Skip to content

Latest commit

 

History

History
190 lines (144 loc) · 5.73 KB

README.md

File metadata and controls

190 lines (144 loc) · 5.73 KB

Raffle

Calculates odds of winning prizes in a raffle. This was written to calculate the expected return from the UK NS&I premium bonds scheme, but can also be applied to any raffle where the prizes are primarily defined by their monetary value.

See it in action!

This project includes a web page for visualising results, but the core logic is available as a component which can be included in other projects under the LGPL license.

Browser Support

Due to the use of Web Workers and modern Javascript syntax, this will only work in modern browsers (Chrome / Safari / FireFox). It may or may not work in Edge and almost certainly will not work in Internet Explorer.

Local Development

To run this project locally, you will need to start a localhost server. If you have nodejs installed, this can be done with a simple command:

npm start

(a localhost server is required because Web Workers cannot be served from the filesystem in all browsers)

WebAssembly

This project uses C for its core logic (compiled to WebAssembly). To change the C code, you will need to install emscripten:

brew install emscripten binaryen

You can also install helpers (e.g. wasm2wat) with:

brew install wabt
# configure git diff for .wasm files:
printf '[diff "wasm"]\n\ttextconv = wasm2wat\n' >> ~/.gitconfig

To rebuild the WebAssembly files, run:

npm run build

Testing

There is also a suite of Jasmine tests (spec/**/*.spec.js) and a linter which can be run with:

npm run test            # run Jasmine & C tests
npm run lint            # run linter
npm run check           # run linter and tests

Using the Library

// First, build a raffle with prizes and a total audience size:

const raffle = new Raffle({
  audience: 100,
  prizes: [
    {count:  2, value: 1000},
    {count: 10, value:  100},
    {count: 20, value:   10},
  ],
  pCutoff: 1e-10, // Optimisation (defaults to 0)
});

// Now enter the raffle with a number of tickets:

raffle.enter(5).then((results) => {
  console.log('Tickets: ' + results.tickets());
  console.log('Audience: ' + raffle.audience());
  console.log(
    'Prize range: £' + results.min().toFixed(2) +
    ' - £' + results.max().toFixed(2)
  );
  console.log('Median winnings: £' + results.median());

  for(let i = results.min(); i <= results.max(); ++ i) {
    const pE = results.exact_probability(i);
    if(pE < 0.00000001) {
      continue;
    }
    const pR = results.range_probability(i, Number.POSITIVE_INFINITY);
    console.log(
      '= £' + i.toFixed(2) + ': ' +
      (pE * 100).toFixed(4) + '%' +
      ' >= £' + i.toFixed(2) + ': ' +
      (pR * 100).toFixed(4) + '%'
    );
  }
});

// We can enter the same raffle any number of times.
// All results will be independent:

raffle.enter(2).then((results) => {
  // ...
});

Explanation

Theory

To calculate the probabilities, all distributions of prizes are simulated (with a configurable cutoff for incredibly unlikely outcomes for performance reasons). The process is:

  1. Clean up the prize list (e.g. by combining prizes of the same value)

  2. Add 0-value prizes to the prize list so that every entry will win something (even if its worth is 0)

  3. Sort the prize list from the rarest to the most common. This is not strictly required but provides a large performance boost in the later stages by keeping the matrices sparse.

  4. Construct a nested array structure. This represents a sparse matrix where the primary ("vertical") dimension represents the number of spent tickets so far, and the secondary ("horizontal") dimension is the total value won so far. The elements of the matrix store the probability of this situation. Initially, only the element [0,0] has a value (1).

    note: a nested array structure is used so that the next stages can run more efficiently.

  5. For each prize (from rarest to most common), iterate down the matrix (all tickets spent to no tickets spent). At each of these rows, calculate the probabilities of winning different quantities of the current prize (e.g. if we are at row 8 of 10, we have 2 tickets remaining, so we can calculate how likely we are to win 0, 1 or 2 of the current prize). Once these probabilities are known, use them to add entries to the rows above the current row (i.e. 1 more ticket spent => value of 1 prize, with current probability * odds of winning, then same for 2, 3, etc.).

    note: in this implementation, a shortcut is taken for the final (most common) prize, since once that stage has completed we will only care about a small part of the matrix.

  6. Read out the probabilities for each monetary value from the top row of the matrix (corresponding to all tickets spent).

  7. Apply post-processing (store probabilities as cumulative probabilities and reduce rounding errors by normalising to [0 1])

To calculate the probability of winning an individual prize (step 5), the equation used is:

(targets C n) * ((total - targets) C (samples - n))
---------------------------------------------------
                 total C samples

Where n C k represents the binomial coefficient:

             n!
n C k = -------------
        k! * (n - k)!

Since these factorials lead to very large numbers, these calculations are performed in log space, using Stirling's log-gamma approximation.

Code

Steps 1-3 above are performed when creating a Raffle object, and the result is stored. When enter is called, steps 4-7 are performed in a worker thread (code for this is in raffle_worker.js). The result is stored in a Result object which is returned. This Result object has various convenience methods for reading the probabilities.