Entropy & – Information



Entropy & – Information

0 0


Entropy-And-Information-Part-I

Introduction to the nature of Information, and its relationship with uncertainty, probabilities and entropy

On Github SUNY-Albany-CCI-LearningEncounters / Entropy-And-Information-Part-I

Entropy &

Information

Created by Luis Ibanez

Entropy and Information by Luis Ibanez is licensed under a Creative Commons by Attribution 4.0 License Apache 2.0 License

Information

A Question...

Something that you

know

Something that you

don't know

Something that you

can guess

How

Likely ?

Why ask?

if you can guess...

The Cost

of asking...

The Reward

of getting

an answer

The Value

of an answer

Let's Try...

Is it Cold

in Albany

on January ?

Is it Cold

on January ?

Is it Cold ?

Would you

Eat a Monkey ?

Probabilities

If you pick

at random

a day in the calendar...

What are the chances

that it will be a

Tuesday ?

1

----

7

What are the chances

that it will be

the first day

of the month?

12

----

365

If you pick

at random

a card from a deck...

What are the chances

that it will be

an Ace ?

4

----

52

What are the chances

that it will be

a Red card ?

26

----

52

Amount

of

Information

Of an event

with probability

p

-log(p)

-----------

log(2)

Tossing

a coin...

probability

of head ?

1

--

2

amount of

information?

-log(1/2)

-----------

log(2)

= 1

Units ?

= Bits

The answer to a

YES / NO

question

Where

YES and NO

are equally likely

Carries

an amount of

Information

= 1 bit

STOP !

Reflect

consider the

question...

Is the card

an Ace ?

The YES answer

has probability

p = 4 / 52

Its amount of

information is...

-log(4/52)

-----------

log(2)

= 3.7 bits

The NO answer

has probability

p = 48 / 52

Its amount of

information is...

-log(48/52)

-----------

log(2)

= 0.115 bits

Why is it that

the most likely

answer...

Carries

less

Information?

Is the sky blue ?

How good

is the question ?

Back

to the cards

Is the card

an Ace ?

p(Y) = 4/52

p(N) = 48/52

I(Y) = 3.700 bits

I(N) = 0.115 bits

If we ask

many times...

In average...

We will get a YES

4 out of 52 times

We will get a NO

48 out of 52 times

So...

4 out of 52 times

We get 3.7 bits

48 out of 52 times

We get 0.115 bits

In average

We get

( 4 x 3.7 bits + 48 x 0.115 bits)

------------------------------------------

52

= 0.39 bits

The Moment

of Zen

The best

YES / NO

question

that we can

possibly ask

Is the one

where

YES and NO

are equally likely

That question...

Will carry...

an expected

amount of

information...

equal to...

( 1 x 1 bits + 1 x 1 bits)

-------------------------------

2

= 1.0 bits

Breath !

Smile !