Status: work-in-progress - 20% complete
Last modified: 2016-12-01

Functional Programming - Why and How

If you have only ever programming in a procedural language (e.g. Fortran, C, Pascal, Cobol) or only in an object-oriented language (Java, C++, etc.), you may find functional programming languages a bit jarring. Many of your “go to” programming patterns won’t work. This chapter will help you reorient your thinking so you can be more productive and less frustrated.

What is a Functional Language?

This is the “for dummies” answer.

A. Pure functional langagues use only immutable variables. Once a variable is created through an assignment, you can’t change its value. You can’t do this:


var x = 4;
x = 6;   // This won't run
print(x);

Any later assigments with the same variable name in the same scope will replace the old variable with a new variable, not modify the old one.


var x = [0,3,6];
var y = x[0] + x[1] + x[2];
if (x[0] == 0){
   var x = [1,2,3];
   var y = x[0] + x[1] + x[2];
   
}
print(" x = "+ x + "; y = " + y);

This prints x = 0,3,6; y = 9 rather than x = 1,2,3; y = 6, because the assignments within the if statement scope (brackets) are local and don’t affect the x and y variables outside.

B. Every function must have a return value, and the only effect of the function is through the return value. No side effects.

C. Every operation is a function call, even simple operators. In LISP, 1 + 3 (infix notation for addition) looks like this: add(1 3). Scala works the same way, but has transformation operations (a.k.a. “syntactic sugar”) that allow you to use infix notation.

D. In Pure functional langagues there are only three types of statements:

  1. Simple assignments: e.g. var x = 6 + 4; and var total = total(arr);, where total is a function and arr is an array.
  2. Function assignments: for example

var arr = [1,1,1,1];
var total = function(arr){
    return reduce(function(x,acc){return x + acc},0,arr);
}
print(total(arr));
  1. Flow control statements like if-else that do not influence any variable out of their scope (scope = “inside a pair of brackets”). For example

var x = 4;
if (x > 3){
    var x = 5;
} else {
    var x = 1;
}
print(x);

This will print 4 because the assignments to x inside the if scope do not affect the orginal (immutable) x.

This is important: the return values of functions can be functions, not just immutable variables. You just need to apply them. For example:

fix this example to include "apply"


var arr = [1,1,1,1];
var total = function(arr){
    return reduce(function(x,acc){return x + acc},0,arr);
}

var doubletotal = function(x,acc){return (x + acc) * 2;}

print("The cummulative sum is " + total(arr));
print("Compiled form of 'doubletotal' function:\n" + doubletotal);
var test = reduce(doubletotal, 0, arr);
print("Applying 'doubletotal' => " + test);

Why Are WebPPL and other PPLs Functional Languages?

The reason has nothing to do with the aesthetics of programming, usability, productivity, or theoretical reasons. It comes down to this:

  1. WebPPL and other PPLs do their “magic” of inference through side effects. If you, the programmer, are also using side effects in your code, then the WebPPL compiler and runtime can no longer do its own side effects reliably.
  2. Functional languages do not permit side effects within functions.

Imagine if you used a purely procedural language (e.g. C). Yes, you might be a nice programmer and code your functions in such a way that their only effects were through return values. But if there was a PPL built on top of C, there would be no way for the compiler and runtime to be sure this was the case.

Take the case of a simple Java for loop over an array, updating the array values to the cummulative sum:

int [] arr = {1,1,1,1};
int sum = 0;
for (int i = 0; i < arr.length; i ++){
       sum += arr[i];
       arr[i] = sum;
}
System.out.println(arr.toString());

This will print 1,2,3,4. But you can’t use the same pattern in a functional language. Why? Because this loop depends on the variable arr being mutable, meaning you can change the values in memory after it is created, and do so at any time, over and over. Inside the loop you are using the original value of each array element, and then modifying it with the cummulative sum. This is, effectively, a side effect of the loop that fully depends on where each statement sits in the program.

Because WebPPL and other PPLs do their “magic” by rearranging your code to insert their side effects, a complied PPL cannot operate safely on for loops and if - then structures where mutable variables are changed along the way.

Tips

  1. How to Apply Math Functions over Arrays
  2. Recursion Can Be Your Best Friend
  3. Use Recursion to Conditionally Loop
  4. Use Recursion and an Accumulator For ‘Empty’ Function Return Values

Tip 1: How to Apply Math Functions over Arrays

Let’s say you want to apply a math function (e.g. max) to all elements of an array. WebPPL does not have an apply(.) function.

First, you might test if the Javascript math function accepts an array object as an argument:


var arr = [0,2,4,6,3];
Math.max(arr);

Nope. Returns null.

