Other

flip([p])

Draws a sample from Bernoulli({p: p}).

p defaults to 0.5 when omitted.

uniformDraw(arr)

Draws a sample from the uniform distribution over elements of array arr.

display(val)

Prints a representation of the value val to the console.

expectation(dist[, fn])

Computes the expectation of a function fn under the distribution given by dist. The distribution dist must have finite support.

fn defaults to the identity function when omitted.

expectation(Categorical({ps: [.2, .8], vs: [0, 1]})); // => 0.8

marginalize(dist, project)

Marginalizes out certain variables in a distribution. project can be either a function or a string. Using it as a function:

var dist = Infer({model: function() {
  var a = flip(0.9);
  var b = flip();
  var c = flip();
  return {a: a, b: b, c: c};
}});

marginalize(dist, function(x) {
  return x.a;
}) // => Marginal with p(true) = 0.9, p(false) = 0.1

Using it as a string:

marginalize(dist, 'a') // => Marginal with p(true) = 0.9, p(false) = 0.1

forward(model)

Evaluates function of zero arguments model, ignoring any conditioning.

Also see: Forward Sampling

forwardGuide(model)

Evaluates function of zero arguments model, ignoring any conditioning, and sampling from the guide at each random choice.

Also see: Forward Sampling

mapObject(fn, obj)

Returns the object obtained by mapping the function fn over the values of the object obj. Each application of fn has a property name as its first argument and the corresponding value as its second argument.

var pair = function(x, y) { return [x, y]; };
mapObject(pair, {a: 1, b: 2}); // => {a: ['a', 1], b: ['b', 2]}

extend(obj1, obj2, ...)

Creates a new object and assigns own enumerable string-keyed properties of source objects 1, 2, … to it. Source objects are applied from left to right. Subsequent sources overwrite property assignments of previous sources.

var x = { a: 1, b: 2 };
var y = { b: 3, c: 4 };
extend(x, y);  // => { a: 1, b: 3, c: 4 }

cache(fn, maxSize)

Returns a memoized version of fn. The memoized function is backed by a cache that is shared across all executions/possible worlds.

cache is provided as a means of avoiding the repeated computation of a deterministic function. The use of cache with a stochastic function is unlikely to be appropriate. For stochastic memoization see mem().

When maxSize is specified the memoized function is backed by a LRU cache of size maxSize. The cache has unbounded size when maxSize is omitted.

cache can be used to memoize mutually recursive functions, though for technical reasons it must currently be called as dp.cache for this to work.

cache does not support caching functions of scalar/tensor arguments when performing inference with gradient based algorithms. (e.g. HMC, ELBO.) Attempting to do so will produce an error.

mem(fn)

Returns a memoized version of fn. The memoized function is backed by a cache that is local to the current execution.

Internally, the memoized function compares its arguments by first serializing them with JSON.stringify. This means that memoizing a higher-order function will not work as expected, as all functions serialize to the same string.

error(msg)

Halts execution of the program and prints msg to the console.

kde(marginal[, kernelWidth])

Constructs a KDE() distribution from a sample based marginal distribution.

AIS(model[, options])

Returns an estimate of the log of the normalization constant of model. This is not an unbiased estimator, rather it is a stochastic lower bound. [grosse16]

The sequence of intermediate distributions used by AIS is obtained by scaling the contribution to the overall score made by the factor statements in model.

When a model includes hard factors (e.g. factor(-Infinity), condition(bool)) this approach does not produce an estimate of the expected quantity. Hence, to avoid confusion, an error is generated by AIS if a hard factor is encountered in the model.

The length of the sequence of distributions is given by the steps option. At step k the score given by each factor is scaled by k / steps.

The MCMC transition operator used is based on the MH kernel.

The following options are supported:

steps

The length of the sequence of intermediate distributions.

Default: 20

samples

The number of times the AIS procedure is repeated. AIS returns the average of the log of the estimates produced by the individual runs.

Default: 1

Example usage:

AIS(model, {samples: 100, steps: 100})

Bibliography

[grosse16]

Grosse, Roger B., Siddharth Ancha, and Daniel M. Roy. “Measuring the reliability of MCMC inference with bidirectional Monte Carlo.” Advances in Neural Information Processing Systems. 2016.