COMP 310
Fall 2017 |
## Lec21: Extended Visitors Example - Self-balancing Trees |

Binary Search Trees (BST's) are great for storing data and give us great speed...or do they?

- Trees offer much faster access than lists, but only if the tree so long as all the branches are nearly the same length, i.e. "balanced".
- A balanced tree gives O(log n) behavior, while lists give O(n) behavior. (See below.for a quick refresher on "Big-Oh" analysis of programs.)
- Worst case scenario for an unbalanced tree is O(n) behavior.

One really wants BSTs and other trees to be balanced for optimal performance!

Kind of like an *n*-ary tree but with multiple pieces of data on
each tree root (node).

An extension of the idea of a BST where the ordering of the data in the child trees depends on the data in the parent node.

- See the PowerPoint Presentation! (web-based presentation -- animations may not run)
- See the OOPSLA 2002 Poster! (>740KB PNG file!)
- Read the OOPSLA 2002 paper in PDF format!
- See the demo!
- Compare the above demo to a demo of the traditional algorithms. Here is the one of the clearest explanations of the traditional insertion and deletion algorithms from Prof. Neli Zlatereva at Central Connecticut State Univ.
- For comparison: quickie explanation of rotations

The main points of this discussion, besides teaching you about 2-3-4 and B-trees, is for you to see the following in action:

- How stepping back and re-assessing a traditional and "well understood" model leads to greater understanding of the system plus a simplified solution.
- How having a large number of cases does not
imply that there are a large number of
*distinguishable*cases. - How extended visitors can make short work of complex problems by easily handling large numbers of cases.
- How commands can be used to dynamically capture a situation (environment) whose information is to be used at a later time.
- How the technique of "candidate data" can be used to abstract a search into process that minimizes the amount of calculations at any given step.

A running program consumes resources such as time (seconds) and space (bits). Frequently, we abstract our units, and measure steps and objects, instead of seconds and bits.

When comparing programs (or algorithms), you should *first* pay
attention to *gross* differences in time or space consumed, for
example, *n ^{3}* versus

For a few programs, the cost is fixed and can be calculated by examining the program text. More frequently, however, cost depends on characteristics of the input, such as length.

When we make gross comparisons of programs, we often refer to the
``order-of-magnitude'' of the cost. The notation used is sometimes called
``Big-Oh,'' and is always of the form O(*f(n)*) where *f(n)* is
some function over the positive integers.

The Big-Oh notation simply means that the cost function is bounded by
(is less than) some multiple of the function *f(n)*. For example, if
we say

*P* = *n*^{3} + O(*n*^{2})

we mean that *P* equals *n*^{3}, plus some terms that are ``on the order of
*n ^{2}*''---i.e., they don't grow faster than

More precisely,

**Definition**. A function g(n) is said to be O(f(n)), written

*g*(n) = O(*f*(n))

if there are positive integers *c* and *n0* such that

0 <= *g*(n) <= c*f*(n)

for all n >= n0.

In other words, O(*f*(n)) is the **set** of all functions *h*(n)
such that there exist positive integers c and n0 such that

0 <= *h*(n) <= c*f*(n)

for all n >= n0.

For example,

1+2+3+ ... +n = n(n+1)/2 = n^{2}/2 + n/2

1+2+3+ ... +n = n^{2}/2 + O(n)

1+2+3+ ... +n = O(*n ^{2}*)

Here are some equivalences that allow you to manipulate equations involving order-of-magnitude quantities:

*f*(n) = O(*f*(n))- K * O(
*f*(n)) = O(*f*(n)) - O(
*f*(n)) + O(*f*(n)) = O(*f*(n)) - O(
*f*(n)) * O(*g*(n)) = O(*f*(n) **g*(n))

Also, the base to which a logarithm is computed doesn't affect the order
of magnitude, because changing the base of the logarithm from 2 to *c*
changes the value by a constant factor of log2(*c)*.

*(written by Alan Cox)*

Big-Oh tells us how a cost of running a program
(algorithm) scales with respect to *n* for large values of *n*, e.g. linearly,
quadraticly, logarithmically, etc. The slower the cost rises with
*n*,
the better, so long as we are dealing with large values of *n*.

- Summing a list of numbers:
*O(n)*-- single traversal of the list - Sorting a list by inserting first into a sorted rest:
*O(n*^{2}) -- double traversal of the list (a traversal to insert each element) - Finding an element in a perfectly balanced binary
search tree:
*O(log(n))*-- height of a balanced tree is*O(log(n))* - Finding an element is a completely unbalanced tree,
worst case scenario:
*O(n)*-- all elements along one branch = linear structure.

© 2017 by Stephen Wong