Friday, December 6, 2013

Dec 6 2013

Well, this is the last post which I will be giving in this course, so I will really make an effort to write this well.
I have done a lot of practice during this week in terms of recursion. The focus of my study is mostly on the problems which I was unable to solve before midterm 2. Before this week, my Impression about recursion on trees is mostly based on the understanding of list concatenation. During this week, I have learned how to truly handle the combinatorial nature of trees by assigning variables to the recursive calls and perform functions on these calls. With this technique, I was able to solve many hard problems such as finding the max element in a Binary Tree and looking for a path to the minimum value.  Here is a program which I have wrote that handles one of those problems

def max_val(BT):
    if not BT:
        return 0
    else:
        left_max = max_val(BT.left)
        right_max = max_val(BT.right)
    return max(BT.data, left_max, right_max)

As we can see, it does not involved any list comprehensions or concatenation but it is a very straight forward program that solves a problem that would otherwise be very difficult.

Now to sum up this course. In general, I have greatly benefitted from the material in this course. Especially the fact that although there is no text-book, there is a common theme that runs across the entire course. I loved the material about recursion and how it can be applied to solve a wide range of problems. Since this is the first computer science course ever in my life, I am rather satisfied with what I am able to achieve throughout this course. Although there are times which I find the course difficult, and there are times which I have to spend hours trying to solve a problem. But in return I can say that I have acquired a solid foundation in computer science. Comparing with the other courses which I have taken so far, this area is certainly a breath of fresh air.

Well that is pretty much it, I am certain that everyone have enjoyed this course as much as I did and I hope everyone have a successful exam on Tuesday.

Sunday, December 1, 2013

Regular Entry Dec 1 2013

This week's material is rather interesting due to the fact that it concerns a little bit on the hardware part of computer science. First, We have learned MapReduce, which is the practically the essence of the different sorting algorithms we have learned so far. As I have said in the assigned topics, the essence of O(n*logn) algorithms lies in the fact that they employ a recursive structure. This allows the partition of the domain of a particular function and then simultaneously perform computation to different partitions. The net result will be assembled together and the final result will be returned.

What I have described above is practically a mini version of MapReduce. MapReduce is a frame work for computing parallelizable problems across a large number of CPUs. The entire frame work consist of two steps. The Map step uses the master node and takes in the input data. The input problem is then split into sub problems and assigned to different sub nodes. The worker node may perform another separation and so on. The second step is the reducing step. This step consists of the collection of answers from different sub nodes and the combination of these answers to these outputs which will eventually yield the answer to the problem it originally intended to solve. The general theme of this framework is clearly very important in computer science due to the fact that a large portion of this course is practically recursion. And recursion is essentially the same concept on a smaller scale. The beauty of this concept is that it allows distributed simple problems to be solved while they are altogether too complicated to solve. As we can see from the sorting algorithms, a method that splits the original input and recursively apply the function are usually more efficient.

Another thing which we have learned is memorization. This is completely a new concept to me since it is similar to caching. The whole idea is that it may improve all the recursive functions I have programed so far by making the programs not repeating the calculations which have been performed already. This can further save time in real time since no calculations will be repeated, which makes a recursive function even more efficient.
          

Assigned Topics

This particular post will specifically focus on selection algorithms. I have previously touched upon the topic in the post regarding the big-O notation. This post will focus more on the methods which these algorithms employ and compare the advantages of each.

To begin, let us talk about the well known bubble sort. This particular sorting algorithm is a standard but an inefficient one. Each pass through a list makes one less comparison comparing to the previous one. Due to this nature, the bubble sort algorithm can be implemented using recursion. However due to the sheer number of switches which the algorithm has to make in the worst possible case, i.e the case which the list is completely reversed. This method is extremely inefficient, since the number of switches increases in a quadratic sense with respect to the length of the list.

The selection sort is an improvement of the bubble sort. It improves the bubble sort by selecting out only the largest element in the input and place it at the proper position each time it goes through a list. The number of comparisons which the method makes is the same as the bubble sort, hence yielding O(n^2) efficiency. However, it is much more efficient in terms of the number of switches which it makes during each pass of the element, thus saving us time and energy during the application of this method.

Now, let us shift our perspective a bit. Other than passing linearly through a list, we employ a divide and conquer strategy. To demonstrate this strategy, we look at the method of quicksort. The quicksort algorithm yields O(nlogn) efficiency by partitioning each list in 2 with respect to a pivot value and then recursively apply the same procedure to the subsequent lists generated. The division of the lists is that all items that are smaller than the pivot value goes to the left and all the values that are greater goes to the right. Same method will be repeated for the subsequent lists until the length of the list becomes 1. The correctness of this algorithm can be proven using induction. This particular method is much more efficient comparing to the previous one. This is due to the recursive nature of the algorithm allows the computer to solve multiple problems at the same time. This is a common theme in terms of the efficiency of different sorting algorithms. When a specific method allows us to only put one item in order, we are usually faced with two options. One is to actually go through the entire list sorting one at a time. This will usually yield a O(n^2) efficiency. However, if we are able to divide the list up and use recursion to simultaneously solve the problem on different partitions of the list.

