Skip to content

This project (Sorting and Searching Algorithms, AVL/Red-Black Trees , Hashing with Chains) was implemented as part of the Data Structures course during my second year of studies at the Department of Computer Engineering and Informatics (CEID) at the University of Patras.

License

Notifications You must be signed in to change notification settings

Sofosss/DataStructures

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

22 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation


logo
Data Structures Project 2021

A simple introductory project in the field of data structures. It includes the implementation of typical sorting and searching algorithms (along with some alternative versions and optimizations) (PART I), the development of BST trees (AVL and Red-Black) and the implementation of a Hashing structure using chains with linked lists (PART II).


License

Table of Contents

About the Project

The project was implemented as part of the Data Structures course during my second year of studies at the Department of Computer Engineering and Informatics (CEID) at the University of Patras.

In the first part of the project, algorithms for sorting and searching are implemented on a dataset (provided in this repository) containing historical data with daily values for three stocks traded on the NYSE and NASDAQ stock exchanges from 2005 to 2017. Specifically, the implemented sorting algorithms include MergeSort, QuickSort, HeapSort and CountingSort. The searching algorithms encompass Binary Search, Interpolation Search and Binary Interpolation Search (BIS).

In the second part, binary search trees are developed, specifically AVL and Red-Black (RB) trees. Each node of these trees maintains the Date and Volume fields of a stock record from the corresponding input file. The binary search trees are ordered based on a specific field (initially Date, with a later implementation for the Volume field) and they are implemented with dynamic memory management. In case of modification, insertion or deletion of a record (e.g., a tree node), the tree is appropriately readjusted, adhering strictly to the specifications of its structure. Additionally, a Hash Chain structure is provided as an alternative implementation.

PART I: Sorting and Searching Algorithms

  1. MergeSort and QuickSort: In PartI_A, the MergeSort and QuickSort algorithms are implemented with the purpose of sorting dates (Date field) in ascending order based on the opening values (Open field). For QuickSort, three different implementations (variations) are provided. The first implementation corresponds to the Standard QuickSort, the second one to the Randomized QuickSort and the third one to the Median of Three logic.

  2. HeapSort and CountingSort: In PartI_B, the HeapSort and CountingSort algorithms are implemented with the objective of sorting dates (Date field) in ascending order based on the opening values (Open field). For CountingSort, two different implementations (variations) are provided. The first implementation corresponds to the Standard CountingSort and the second one to the Extended CountingSort. The Standard CountingSort operates on the integer values of the Close field (Close: a field of a stock record). However, it does not produce a correctly sorted array (e.g., data records with Close values of 35,510, 35,243, 35,210 appear in the sorted array in that order). The Extended CountingSort operates on the values of 1000*Close, considering them as integers since the Close values of data records have three decimal places. Thus, it produces a correctly sorted array at the expense of requiring a larger space for the C array, which is 1000 times larger than the C array of the standard CountingSort. [ More precisely, it holds that 1.000*[(int(max)–int(min)-1]+1 ≤ size of array C ≤ 1.000*[(int(max)–int(min)+1]-1, where max and min are the maximum and minimum decimal values and (int)max, (int)min are the maximum and minimum integer values (integer part of the decimal value) of the Close field, respectively. ]

  3. Binary Search and Interpolation Search: In PartI_C, the algorithms Binary Search and Interpolation Search are implemented with the purpose of finding the total transaction volume (Volume field) for the stocks in the corresponding file, based on a specific date (Date field).

  4. Binary Interpolation Search and Binary Interpolation Search Improved: In PartI_D, the algorithms Binary Interpolation Search and Binary Interpolation Search Improved are implemented with the purpose of finding the total transaction volume (Volume field) for the stocks in the corresponding file, based on a specific date (Date field). In both variations, the average-case time complexity remains O(lognlogn). The Binary Interpolation Search variation performs jumps of size √n until it locates the block of size √n that potentially contains the search element x. In the next step, the jumps have a size of √√n and so on. This approach improves the worst-case time complexity from O(n) to O(√n). The Improved Binary Interpolation Search variation executes exponentially large jumps (of size √n, 2*√n, 2^2*√n, etc.) until it identifies the block that may contain x. Then, it calls the Binary Search algorithm for that block with elements at positions, for example, m+2^(i-1)*√n to m+2^i*√n in array S. This helps locate the block of size √n that potentially contains x. In one step, at most logn jumps are performed and the Binary Search algorithm requires at most log√n comparisons. Consequently, the total time complexity is O(logn) and in the next step it requires O(log√n) and so on, resulting in a worst-case time complexity of O(logn).

Note

For the creation of the executable files of PART I, a Makefile is provided.

