Gentle Introduction to Functional Programming

Functional Programming
(In an oops culture)

What is functional programming (FP)
 It is a programming paradigm and not a programming technique
 Broadly speaking it’s the philosophy of programming which dictates that the
building blocks of code should be “pure functions”
 The idea is to model computational problems in terms of mathematical
formulation.
 This usually means writing code with functions which excel at doing one thing
and doing it great.
 The data structures in functional programming are usually immutable

Pure functions
 T square(T x) { return x*x; }
 bool find(Container<T> items, T item) { for (auto x : items) if (x == item)
return true; return false; }
 void WillDoSomething(T *value) { *value = DidSomething(); }
 std::sort(T start, T end);
 T WillDoSomething() { return DidSomething(); }
 Data structures in FP based code are mostly immutable.
 Immutable data structures + pure functions = Highly parallelizable code

Making containers parallelizable?
Iterators
 Expose an iterator interface.
For eg all typical stl containers support iterator pattern.
bool Find(const Container<Type> &list, const Type &t) noexcept
{
for(const auto &x : list) { if (x==t) return true;}
return false;
}
Notice this find is actually a “pure” function.

Generators
 Generators are like iterators except that the items are generated on demand.
 (Python demo)
 A relatively good example in c++ of generators would be input iterators
while(cin >> value)
{
// Do something
}
Although this does explains what generators are supposed to do, it doesn’t do
justice to the power of generators.

The true power of generators
Problem : write a function that returns the sum of all primes <= N.
A trivial implementation would be:
int sumOfPrime(int N){
if ( N <= 1) return 0;
int sm = 2;
for (int i = 3;i <= N;i+=2) {
if (isPrime(i)){
sm += i;
}
}
return sm;
}
The Problem with this approach is that for each number i s.t. 0 <=i <= N one will necessarily have
to make sqrt(i) iterations taking the worst case to O(N*sqrt(N))

void primeSieve(int N, function doOnPrimes){
vector<bool> sieve(N + 1, 1);
sieve[2] = true;
doOnPrimes(2);
for (int i = 3; i <= N; i+=2){
if(sieve[i]){
doOnPrimes(i);
for(j=3*i; j <= N; j+=2*i) sieive[j] = false;
}
}
}
int sumOfPrimes(int N){
if (N <=1 ) return 1;
int sm = 0;
primeSieve(N, [&sm](int prime){ sm+= prime; });
return sm;
}
Using the sieve the asymptotic complexity remains the same but fewer iterations are made

But….
 Still not a good example of generator
 sumOfPrime needs to keep a state which sort of violates the principles of FP
 That is an important limitation for scalability

Yield keyword
generator<int> primeSieve(int N){
yield 2;
vector<bool> sieve(N + 1, 1);
for (int i = 3; i <= N; i+=2){
if(sieve[i]){
yield i;
for(j=3*i; j <= N; j+=2*i) sieve[j] = false;
}
}
yield break;
}

int sumOfPrimes(int N)
{
int sm = 0;
for (auto var : primeSieve(N)) sm += var;
return sm;
}
(Code shared here: https://github.com/bashrc-
real/Codearchive/tree/master/YeildTry)

Gentle Introduction to Functional Programming

Lambdas and closures
 Lambdas are anonymous functions
 Used for returning from higher order functions* or to be passed to higher
order functions
 Closures are lambdas with state
auto square = [](int x) { return x*x; } -> Lambda
auto squareAndSum = [value](int x) { return (x*x) + value; } -> Closure

MapReduce
 Map: Given a list map each value to another value. The mapping function must not
depend on the state of the list
 reduce : given a list transform the value to a single value
 For eg
Problem : Find the sum of all squares of [0,b]
List -> {0, 1, 2, 3…….b} [equivalent of vector<int> l(b); iota(l.begin(), l.end(), 0); }
Map-> [](int x) { return x*x; }
Reduce -> [](list values) { return accumulate(values.begin(), values.end(), 0); }
Combining the two:
transform(input.begin(), input.end(), output.begin(), [](int x) { return x*x; })
accumulate(output.begin(), output.end(), 0)

MapReduce(contd.)
 Notice that the steps of map reduce can be concatenated
Problem: Find whether a word exists among all words on the internet
Word: “Chewbacca”
MapReduce solution:
1. Define a map function such that [](string x) { int sm = 0; for(auto var:x)
sm+=var; return {x, sm%MOD;} }
2. After running the map on all strings the result will look something like
[DarthVader,7] [Yoda, 0] [Luke,5] [anakinSkyWalker,0]…..
Now we define a reduce function which returns a list of list of the form:
[0->{AnakinSkyWalker, Yoda}, 1->{HanSolo, …}, 2->{…},….]
3. We know the value of Chewbacca = 1 from the map function

4. We take the list with “key value” as 1
[0->{AnakinSkyWalker, Yoda}, 1->{HanSolo, …}, 2->{…},….]
5. We can now apply steps 1-4 each time with probably a new hash function and
at each step after reduce the size of the list will keep getting smaller
6. Eventually the list will be so small that we can iterate through the list and
match the strings character by character to our search string
7. Chewbacca does exist in star wars universe 

Conclusion
 Small stateless functions = testability, easier debugging, less bugs, such
amaze much wow.
 Easy to scale and expand the program to take advantage of multi-cores
without changing the logic
 More power to the compiler

References
 https://wiki.haskell.org/Functional_programming
 http://stackoverflow.com/questions/715758/coroutine-vs-continuation-vs-
generator
 https://en.wikipedia.org/wiki/Higher-order_function

Gentle Introduction to Functional Programming

Recommended

More Related Content

What's hot (20)

Similar to Gentle Introduction to Functional Programming (20)

Recently uploaded (20)

Gentle Introduction to Functional Programming