To have something that goes along with this theme, let us look at merge sort. Similar to the quick sort, this algorithm yield a O(n*log n) efficiency. This is due to the same reason as above, there are long n number of ways to split a list in smaller lists of length 2. This method works by completely breaking down a large list into smaller lists of size one. The lists will be sorted with the neighbouring lists, and recursively, bigger lists will be sorted according to the same method and finally yields the sorted master list. Again, together with last algorithm, the beauty of this particular method lies with the beauty of  recursion. Bigger problems are being divided down into smaller lists so that simultaneous application of the function to different partitions of the domain is allowed.

Let us again shift our attention a bit and talk about count sort. Count sort takes a slightly different approach comparing to the bubble sort. I put this sorting algorithm together with bubble sort due to their non recursive nature. Comparing to the bubble sort or insertion sort, this method is highly efficient for a non-recursive function. This is because this entire method is based on the key space of the input list. The bucket list is first generated during the first pass through the list. This list contains a range of possible value that is contained within the list. I say range of possible values due to the fact that in case of integers, we might want to generate a list that has values from the smallest to the largest value of the input. In case of the elements are a deck of cards, we might want a list of all possible representations of the cards. The second step of this method is counting the occurrence of each  element in the bucket-list. The third step, which consist of only one pass through the bucket list, is to rebuild the sorted list from the bucket list. Since the bucket list will be produced sorted by default and it contains the number of occurrences of each element,  the resulting list will be sorted. Note that the beauty of this sort is that really requires one pass through a list and thus the efficiency is linear O(n + k) with n being the number of elements in the input list and k being the number of elements in the bucket list. One disadvantage of this method is that it requires the initial generation of the bucket list, which is entirely dependent on the size of the key space. If the key space is large, then great cost will come with the application of this algorithm.    

Sunday, November 24, 2013

This week we have covered 3 different topics: equality, hiding attributes and reduce. The topic this week which particularly interests me was the build in reduce function. In a typical computer science exercise, we may be asked to find the maximum element within a nested list. With the build in function max(), this is a very easy exercise. However, many times in practice, we may not have tasks as nice as this. For instance, we maybe asked to pick, out of a nested list of string, the string that contains the most vowels. Although still simple, it is a more complicated task comparing to the maximum problem. And in this case, we may utilize the reduce function in other to carry out the task. The code for this task is posted is the following:

def most_vowels1(x,y):
    return max([[sum([x.count(i) for i in 'aeiou']),x], [sum([y.count(i) for i in 'aeiou']),y]])[1]
def most_vowels(L:"iterable"):
    return None if L == [] else reduce(most_vowels1, [most_vowels(i) if isinstance(i, list) else i for i in L])

In my personal perspective, I think using the reduce function in conjunction with recursion can truly yield elegant solutions to a wide range of problems. In the above code, recursion within a list comprehension was used in other to deal with nested lists. The most_vowels1 is a helper function which helps to find the string with the highest occurrence of vowels out of  only two inputs. The most vowels function implements the reduce function and greatly simplifies the code due to the built in algorithm. This approach can usually simplify a problem due to the fact that what is needed to be performed and compared among ALL elements of a nested list can be broken down to what is needed to be performed to only 2 elements. This make the coding much easier and the recursive nature of the final input much easier to handle.
Another thing to note is that the approach above in solving the vowel problem is rather generic. Compare the above code with the following code which sums all elements within a list:

def my_add(x,y):
    return x+y
def my_sum1(L:"iterable"):
    return 0 if L == [] else reduce(my_add,[my_sum1(i) if isinstance(i,list)  else i for i in L])

Notice the rather remarkable similarity to the first block of code. Especially in the second function, the syntax is almost exactly the same. This emphasizes the generality of the reduce function and the fact that it may be used for any arbitrary function which defined with respect to two inputs of the same type and returns data with the same type. Therefore , in other to use the reduce function to the fullest extend, this is something good to keep in mind.

Monday, November 18, 2013

November 18 2013

Today I will be talking about last week's test since it was the only event that occurred that week. In general, I found it to be a fair test. Since all of the questions are of the same kind, the test turned out to be surprisingly short. I have achieved the desired results with what I have coded during the test. However, my methods are not the most efficient there is. For instance, in the last question, instead of directly creating the linked list, I created a helper function that gives out a list and the main function converts the list into a linked list. This method works well if the amount of data which the program is handling is small. However, if we have a large tree and that the amount of data we have to deal with is large, this program will have memory overflow error. This is due to the fact that although there is no upper limit in terms of the number of elements may be contained within each list, there is a max in the storage capacity of a memory slot. Therefore, with only a small or average sized tree, my method will work well, otherwise it will cause exceptions.
 