PART II: Binary Search Trees and Hashing

Binary Search Tree implemented as AVL and RED-BLACK based on the Date field

In PartII_A_AVL | PartII_A_RB, a Binary Search Tree (BST) is implemented as an AVL | RED-BLACK node-oriented tree for the records of the corresponding input stock file. Each node of the tree maintains the Date and Volume fields of a stock record.

The operations to be executed on the structure of each individual BST are:

  1. Displaying the Binary Search Tree (BST) with an in-order traversal. Each representation should include a header with the titles of the elements of the records being depicted.
  2. Searching for the volume trading price (Volume) based on the date (Date) provided by the user.
  3. Modifying the volume trading price corresponding to a specific date, which will be provided by the user.
  4. Deleting a record corresponding to a specific date, which will be provided by the user.

The structure of a node in AVL:

struct dateVolume // Data record stored in binary tree node 
{  
    char Date[11];
    int Volume;
};
typedef struct dateVolume dataItem;

struct binaryTreeNode // Binary Search Tree node implemented as AVL tree node
{  
    dataItem data; 
    struct binaryTreeNode *left;
    struct binaryTreeNode *right;
    int height;
};
typedef struct binaryTreeNode btNode;

btNode *root = NULL; // Root of the tree initially empty

The structure of a node in Red-Black(RB):

struct dateVolume // Data record stored in a binary tree node
{  
    char Date[11];
    int Volume;
};
typedef struct dateVolume dataItem;

struct binaryTreeNode // Binary Search Tree node implemented as RED-BLACK tree node
{  
    dataItem data; 
    struct binaryTreeNode *left;
    struct binaryTreeNode *right;
    struct binaryTreeNode *parent;
    char color;
};
typedef struct binaryTreeNode btNode;

btNode *root = NULL; // Root of the tree initially empty

The requested operations are implemented through the following functions:

void inorderBinTree(btNode *r)
btNode *searchBinTree(btNode *r, char x[11])
deleteFromBinTree(btNode *r, char x[11])

(For deleteFromBinTree(), in the case of an AVL tree, it returns a pointer to a node in the tree, while in the case of a RED-BLACK tree it does not return any value).

The modifyBinTree() function has not been implemented separately.

For each different structure (AVL, RED-BLACK), a set of auxiliary functions is used (e.g., rotateL(), rotateR(), btNodeHeight(), btNodeBalance(), uncle(), sibling(), swapColors(), fixRedRed(), fixDoubleBlack(), etc.) to implement the basic operations.

📌 The creation of the tree is done with the help of the corresponding function insertToBinaryTree(dataItem x), which is called by readFileToBinaryTree() every time a data record is read. In the case of the AVL tree, the function returns a pointer to a tree node, while in the case of the RED-BLACK tree, it does not return any value. The insertion of duplicate records (data records with the same value in the Date field) is not allowed. Practically, since all values in the Date field are unique, there are no duplicate records (the precaution against inserting duplicates is taken for completeness reasons).

📌 The rebalancing of the subtree with the root being the node T1 is performed using the functions btNode *rotateL(btNode *T1) and btNode *rotateR(btNode *T1), which implement the single left and right rotations. Double rotation involves either a single left and a single right rotation or a single right and a single left rotation.

Note

In the case of the AVL tree, the insertion of a new node is concluded with at most a single or double terminal rotation, while in the deletion of a node, rotations may propagate up to the root of the tree.

Note

In the case of the RED-BLACK tree, for each insertion or deletion operation of a node at most one single and one double terminal rotation are executed.

Caution

When a violation of the properties of the RED-BLACK tree occurs in the node 'r', the functions void fixRedRed(btNode *r) and void fixDoubleBlack(btNode *r) are called to correct the problem.

  • fixRedRed() is called during the insertion of a new node into the tree when two consecutive red nodes are created (red parent and red child).
  • fixDoubleBlack() is called during the deletion of a black node that leads to a reduction by 1 in the number of black nodes on the paths from the root of the tree that, before the deletion, they(the paths) passed through the deleted node.

Both functions (fixRedRed() and fixDoubleBlack()) are recursive, operate level-wise and perform color changes on the nodes, which may propagate up to the root of the tree along with O(1) rotations.

For the representation of the nodes in each tree, the functions inorderBinTree() and printBinTree() are utilized:

In inorderBinTree(), the fundamental information (Date, Volume) of each data record stored in the tree nodes is displayed, along with a set of auxiliary information primarily for testing purposes. In the AVL tree implementation, additional information such as the height and balance factor of each node is displayed. In the RED-BLACK tree implementation, for each node with one or zero children (leaf), the color of the node, the count of black nodes in the path from the node to the tree's root, as well as the total count of red and black nodes in the path are displayed.

The function printBinTree() prints the structure of the tree from left to right with the Volume information of each node.

