Explore
For Enterprise
Join for Free
Log In

艺术与人文
商务
计算机科学
数据科学
信息技术
健康
数学和逻辑
个人发展
物理科学与工程
社会科学
语言学习
學位
證書

瀏覽 Coursera 的全部課程

瀏覽
搜索
企業版
登錄
免費加入

Proof: Upper Bound for Chain Length (Optional)

Loading...

数据结构

加州大学圣地亚哥分校

4.7（2,109 個評分） | 91K 名學生已註冊

課程 1（共 6 門，数据结构与算法專項課程）

A good algorithm usually comes together with a set of good data structures that allow the algorithm to manipulate the data efficiently. In this course, we consider the common data structures that are used in various computational problems. You will learn how these data structures are implemented in different programming languages and will practice implementing them in our programming assignments. This will help you to understand what is going on inside a particular built-in implementation of a data structure and what to expect from it. You will also learn typical use cases for these data structures. A few examples of questions that we are going to cover in this class are the following: 1. What is a good strategy of resizing a dynamic array? 2. How priority queues are implemented in C++, Java, and Python? 3. How to implement a hash table so that the amortized running time of all operations is O(1) on average? 4. What are good strategies to keep a binary tree balanced? You will also learn how services like Dropbox manage to upload some large files instantly and to save a lot of storage space!

查看授課大綱

您將學習的技能

Binary Search Tree, Priority Queue, Hash Table, Stack (Abstract Data Type), List

審閱

4.7（2,109 個評分）

5 stars
1,552 ratings
4 stars
453 ratings
3 stars
72 ratings
2 stars
12 ratings
1 star
20 ratings

VK

Jul 16, 2018

Amazing course.\n\nInteresting exercises help with learning and understanding of internal mechanisms basic data structures, how some of them work, the pros and cons of each.\n\nThanks!

SC

Feb 27, 2019

Perfect course for learning more about fundamental data structures, except for presentations on few difficult topics like splay trees, where explanations can be made more elaborate!

從本節課中

Hash Tables

In this module you will learn about very powerful and widely used technique called hashing. Its applications include implementation of programming languages, file systems, pattern search, distributed key-value storage and many more. You will learn how to implement data structures to store and modify sets of objects and mappings from one type of objects to another one. You will see that naive implementations either consume huge amount of memory or are slow, and then you will learn to implement hash tables that use linear memory and work in O(1) on average! In the end, you will learn how hash functions are used in modern disrtibuted systems and how they are used to optimize storage of services like Dropbox, Google Drive and Yandex Disk!

Phone Book Problem4:16

Phone Book Problem - Continued6:49

Universal Family9:42

Hashing Integers9:17

Proof: Upper Bound for Chain Length (Optional)8:43

Proof: Universal Family for Integers (Optional)11:40

Hashing Strings9:33

Hashing Strings - Cardinality Fix7:00

教學方

Alexander S. Kulikov
Visiting Professor
Michael Levin
Lecturer
Daniel M Kane
Assistant Professor
Neil Rhodes
Adjunct Faculty

腳本

Hi, in this video, you will learn how to prove the upper bound on the expected

chain length in the hash table using chaining and

a hash function from the universal hash family.

This proof is needed to be confident that our hash tables

will work fast in practice.

The proof will use some advanced math, which we won't cover, but

this video is optional.

We will use probabilities, mathematical expectation, and

the linearity of expectations.

First, let us recall what is a universal family.

So if U is the universe, the set of all possible keys that we want to store and

calligraphic H is the set of function from this universe to the set of numbers from 0

to m- 1, where m is the selected cardinality of the hash function.

Then this set, calligraphic H, is called universal family, if for

any two different keys from the universe, the probability of collision for

a random hash function on these two keys is at most 1 over m.

What do we mean by probability?

It is taken over the random choice of a hash function from the set calligraphic H.

Or in other words, the reformulation of this definition is that for

any two different keys, x and y from the universe, at most 1 over m part of

all hash functions in calligraphic H produce a collision on these two keys.

Let's also recall what is a load factor.

You have a hash table of size m, storing n keys.

Then n/m is denoted by alpha and called load factor.

That measures how filled up is your table.

If you store just one key in a hash table of size 10, then load factor is 0.1.

And if you store nine keys in a hash table of size 10, then the load factor is 0.9.

We'll also use the property called linearity of expectation.

It says that for any finite set of random variables, x1 through xk,

the random variable y, which is equal to sum of these random variables,

has expectation which is equal to the sum of expectations of these random variables.

We won't prove it, we will use it as a fact from probability theorem.

