Lazy Variables for Dart

Introduction

The package lazy_memo provides generic classes that can be used to define lazy cached variables and memoized functions.

Minimizing CPU and memory usage are two important goals of software optimization. If sufficient memory is available, costly operations (such as sorting a large list) can be avoided by storing the result and reusing it as long as the relevant input (e.g. the unsorted list) has not changed.

The technique of storing the result of function calls was coined memoization.

A different strategy to minimize CPU usage is to delay the initialization of variables. Lazy initialization is particularly useful in event driven scenarios where there is no definite execution path and a certain variable might never be used.

Usage

To use this library include lazy_memo as a dependency in your pubspec.yaml file.

1. Lazy Variables

Note: To define variables that are going to be initalized once simply use Dart's late modifier.

To define cached lazy variables that can be marked for re-initialization use the generic class Lazy<T>. I often find it useful to declare lazy variables using the late modifier since it makes it possible to initialize final instance variables at the point of definition.

1. Lazy variables are declared using the constructor of the generic class Lazy<T>.

2. The constructor requires a callback, ObjectFactory, that returns an object of type T. To prevent (inadvertent) modification of the cached variable it is advisable to have ObjectFactory return an immutable object. For more info see the section Lazy Collections below.

3. To access the cached object, the lazy variable is called like a function (see example below).

4. The optional parameter updateCache can be used to request an update of the cached object. If updateCache is true, the object is re-initialized by calling the object factory ObjectFactory with the current value of the input arguments.

import 'dart:math';

import 'package:lazy_memo/lazy_memo.dart';

// To run this program navigate to
// the root folder of your local copy of 'lazy_memo' and use the command:
//
// # dart example/bin/lazy_example.dart
void main() {
  print('Running lazy_example.dart.\n');

  final random = Random();
  final mean = 4.0;

  print('Generating a random sample with size 5000 and mean: 4.0:');
  // Generating a random sample
  final sample = List<double>.generate(
      5000, (_) => -mean * log(1.0 - random.nextDouble()));

  // Initializing lazy variables
  final sampleSum = Lazy<double>(
    () => sample.reduce((sum, current) => sum += current),
  );

  // Calculating the sample mean
  final sampleMean = Lazy<double>(
    () => sampleSum(updateCache: true) / sample.length,
  );

  print('  Initial value of sampleMean: ${sampleMean()}');
  print('  Initial value of sampleSum: ${sampleSum()}\n');
  print('Adding outliers to random sample: [1500.0, 1200.0]');

  // Adding outliers
  sample.addAll([1500.0, 1200.0]);

  print('  Updated value of sampleMean: '
      '${sampleMean(updateCache: true)}');
  print('  Updated value of sampleSum: ${sampleSum()}');
}

2. Dependent Lazy Variables

It is possible to declare dependent lazy variables by using an expression containing one lazy variable to declare another lazy variable.

// Calculating the sample mean
  final sampleMean = Lazy<double>(
    () => sampleSum(updateCache: true) / sample.length,
  );

In the example above, sampleMean depends on sampleSum since the callback passed to the constructor of sampleMean references sampleSum.

The optional parameter updateCache can be used strategically to trigger an update of cached variables along the dependency tree. Since sampleSum(updateCache: true) is called every time sampleMean is updated, an update of sampleMean triggers an update of sampleSum.

Note: An update of a lazy variable can also be requested by calling the instance method: updateCache().

Click to show console output.

$ dart example/bin/lazy_example.dart
Running lazy_example.dart.