  • In the AVL tree implementation, the balance factor of each node is additionally printed.
  • In the RED-BLACK tree implementation, the color of each node is printed.

Binary Search Tree implemented as AVL and RED-BLACK based on the Volume field

As in the previous case, in PartII_B_AVL | PartII_B_RB the tree is constructed with the function insertToBinTree(). The insertion of duplicate records (data records with identical values in the Date and Volume fields) is not allowed. As mentioned earlier, since all values in the Date field are unique, there are no duplicate records (the precaution against inserting duplicates is taken for completeness reasons).

In this case (Volume based BST), the operations to be executed on the structure of each individual BST are:

  1. Finding the day/days with the MINIMUM trading volume.
  2. Finding the day/days with the MAXIMUM trading volume.

The requested operations are implemented through the following functions:

  • btNode *minValuebtNode(btNode *r): Returns a pointer to the node in the tree with the smallest Volume value (leftmost node of the tree with only a right child or is a leaf) or NULL if the tree is empty.

  • btNode *maxValuebtNode(btNode *r): Returns a pointer to the node in the tree with the largest Volume value (rightmost node of the tree with only a left child or is a leaf) or NULL if the tree is empty.

  • void reportBinTree(btNode *r, int x): Prints the Date values of data records that have a Volume value equal to x. [ reportBinTree() is quite general: it can easily be modified to print the Date values of data records with Volume values between x1 and x2, where x1 < x2. The time complexity for a tree with n data records is O(logn + k), where k is the size of the answer (the number of data records with Date values within the given range). ]

Hashing with chained (linked) lists

Finally, in PartII_C, for the records of the corresponding input stock file, we replace the Binary Search Tree (BST) based on the Date field using a Hashing structure with chained (linked) lists. The hash function is computed as the modulus of the sum of ASCII codes of individual characters forming the DATE, divided by an odd number m, which represents the number of buckets. For instance, with DATE = "2014-02-13" and m=11, the formula is:

Hash("2014-02-13") = [ASCII(‘2’)+ ASCII(‘0’)+ ASCII(‘1’)+ ASCII(‘4’)+ ASCII(‘-’)+ ASCII(‘0’)+ ASCII(‘2’)+ ASCII(‘-’)+ ASCII(‘1’)+ ASCII(‘3’)] mod 11

The structure of the data stored in the buckets of the hash chain is as follows:

#define M 11 // Size of Hash table


struct dateVolume // Data record stored in Hash table Bucket list
{
    char Date[11];
    int Volume;
};
typedef struct dateVolume dataItem;

The structure of the buckets of the hash chain is as follows:

struct listNode // Bucket list node
{
    dataItem data;
    struct listNode *next;
};
typedef struct listNode lNode;

lNode *hashTable[M] = {NULL}; // Hash table of M buckets initially empty

In this case (Hashing structure with chained (linked) lists), the operations to be executed are:

  1. Search for the transaction volume based on the date provided by the user.
  2. Modify the record entries based on the date provided by the user. The modification specifically concerns ONLY the transaction volume field.
  3. Delete a record from the hash table based on the date provided by the user.

Note

The creation of the hash structure (buckets with linked lists) is done with the help of the function void insertToHashTable(dataItem x), which is called by readFileToHashTable() each time a data record is read.

Note

The insertion of duplicate records (data records with the same Date) is not allowed. As mentioned earlier, since all values in the Date field are unique, there are no duplicate records (the handling of not allowing duplicate records is here for completeness).

The requested functionalities are implemented through the operations:

lNode *searchHashTable(char x[11])
void deleteFromHashTable(char x[11])

Before the user exits the program, we print the contents of the Hash Table using void displayHashTable().

Note

For the creation of the executable files of PART I, a Makefile is provided.

Important

In the directory, the files PartII_AB_AVL, PartII_AB_RB, PartII_ABC_AVL and PartII_ABC_RB are also provided:

  • PartII_AB_AVL.cpp: It supports the functions of the first two subsections of Part II on the AVL BST structure.
  • PartII_AB_RB.cpp: It supports the functions of the first two subsections of Part II on the Red-Black BST structure.
  • PartII_ABC_AVL.cpp: It supports the functions of the first two subsections of Part II on the AVL BST structure, as well as the Hashing with Chained Linked Lists structure along with the corresponding tasks.
  • PartII_ABC_RB.cpp: It supports the functions of the first two subsections of Part II on the Red-Black BST structure, as well as the Hashing with Chained Linked Lists structure along with the corresponding tasks.

Contact

Authors:

License

Distributed under the MIT License. See LICENSE.md for more details.

About

This project (Sorting and Searching Algorithms, AVL/Red-Black Trees , Hashing with Chains) was implemented as part of the Data Structures course during my second year of studies at the Department of Computer Engineering and Informatics (CEID) at the University of Patras.

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published