With a web search, you find the standard Javascript way to apply a math function over an array, which looks like this:


var arr = [0,2,4,6,3];
Math.max.apply(null,arr);

Does this work? No. It compiles and runs, but gives the wrong answer. The answer should be 6 but this function returns 2. Why? Because WebPPL reconfigures this apply during its compilation process and does so that it no longer works as you’d expect in generic Javascript.

Solution: use reduce, like this:


var arr = [0,2,4,6,3];
reduce(function(x,acc){return x > acc ? x : acc},-Infinity,arr);

This returns the right answer: 6.

Note: The accumulator is initialized to -Infinity so that it will be replaced by any valid number in the array. If you pick some other number, like -1000000, there is always a chance your array will have a bigger negative number. Also, don’t fall into the mental trap of initializing the accumulator to 0, because you’ll only get a valid max if one array value is positive.

Lesson: be cautious about using any Javascript functions that are “compound” (functions acting on functions). In this case, it is apply acting on Math.max. The WebPPL compilation process may reconfigure them in a way that produces incorrect results.

Tip 2: Recursion Can Be Your Best Friend

(Note: Experienced programmers can skip this section)

Recursion is a programming pattern where a function calls it self. Specifically, in functional programming, a recursive function calls itself in the return statement, since the only effects of functions are through their return values.

If you’ve never encountered this idea before, it might seem ghoulish – like a snake eating it’s own tail. Why wouldn’t this lead to an infinite loop? And why would you ever want to do something like this?

It can lead to an infinite loop if the programmer isn’t careful. But there is the same risk in a Javascript for loop or while loop. In all these cases, the programmer is responsible for including termination logic.

The benefit of recursion: it is a good fit for algorithms that have identical sub-algorithms, where each sub-algorithm takes on a subset of the work, and the final resuts are nested inside of each other, like Russian dolls.

(Image by James Jordan, Source: Photographyblogger.net)

A classic example of such an algorithm is depth-first search of graphs, used to find connected components in the graph and to solve mazes with only one solution path. Here is a Java implementation. Note that each iteration of the function prints a result rather than through a return statement.

// Recursive DFS 
public void dfs(int adjacency_matrix[][], Node node) {
    System.out.print(node.data + "\t"); 
    ArrayList<Node> neighbours=findNeighbours(adjacency_matrix,node); 
    for (int i = 0; i < neighbours.size(); i++) {
        Node n=neighbours.get(i); 
        if(n!=null && !n.visited) {
            dfs(adjacency_matrix,n); n.visited=true; 
        } 
    }
}

(Source:Java2blog. See that web page for full code and explanations.)

How Does the Computer Execute Recursion?

Consider the following recursive function in Java:

Modify this code so it is a better example of recursion and call stack effects

int [] data = {1,2,3};
int index = 0;
public void total(result){
    index ++;
    double sqrtTotal = Math.sqrt(result);     
    if (index >= arr.length || sqrtTotal > 1.5){
        return result;
    } else {
        return total(result + arr[index]);
    }
}
System.out.println("Total = " + total(0));

This function computes a cummulative sum, but stops when the square root of the total exceeds 1.5. You’ll notice that index is incremented and there is a local variable for square root in line 5. What happens to this every time you execute the function? Is it overwritten (Java has mutable variables)? Or somehow preserved?

The answer is that the state of each function is preserved in a data structure called “call stack”. Everytime you make a function call, the current state is pushed on to the call stack (maintained by the operating system). When you exit the function, the state is “popped” off the top of the call stack, so you can continue executing just as before.

As spiffy as this is, it has problems. First, it takes execution time to push and pop the program stack. More worrying is that you can run out of stack space if you have a very large number of nested function calls, recursive or otherwise.

For these two reasons, some programmers stay away from recursive functions. However, these are not problems in functional languages. Because there are no mutable variables, there is no variable state that needs to be preserved. And because of a pattern called “tail recursion”, it can be implemented without a call stack that requires more and more memory with each recursive function call. This means that recursive functions do not cost more (performance or memory) than non-recursive functions doing the same algorithm. You can have nearly infinite depth of recursive function calls, limited only by the amount of CPU time you want to use.

Here is the same recursive function in WebPPL:


var data = [1,2,3];
var total = function(index, result, accData){
    // a DIFFERENT "newIndex" and "sqrtTotal" is created every iteration of the function
    var newIndex = index + 1;
    var sqrtTotal = Math.sqrt(result);     
    if (index >= data.length || sqrtTotal > 1.5){
        return {result:result,
                data : accData};
    } else {
        var newAccData = Array.prototype.concat(accData,data[index]);
        return total(newIndex, result + data[index],newAccData);
    }
}

var test = total(0,0,[]);
print("Total of " + test.data + " is " + test.result);

Here are two more tips that use recursion to solve common problems.

Tip 3: Use Recursion to Conditionally Loop

Tip 4:Use Recursion and an Accumulator For ‘Empty’ Function Return Values