Publications: 82 | Followers: 0

Getting Started - cct.lsu.edu

Publish on Category: Birds 0

Using Sequential Containers
Lecture 8Hartmut Kaiserhkaiser@cct.lsu.eduhttp://www.cct.lsu.edu/˜hkaiser/spring_2015/csc1254.html
Programming Principle of the Day
Principle of leastastonishment (POLA/PLA)Theprinciple of least astonishment is usually referenced in regards to the user interface, but the same principle applies to written code.Codeshouldsurprisethe reader as little as possible.Thatmeans following standard conventions, code should do what the comments and name suggest, and potentially surprising side effects should be avoided asmuchas possible.http://en.wikipedia.org/wiki/Principle_of_least_astonishment
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
2
Abstract
We start looking beyond vector and string. We will focus on sequential containers and demonstrate a couple of problems we can solve when applying them.The standard library’s architecture will start to get visible. That will help us to start understanding how to use all of the different containers in the standard library.
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
3
Separating Students into Categories
Sort out failed studentsWho failed?Remove from our dataCreate a new vector ofstudent_datacontaining only students who succeeded:// predicate to determine whether a student failedboolfail_grade(student_infoconst& s){returngrade(s) < 60;}Pushstudent data onto one of two containers based on this predicate
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
4
Separating Students into Categories
What’s wrong here? (Hint: what’s the memory consumption?)// separate passing and failing student records: first tryvector<student_info>extract_fails(vector<student_info>& students){vector<student_info> pass, fail;for(vector<student_info>::size_type i = 0;i != students.size(); ++i){if(fail_grade(students[i]))fail.push_back(students[i]);elsepass.push_back(students[i]);}students = pass;returnfail;}
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
5
Separating Students into Categories
Requires twice as much memoryEach record is held twiceBetter to copy failed students, removing the data from original vectorHow to remove elements from a vector?Slow, too slow for larger amounts of data.Why?What happens if all students have failed?This can be solved by either using a different data structure or by modifying the algorithm
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
6
ErasingElementsinPlace
Slow, but direct solution (Why is it slow?)// second try: correct but potentially slowvector<student_info>extract_fails(vector<student_info>& students){vector<student_info> fail;vector<student_info>::size_typei= 0;// invariant: elements [0,i) of students represent passing gradeswhile(i!=students.size()) {if(fail_grade(students[i])) {fail.push_back(students[i]};students.erase(students.begin() +i);}else++i;}returnfail;}
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
7
Erasing Elements in Place
The erase() function takes a special type ‘pointing’ (referring) to the element to erase:students.erase(students.begin() +i);
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
8
Elements we’ve already seen
FAIL
Elements we haven’t processed
Elementi
students.size() == n
Elements we’ve already seen
Elements we haven’t processed
students.size() == n - 1
(These elements are copied)
Erasing Elements in Place
Caution: why will this fail?// this code will fail because of misguided optimizationautosize =students.size();while(i!= size) {if(fail_grade(students[i])) {fail.push_back(students[i]);students.erase(students.begin() +i);}else++i;}
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
9
SequentialVersus Random Access
Both versions share a non-obvious propertyThe elements are accessed sequentially onlyWe used integer ‘i’ as an index, which hides thatNeed to analyze every operation on ‘i’ to verifyWe might access student data in arbitrary orderEvery container type has its performance characteristics for certain operationsBy knowing what access pattern we use we can utilize the ‘best’ container type
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
10
Sequential Versus Random Access
Let’s restrict our access to being sequentialThe standard library exposes special types we can use to express this intent:IteratorsBy choosing the right type of iterator we ‘tell’ the library what access pattern we need to supportAllows for optimal selection of the underlying algorithm implementation
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
11
Iterators
Our code uses the index forAccess of an elementfgrade(students[i])Moveto the nextelement (increment ‘i’)while(i!=students.size()) {// work gets done here; but doesn't change the value ofii++;}We useindex for sequential access only!But there is no way of telling the library about this
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
12
Iterators
