Name: Data structures: Binary Search Tree
Uploaded: 2014-06-15T18:26:55.000Z
Duration: 19 min 28 s
Description: Thousands of YouTube videos with English-Chinese subtitles! Now you can learn to understand native speakers, expand your vocabulary, and improve your pronunciation...

In our previous lesson, we talked about binary trees in general. Now, in this lesson we are

going to talk about binary search tree, a special kind of binary tree which is an efficient

Sentences with advanced comparative

structure to organize data for quick search as well as quick update. But before I start

talking about binary search tree, I want you to think of a problem. What data structure

will you use to store a modifiable collection? So, lets say you have a collection and it

can be a collection of any data type, records in the collection can be of any type. Now,

you want to store this collection in computers memory in some structure and then you want

to be able to quickly search for a record in the collection and you also want to be

able to modify the collection. You want to be able to insert an element in the collection

or remove an element from the collection. So, what data structure will you use? Well,

you can use an array or a linked list. These are two well known data structure in which

we can store a collection. Now, what will be running time of these operations - search,

insertion or removal, If we will use an array or a linked list. Lets first talk about arrays

and for sale of simplicity, lets say we want to store integers. To store a modifiable list

or collection of integers, we can create a large enough array and we can store the records

in some part of the array. We can keep the end of the list marked. In this array that

I am showing here, we have integers from 0 till 3. We have records from 0 till 3 and

rest of the array is available space. Now to search some X in the collection, we will

have to scan the array from index 0 till end and in worst case, we may have to look at

all the elements in the list. If n is the number of elements in the list, time taken

will be proportional to n or in other words, we can say that time complexity will be big-oh

of n. Ok, now what will be the cost of insertion. Lets say we want to insert number 5 in this

list. So, if there is some available space, all the cells in yellow are available, we

can add one more cell by incrementing this marker 'end' and fill in the integer to be

added. Time taken for this operation will be constant. Running time will not depend

upon number of elements in the collection. So, we can say that time complexity will be

big-oh of 1. Ok, now what about removal. Lets say we want to remove 1 from the collection.

What we'll have to do is, we'll have to shift all records to the right of one by one position

to the left and then we can decrement end. The cost of removal once again will be big-oh

of n. In worst case, we will have to shift n-1 elements. Here, the cost of insertion

will be big-oh of 1, if the array will have some available space. So, the array has to

be large enough. If the array gets filled, what we can do is, we can create a new larger

array, typically we create an array twice the size of the filled up array. So, we can

create a new larger array and then we can copy the content of the filled up array into

this new larger array. The copy operation will cost us big-oh of n. We have discussed

this idea of dynamic array quite a bit in our previous lessons. So, insertion will be

big-oh of 1 if array is not filled up and it will be big-oh of n if array is filled

up. For now, lets just assume that the array will always be large enough. Lets now discuss

the cost of these operations if we will use a linked list. If we would use a linked list,

I have drawn a linked list of integers here, data type can be anything, the cost of search

operation once again will be big-oh of n where n is number of records in the collection or

number of nodes in the linked list. To search, in worst case, we will have to traverse the

whole list. We will have to look at all the nodes. The cost of insertion in a linked list

is big-oh of 1 at head and its big-oh of n at tail. We can choose to insert at head to

keep the cost low. So, running time of insertion, we can say is big-oh of 1 or in other words,

we will take constant time. Removal once again will be big-oh of n. We will first have to

traverse the linked list and search the record and in worst case, we may have to look at

all the nodes. Ok, so this is the cost of operations if we are going to use array or

linked list. Insertion definitely is fast. But, how good is big-oh of n for an operation

like search. What do you think? if we are searching for a record X, then in the worst

case, we will have to compare this record X with all the n records in the collection.

Lets say, our machine can perform a million comparisons in one second. So, we can say

that machine can perform 10 the power 6 comparisons in one second. So, cost of one comparison

will be 10 to the power -6 second. Machines in today's world deal with really large data.

Its very much possible for real world data to have 100 million records or billion records.

A lot of countries in this world have population more than 100 million. 2 countries have more

than a billion people living in them. If we will have data about all the people living

in a country, then it can easily be 100 million records. Ok, so if we are saying that the

cost of 1 comparison is 10 the power -6 second. If n will be 100 million, time taken will

be 100 seconds. 100 seconds for a search is not reasonable and search may be a frequently

performed operation. Can we do something better? Can we do better than big-oh of n. Well, in

an array, we can perform binary search if its sorted and the running time of binary

search is big-oh of log n which is the best running time to have. I have drawn this array

of integers here. Records in the array are sorted. Here the data type is integer. For

some other data type, for some complex data type, we should be able to sort the collection

based on some property or some key of the records. We should be able to compare the

keys of records and the comparison logic will be different for different data types. For

a collection of strings, for example, we may want to have the records sorted in dictionary

or lexicographical order. So, we will compare and see which string will come first in dictionary

order. Now this is the requirement that we have for binary search. The data structure

should be an array and the records must be sorted. Ok, so the cost of search operation

can be minimized if we will use a sorted array. But, in insertion or removal, we will have

to make sure that the array is sorted afterwards. In this array, if I want to insert number

5 at this stage, i can't simply put 5 at index 6. what I'll have to do is, I'll first have

to find the position at which I can insert 5 in the sorted list. We can find the position

in order of log n time using binary search. We can perform a binary search to find the

first integer greater than 5 in the list. So, we can find the position quickly. In this

case, its index 2. But then, we will have to shift all the records starting this position

one position to the right. And now, I can insert 5. So, even though we can find the

position at which a record should be inserted quickly in big-oh of log n, this shifting

in worst case will cost us big-oh of n. So, the running time overall for an insertion

will be big-oh of n and similarly, the cost of removal will also be big-oh of n. We will

have to shift some records. Ok, so when we are using sorted array, cost of search operation

is minimized. In binary search for n records, we will have at max log n to the base 2 comparisons.

So, if we can perform million comparisons in a second, then for n equal 2 to the power

31 which is greater than 2 billion, we are going to take only 31 micro-seconds. log of

2 to the power 31 to base 2 will be 31. Ok, we are fine with search now. we will be good

for any practical value of n. But what about insertion and removal. They are still big-oh

of n. Can we do something better here? Well, if we will use this data structure called

binary search tree, I am writing it in short - BST for binary search tree, then the cost

of all these 3 operations can be big-oh of log n in average case. The cost of all the

operations will be big-oh of n in worst case. But we can avoid the worst case by making

sure that the tree is always balanced. We had talked about balanced binary tree in our

previous lesson. Binary search tree is only a special kind of binary tree. To make sure

that the cost of these operations is always big-oh of log n, we should keep the binary

search tree balanced. We'll talk about this in detail later. Let's first see what a binary

search tree is and how cost of these operations is minimized when we use a binary search tree.

Binary search tree is a binary tree in which for each node, value of all the nodes in left

sub-tree is lesser and value of all the nodes in right sub-tree is greater. I have drawn

binary tree as a recursive structure here. As we know, in a binary tree, each node can

Subtitles ListPlay Video

Data structures: Binary Search Tree

sort

time

leave

great