Now the main theorem is that if you select,

at random, a hash function from a universal family, calligraphic H and

it is used to hash n keys into hash table T of size m giving load factor alpha.

Then for any key k, the expected length of the chain in

table t which contains key k if it stored in it,

is at most 1 + alpha.

What it means is that on average the length of the chain containing

any key will be just 1 + alpha, which is not too long.

Of course, in the worst case, the chain containing some key can be very,

very long such as n keys, and all of them are in the same chain.

But, on average, if it's like the random hash function from the universal family,

the chain length will be at most 1 + alpha, where alpha is the lowest factor.

So to prove that, we fix some arbitrary key k and consider all other keys l

and define a random variable Xkl which takes value of 1 if

the hash value of k is equal to hash value of l,

if there's a collision between keys k and l.

And it takes value zero otherwise.

Let's estimate the expectation of this random variable.

It just takes two different value, so we need first to multiply,

the first one is zero by the probability of zero, but we will anyway get zero.

And we need to add 1 multiplied by the probability of 1,

which is the same as the probability of collision and we know that for

a universal family that probability is at most 1 over m.

So expectation of Xkl for k and any l different from k is at most 1 over m.

Now let's estimate the number of collisions, and we denote by

Yk the number of collisions that k has with different keys l in the table.

And to do that, we can just sum variables Xkl for

all l which are different from key, and are in the table T.

Why is that?

Because when a collision happens for some key l different from k,

variable Xkl takes value 1.

Otherwise, it takes value 0.

So the sum of all these variables is just the number of collisions with key k.

Then the length of the chain containing key k, potentially, in the table T, is 1.

Because this key itself could be in the table.

Plus the number of collisions with different keys, which is Yk.

So the expectation of the number of collisions is by the property of linearity

of expectation, equal to the sum of expectations of variables Xkl.

And we have an upper bound on the expectation of each of those variables,

which is 1 over m.

So this sum is at most sum for

all keys l different from k which are in the table T of value 1 over m.

And there are at most n summands in this sum.

Because the size, the number of keys in the hash table T, Is n.

So at most n of them are both on the table and different from k.

So this sum is at most n over m.

And that is the same as alpha, the load factor.

And so the expectation of the chain length is the expectation of 1 + Yk,

and by linearity, is the same as the expectation of 1 plus expectation of Yk.

But expectation of constant 1 is just 1, and

expectation of Yk we've just estimated from above, and so

we have an upper bound on the expected chain length, which is 1 + alpha.

So we proved our theorem.

Now here is a corollary from this theorem,

which basically says that if you work with your hash table right,

then your operations will on average run in constant time.

More formally, if you do some n operations with your hash table which

include insertions, deletions and modifications, but

the total number of insertions is big O of m, where m is the size of the hash table.

Then, the total time to make all those operations will be big theta of n.

Well, of course it will be at least m to make n operations, but

it will be at most proportional to m.

And thus the amortized time for each operation will be big O of 1 on average.

Let's prove this corollary.

We have big O of m insertions, so

the total number of keys stored at any moment is, at most, big O of m.

n is big O of m.

And then alpha, which is m over m, must be big O of 1, must be constant.

And then 1 +alpha,

our upper bound on the expected chain length, is also big O of one.

And it means that on average,

the expected training time of the operation, is big O of one.

And then when we sum the expectations for

running time of each operation we get the expected running time of all n operations,

again by linearative expectations.

And so the expected running time of n operations is big theta of n,

because it's big O of n, and also it is at least n.

In conclusion,

we've just proved upper bound 1 + alpha on the expected chain length.

In the case when we use a hash table with chaining scheme and

select a random hash function from a universal family.

And we've also proven constant amortized expected running time for

operations with a hash table, If you use universal family and

chaining and you don't put too many keys on your hash table.

Basically if you keep your loss factor below 1, as we discussed in one of

the previous videos, then your operations on average run in constant time.

And in the next video, we'll also prove that the set of hash functions that was

suggested as a universal family for integers is really a universal family.

We will prove that formally.

探索我們的目錄

免費加入並獲得個性化推薦、更新和優惠。

Coursera 致力於普及全世界最好的教育，它與全球一流大學和機構合作提供在線課程。

© 2019 Coursera Inc. 保留所有權利。

通過 App Store 下載

通過 Google Play 獲取

Coursera
關於
管理團隊
工作機會
目錄
證書
MasterTrack™ 證書
學位
企業版
政府版

社區
Learners
Partners
Developers
Beta Testers
Translators

連接
博客
Facebook
領英
Twitter
Instagram
YouTube
技術博客

更多
條款
隱私
幫助
內容訪問
媒體
聯繫我們
目錄
附屬公司