Iterators are special typesIdentifya container and an element in thecontainerLetus examine the value stored in that elementProvideoperations for moving between elements in the containerRestrictthe available operations in ways that correspond to what the containercan handleefficiently
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
13
Iterators
Code using iterators is often analogous to index based code:// code based on indiciesfor(vector<student_info>::size_typei = 0;i!= students.size(); ++i){cout<< students[i].name <<endl;}// code based on iteratorsfor(vector<student_info>::const_iteratoriter=students.begin();iter!=students.end(); ++iter){cout<< (*iter).name <<endl;}
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
14
Iterator Types
Every standard container, such as vector, defines two associated iterator types:container_type::iteratorcontainer_type::const_iteratorWherecontainer_typeis the container (vector<student_info>)Useiteratorto modify the element,const_iteratorotherwise (read only access)Note, that we don‘t actually see the actual type, we just know what we can do with it.Abstraction is selectiveignorance!
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
15
Iterators, C++11
Code using iterators is often analogous to index based code:// code based on indiciesfor(autoi = 0; i != students.size(); ++i){cout<< students[i].name <<endl;}// code based on iterators using C++11 (VS2010, g++4.2)for(autoiter=students.begin();iter!=students.end(); ++iter){cout<< (*iter).name <<endl;}
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
16
Iterator Types
Everycontainer_type::iteratoris convertible to the correspondingcontainer_type::const_iteratorstudents.begin() returns an iterator, but we assign it to aconst_iteratorOpposite is not true! Why?
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
17
Iterator Operations
Containers not only expose their (specific) iterator types, but also actual iterators:students.begin(),students.end()begin(): ‘points’ to the first elementend(): ‘points’ to the element after the last oneIterators can becompared:iter!=students.end()Tests, whether both iterators refer to the same elementIterators can beincremented:++iterMake the iterator ‘point’ (refer) to the next element
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
18
Iterator Operations
Iterators can bedereferenced:*iterEvaluates to the element the iterator refers toIn order to access a member of the element the iterator refers to, we write:(*iter).name(why not:*iter.name?)Syntactic sugar, 100% equivalent:iter->name
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
19
Iterator Operations
Some iterators can get a number addedstudents.erase(students.begin() +i);Overloaded operator+, makes the iterator refer to the ‘i’ –s element after beginEquivalent to invoking ++ ‘i’ timesDefined only for iterators fromrandom accesscontainersvector, string are random access (indexing is possible)Will result in compilation error for sequential containers
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
20
ErasingElementsinPlace
Slow, but direct solution// second try: correct but potentially slowvector<student_info>extract_fails(vector<student_info>& students){vector<student_info> fail;autoi= 0;// invariant: elements [0,i) of students represent passing gradeswhile(i!=students.size()) {if(fail_grade(students[i])) {fail.push_back(students[i]};students.erase(students.begin() +i);}else++i;}returnfail;}
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
21
Erasing Elements in Place
Still slow, but without indexing:// version 3: iterators but no indexingvector<student_info>extract_fails(vector<student_info>& students){vector<student_info> fail;autoiter=students.begin();while(iter!=students.end()) {if(fail_grade(*iter)) {fail.push_back(*iter);iter=students.erase(iter);// watch out! Why?}else++iter;}returnfail;}
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
22
Iterator Invalidation
What happens to an iterator if the element it refers to is deleted?It is invalidatedCertain containers invalidate all iterators after the deleted element as well (vectors)For that reason erase() returns the next iterator:iter=students.erase(iter);
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
23
Same problem as Before
Why does this code fail:// this code will fail because of misguided optimizationautoiter=students.begin();autoend_iter=students.end();while(iter!=end_iter) {// ...}End iterator is invalidated as well when element is erased!
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
24
What’s the problem with vector
For small inputs, vector works just fine, larger inputs cause performance degradationVector is optimized for fast access to arbitrary elements and for fast addition to the endInserting or removing from the middle is slow.All elements after the inserted/removed element need to be moved in order to preserve fast random accessOur algorithm has quadratic performance characteristicsLet’s utilize a different data structure:Next lecture:The list type
19/2/2015, Lecture 8
CSC 1254, Spring 2015, Using Sequential Containers
25

0

Embed

Share

Upload

Make amazing presentation for free
Getting Started - cct.lsu.edu