Tuesday, November 12, 2013

Since the university is closed on Monday and Tuesday, today's post will focus on last week's material.
Last week's lecture covered mostly on big-O analysis. One thing that I have learned is that although big-O analysis gives a rough idea about a specific algorithm's performance, it cannot alone determine an algorithm's efficiency. For instance, the selection sort and the bubble sort are both O(n^2) in terms of the number of comparisons made through the passes. But due to the nature of the bubble sort, many more exchanges occur during each pass comparing to the selection sort. Taking this into consideration, the selection sort is a much more efficient algorithm although it does not appear so from the big-O analysis. Another thing that I have learned is that when performing tasks such as searching through a list, many times the input format of the list does not affect the results of a big-O analysis. This means that the worst case efficiency for many cases is the same. For example, when searching through a unsorted list or searching through a sorted list, the worst case scenario for these two input formats are the same. In both cases, a pass through a list is required and n comparisons are required in total. Therefore both search would yield O(n). If the best cases are considered, the sorted input would be more efficient since if the item is smaller than the first item in the list no pass is required. Thus many times if want to evaluate the efficiency of an algorithm, not only that we need to consider the worst case, the best cases also need to be considered. In order to choose the ideal method, one must evaluate the form of input, not just at one instance in time but also over an entire period of time. The average of the form of the input may then be obtained and could be used to make this decision.
In terms of the assignment, I have found the matching part of the regular expression to be rather challenging. The concept of regular expression deals with the combinatorial explosion of possibilities when the length of the string is enlarged. During the assignment, I have found out that the regular expression of a string is by no means unique. Therefore, when matching an regular expression to a string , decomposing the target string itself may result in ambiguities depending on the method that is used to decompose the string. Thus a better approach would be to start from the regular expression itself and generate the possibilities of the potential matches. When it is finished, the target string can be searched within the generated list of possibilities. Although inefficient for long strings, this method always work since it decodes the regular expression itself. With this approach, I was able to math the basic regular expression with their corresponding targets. However, due to time constraints, I was unable to finish the entire 2nd part. If time permits in the future, I will try to finish the assignment and hopefully come up with a more efficient approach to this problem.     

My friend in CSC148 have talked about the Big-Theta Notation on his blog which can supplement my info that I have provided. Here is the link to his blog. http://athameemcsc148.wordpress.com/ 

Sunday, November 3, 2013

Nov 3rd 2013

This week's material is an extension of the material from the previous week. Using the concept from the previous week, we have learned a new ADT: binary search tree. By implementing binary search tree ADT, searching through can be much more efficient. This is due to the fact that depending on the specific data value with respect to the root, certain parts of the tree can be ignored. While this seems to be rather trivial as a student, but its effectiveness can be certainly appreciated when it comes to large data sets such as trees that exist in operating systems.

One thing that I really enjoyed over the past week is the introduction of the big O notation. I have personally really enjoyed this material due to my Mathematics background and that it really gives me a sneak peak at the theoretical aspects of computer science. The big O notation is a concept that is widely applied throughout mathematics. In essence, it captures the limiting behaviour of a specific function.
Formally speaking: f (x) -> O ( g (x) )  as x -> infinity  <=> lim x->infinity sup|f(x)/g(x)| < infinity
where the sup notation denotes the least upper bound of a specific set. and | | denotes not necessarily the absolute value but the norm of R1. From this more concrete definition, we can have a better understanding of why T1(n)=(0.3365n2 + 0.17n + 0.32) s and T2(n)=(0.47n2 + 0.08n) s are both O ( n^2). By taking the limit of Ti(n)/n^2 as x -> infinity i = 1, 2 , by L'Hospital's Rule, T1(n)/n^2=0.3365 and T2(n) = 0.47. Since both of which are real numbers, we can say that n^2 describes the limiting behaviour of both functions within a certain constant and therefore both algorithms' efficiency are O(n^2).  
In terms of the assignment, I have encountered a challenge this week. The challenge is how to create a recursive data type from a know linear data type such as a string. The second assignment requires us to initialize the tree expression for a regular expression string. But the creation of a tree requires a data type that is recursive in nature. Therefore, the approach that I have decided to take is that first, recover the tree list representation of the regular expression string. secondly, use a recursive algorithm to link the data as nodes. As it turns out, by using a recursive algorithm to peel off the brackets in the string one by one, one can really recover the treelist representation of a regex. From there, it should be very straight forward to link the lists together using the exsting definitions of nodes provided by the course document.