X and Y are independent
Example: experiment 1 (selecting a dice),
experiment 2 (rolling the selected dice)
P(y): y=D1or D2
P(i, D1)=P(i| D1)P(D1)
P(i| D1)=P(i| D2), independent events
-
8/10/2019 lec-1 probabilistic models.ppt
8/29
Marginal probability
P(X)=YP(X|Y)P(Y)
Example: experiment 1 (selecting a dice),
experiment 2 (rolling the selected dice)
P(y): y=D1or D2
P(i) =P(i| D1)P(D1)+P(i| D2)P(D2)
P(i| D1)=P(i| D2), independent events
P(i)= P(i| D1)(P(D1)+P(D2))= P(i| D1)
-
8/10/2019 lec-1 probabilistic models.ppt
9/29
Probability models
A system that produces different outcomes
with different probabilities.
It can simulate a class of objects (events),
assigning each an associated probability.
Simple objects (processes)probability
distributions
-
8/10/2019 lec-1 probabilistic models.ppt
10/29
Example: continuous variable
The whole set of outcomes X (xX) canbe infinite.
Continuous variablex[x0,x
1]
P(x0xx1) ->0
P(x-dx/2 x x+dx/2) = f(x)dx; f(x)dx=1
f(x)probability density function (density, pdf)
P(xy)= y
x0f(x)dxcumulated density function (cdf)
x0
x1
x1
dx
-
8/10/2019 lec-1 probabilistic models.ppt
11/29
Mean and variance
Mean
m=xP(x)
Variance
2= (k-m)2P(k)
: standard deviation
-
8/10/2019 lec-1 probabilistic models.ppt
12/29
Typical probability distributions
Binomial distribution
Gaussian distribution
Multinomial distribution Dirichlet distribution
Extreme value distribution (EVD)
-
8/10/2019 lec-1 probabilistic models.ppt
13/29
Binomial distribution
An experiment with binary outcomes: 0 or 1;
Probability distribution of a single experiment:
P(1)=p and P(0) = 1-p;
Probability distribution of N tries of the same
experiment
Bi(k 1s out of N tries) ~kNk
ppk
N
)1(
-
8/10/2019 lec-1 probabilistic models.ppt
14/29
Gaussian distribution
N -> , Bi -> Gaussian distribution
Define the new variable u = (k-m)/ f(u)~ 2/exp 2
2
1 u
-
8/10/2019 lec-1 probabilistic models.ppt
15/29
Multinomial distribution
An experiment with K independent outcomes
with probabilities i, i =1,,K, i =1.
Probability distribution of N tries of the same
experiment, getting nioccurrences of outcome i,
ni =N.
M(N|) ~
K
i
n
i
i
i
i
n
N
1!
!
-
8/10/2019 lec-1 probabilistic models.ppt
16/29
Example: a fair dice
Probability: outcomes (1,2,,6) are equal ly
l ikelyto occur
Probability of rolling 1 dozen times (12) andgetting each outcome twice:
~3.410-3 1261
2
!126
-
8/10/2019 lec-1 probabilistic models.ppt
17/29
Example: a loaded dice
Probability: outcomes (1,2,,6) are
unequal ly l ikelyto occur: P(6)=0.5,
P(1)=P(2)==P(5)=0.1
Probability of rolling 1 dozen times (12) and
getting each outcome twice:
~1.8710-4
102
2
!12
1.05.06
-
8/10/2019 lec-1 probabilistic models.ppt
18/29
Dirichlet distribution
Outcomes: =(1, 2,,K)
Density: D(|a)~
(a1, a2,,aK) are constantsdifferent agives different probability distribution over
.
K=2Beta distribution
1
11
1K
i
i
K
i
ii
da
-
8/10/2019 lec-1 probabilistic models.ppt
19/29
Example: dice factories
Dice factories produces all kinds of dices:
(1), (2),, (6)
A dice factory distinguish itself from the
others by parameters aa1,a2 ,a3 ,a4 ,a5 ,a6
The probability of producing a dice in thefactory ais determined by D(|a)
-
8/10/2019 lec-1 probabilistic models.ppt
20/29
Extreme value distribution
Outcome: the largest number among N
samples from a density g(x)is larger than
x;
For a variety of densities g(x),
pdf:
cdf:
-
8/10/2019 lec-1 probabilistic models.ppt
21/29
Probabilistic model
Selecting a model Probabilistic distribution
Machine learning methods Neural nets
Support Vector Machines (SVMs) Probabilistic graphical models
Markov models
Hidden Markov models
Bayesian models
Stochastic grammars Modeldata (sampling)
Datamodel (inference)
-
8/10/2019 lec-1 probabilistic models.ppt
22/29
Sampling
Probabilistic model with parameter P(x| )for event x;
Sampling: generate a large set of eventsxiwithprobability P(xi|);
Random number generator ( function rand()picks a number randomly from the interval [0,1)with the uniform density;
Sampling from a probabilistic model
transforming P(xi| ) to a uniform distribution For a finite set X (xiX), find i s.t. P(x1)++P(xi-1)