Caching functions for Webppl

Provides the following conveniences for caching in WebPPL

localStorageClearAll

Webppl function to completely clear out the localStorage

localStorageClear

Webppl function to clear out particular elements from the localStorage

saveCacheToStore

Webppl function to save a particular function's cache to localStorage

restoreCacheFromStore

Webppl function to restore particular function's cache from localStorage

cacheLS

Webppl function memoizer that enables saving/resoring with localStorage

Example

var modelName = 'simple';

var _simpleConditioning = function(p) {
  return Enumerate(function() {
    var x = flip(p);
    var y = flip(p);
    factor((x || y) ? 0 : -Infinity);
    return x;
  })
};

// localStorageClear(modelName) // clears out persistent cache when required
restoreCacheFromStore(modelName);

var simpleConditioning = cacheLS(modelName, _simpleConditioning);
var p = 0.5;
var result = simpleConditioning(p);

saveCacheToStore(modelName);

// First run of this program will compute the value
// // Every subsequent invocation of program will reuse from cache
console.log('Distribution for filp-probability = ' + p + ' is:');
result.print();

stochasticCacheLS

Webppl function memoizer that is aggregative and stochastic.

• aggregative: takes an aggregator that specifies how to combine previously cached values and new value
• stochastic: takes a recomputation probability that is used to decide if, when a candidate value is present in the cache, whether to recompute and aggregate, or simply return cached value -- trading-off speed for convergence

Example

var modelName = 'gaussianMean';
var recomputeProb = 0.6;

var meanAgg = function(cachedV, newV) {
  var totalwt = cachedV.wt + newV.wt;
  return {mean: (cachedV.mean * cachedV.wt + newV.mean * newV.wt) / totalwt,
          wt: totalwt}
};

var myFn = function (mu) {
  var erpA = ParticleFilter(function() {
    var x = gaussian(mu, 2.0);
    return x
  }, 100);
  return {mean: expectation(erpA), wt: 1};
}

// localStorageClear(modelName); // clears out persistent cache when required
restoreCacheFromStore(modelName);

var myCachingFn = stochasticCacheLS(modelName, myFn, meanAgg, recomputeProb);
var result = myCachingFn(1.0);

saveCacheToStore(modelName);

// expection should converge to the mean over time (repeated invocation)
console.log('Expected Value of Model:', result.mean);

Dependencies

• node-localstorage