Generating a random sample with size 5000 and mean: 4.0:
 Initial value of sampleMean: 4.048803375544851
 Initial value of sampleSum: 8097.606751089702

Adding outliers to random sample: [1500.0, 1200.0]
 Updated value of sampleMean: 5.393409965579271
 Updated value of sampleSum: 10797.606751089701

3. Lazy Collections

Lazy variables can be used to cache objects of type List, Set, Map, etc. However, as the example below demonstrates, the cached object can be modified.

final lazyList = Lazy<List<int>>(() => [1, 2, 3]);
final list = lazyList();
list.add(4); // lazyList() now returns: [1, 2, 3, 4]

In order to prevent users from (inadvertently) modifying the cached object one should instead use the classes LazyList<T>, LazySet<T>, and LazyMap<K, V>. These classes cache a collection and return an unmodifiable view of the collection.

4. Memoized Functions

Memoized functions maintain a lookup table of previously calculated results. When called, a memoized function checks if it was called previously with the same set of arguments. If that is the case it will return a cached result.

Memoizing a function comes at the cost of additional indirections, higher memory usage, and the complexity of having to maintain a function table. For this reason, memoization should be used for computationally expensive functions that are likely to be called repeatedly with the same set of input arguments. Examples include: repeatedly accessing statistics of a large data sample, calculating the factorial of an integer, repeatedly evaluating higher degree polynomials.

The example below demonstrates how to define the memoized functions factorial(n) and combinations(n, k), k-combinations of n objects. Note: For the sake of simplicity, validation of arguments is omitted. A complete version of these functions is provided with the library utils.dart.

Click to show souce code.

 import 'package:lazy_memo/lazy_memo.dart';

 /// Computationally expensive function with a single argument.
 int _factorial(int x) => (x == 0 || x == 1) ? 1 : x * _factorial(x - 1);

 /// Returns the factorial of a positive integer.
 final factorial = MemoizedFunction(
   _factorial,
   functionTable: {8: 40320}, // Optional initial function table.
 );

 /// Computationally expensive function with two arguments.
 int _combinations(int n, int k) {
   if (k > n ~/ 2) {
     return _combinations(n, n - k);
   } else if (k > n) {
     return 0;
   } else {
     int result = 1;
     int m = 1;
     for (var i = n; i > n - k; i--) {
       result = (result * i) ~/ m;
       m++;
     }
     return result;
   }
 }

 /// Returns the number of k-combinations of n distinct objects. More formally,
 /// let S be a set containing n distinct objects.
 /// Then the number of subsets containing k objects is given
 /// by combinations(n, k).
 /// * combinations(n, n) = 1
 /// * combinations(n, k) = combinations(n, n - k)
 /// * combinations(n, 0) = 1
 final combinations = MemoizedFunction2(_combinations);

 // To run this program navigate to
 // the root folder of your local copy of 'lazy_memo' and use the command:
 //
 // # dart example/bin/lazy_function_example.dart
 void main() {
   print('Running lazy_function_example.dart.\n');

   print('------------- Factorial --------------');
   print('Calculates and stores the result');
   print('factorial(12) = ${factorial(12)}\n');

   // The current function table
   print('Function table:');
   print(factorial.functionTable);
   print('');

   // Returning a cached result.
   print('Cached result:');
   print('factorial(12) = ${factorial(12)}');

   print('\n----- k-combinations of n objects -----');

   print('Calculates and stores the result of: ');
   print('combinations(10, 5): ${combinations(10, 5)}');
   print('');

   print('The current function table');
   print(c.functionTable);
   print('');

   print('Returns a cached result.');
   print('combinations(10, 5): ${combinations(10, 5)}');
 }

Click to show console output.

$ dart example/bin/lazy_example.dart
Running lazy_function_example.dart.

------------- Factorial --------------
Calculates and stores the result
factorial(12) = 479001600

Function table:
{8: 40320, 12: 479001600}

Cached result:
factorial(12) = 479001600

----- k-combinations of n objects -----
Calculates and stores the result of:
combinations(10, 5): 252

The current function table
{10: {5: 252}}

Returns a cached result.
combinations(10, 5): 252

Examples

The source code listed above is available in the folder example.

Features and bugs

Please file feature requests and bugs at the issue tracker.