Merge pull request probmods#393 from null-a/erp-new-sample-interface

New ERP Interface

examples/hmmSampleWithFactor.wppl

@@ -10,7 +10,7 @@ var trueObs = [false, false, false]
 
 var hmmRecur = function(n, states, observations) {
   var newState = transition(states[states.length - 1])
-  var newObs = sampleWithFactor(observe(newState), [],
+  var newObs = sampleWithFactor(observe(newState),
                                 function(v) {return v == trueObs[observations.length] ? 0 : -Infinity})
   var newStates = states.concat([newState])
   var newObservations = observations.concat([newObs])

examples/ldaCollapsed.wppl

@@ -14,13 +14,13 @@ var makeDirichletDiscrete = function(pseudocounts) {
   globalStore[ddname] = pseudocounts;
   var ddSample = function() {
     var pc = globalStore[ddname];  // get current sufficient stats
-    var val = sample(discreteERP, [pc]);  // sample from predictive. (doesn't need to be normalized.)
+    var val = sample(discreteERP({ps: pc}));  // sample from predictive. (doesn't need to be normalized.)
     globalStore[ddname] = addCount(pc, val); // update sufficient stats
     return val;
   };
   var ddObserve = function(val) {
     var pc = globalStore[ddname];  // get current sufficient stats
-    factor(discreteERP.score([normalize(pc)], val));
+    factor(discreteERP({ps: normalize(pc)}).score(val));
     // score based on predictive distribution (normalize counts)
     globalStore[ddname] = addCount(pc, val); // update sufficient stats
   };

examples/linearRegression.wppl

@@ -4,15 +4,15 @@ var ys = [0, 1, 4, 6]
 var model = function() {
   var m = gaussian(0, 2)
   var b = gaussian(0, 2)
-  var sigmaSquared = gamma(1, 1)
+  var sigma = gamma(1, 1)
 
   var f = function(x) {
     return m * x + b
   }
 
   map2(
       function(x, y) {
-        factor(gaussianERP.score([(f(x)), sigmaSquared], y))
+        factor(gaussianERP({mu: f(x), sigma: sigma}).score(y))
       },
       xs,
       ys)

examples/logisticRegression.wppl

@@ -4,18 +4,18 @@ var labels = [false, false, true, true, true]
 var model = function() {
   var m = gaussian(0, 1)
   var b = gaussian(0, 1)
-  var sigmaSquared = gamma(1, 1)
+  var sigma = gamma(1, 1)
 
   var y = function(x) {
-    return gaussian(m * x + b, sigmaSquared)
+    return gaussian(m * x + b, sigma)
   }
   var sigmoid = function(x) {
     return 1 / (1 + Math.exp(-1 * y(x)))
   }
 
   map2(
       function(x, label) {
-        factor(bernoulliERP.score([sigmoid(x)], label))
+        factor(bernoulliERP({p: sigmoid(x)}).score(label))
       },
       xs,
       labels)

examples/multivariateRegression.wppl

@@ -11,15 +11,15 @@ var model = function() {
   var cov = numeric.diag([1, 1])
   var m = multivariateGaussian([0, 0], cov)
   var e = gaussian(0, 2)
-  var sigmaSquared = gamma(1, 1)
+  var sigma = gamma(1, 1)
 
   var f = function(x) {
     return numeric.dot(m, x) + e
   }
 
   map2(
       function(x, y) {
-        factor(gaussianERP.score([(f(x)), sigmaSquared], y))
+        factor(gaussianERP({mu: f(x), sigma: sigma}).score(y))
       },
       xs,
       ys)

examples/pragmaticsWithSemanticParsing.wppl

@@ -160,7 +160,7 @@ var speaker = cache(function(world) {
   Enumerate(function() {
     var utterance = utterancePrior()
     var L = literalListener(utterance)
-    factor(L.score([], world))
+    factor(L.score(world))
     return utterance
   }, 100)
 })
@@ -170,7 +170,7 @@ var listener = function(utterance) {
     var world = worldPrior(2, function(w) {return 1}) //use vacuous meaning to avoid any guide...
     //    var world = worldPrior(2, meaning(utterance)) //guide by literal meaning
     var S = speaker(world)
-    factor(S.score([], utterance))
+    factor(S.score(utterance))
     return isall(world)
   }, 100)
 }

examples/scalarImplicature.wppl

@@ -31,7 +31,7 @@ var speaker = cache(function(world) {
   Enumerate(function() {
     var utterance = utterancePrior()
     var L = literalListener(utterance)
-    factor(L.score([], world))
+    factor(L.score(world))
     return utterance
   })
 })
@@ -40,7 +40,7 @@ var listener = function(utterance) {
   Enumerate(function() {
     var world = worldPrior()
     var S = speaker(world)
-    factor(S.score([], utterance))
+    factor(S.score(utterance))
     return world
   })
 }

examples/semiMarkovRandomWalkConstrained.wppl

@@ -42,7 +42,7 @@ var transition = function(lastPos, secondLastPos) {
 
 var observeConstrained = function(pos, trueObs) {
   return map2(
-      function(x, obs) { factor(gaussianERP.score([x, 5], obs)); },
+      function(x, obs) { factor(gaussianERP({mu: x, sigma: 5}).score(obs)); },
       pos,
       trueObs
   );

src/aggregation/distribution.ad.js

@@ -42,7 +42,7 @@ function normalize(dist) {
 }
 
 Distribution.prototype.toERP = function() {
-  return erp.makeMarginalERP(normalize(this.dist));
+  return new erp.marginal({dist: normalize(this.dist)});
 };
 
 module.exports = Distribution;

src/aggregation/histogram.js

@@ -27,7 +27,7 @@ function normalize(hist) {
 }
 
 Histogram.prototype.toERP = function() {
-  return erp.makeMarginalERP(normalize(this.hist));
+  return new erp.marginal({dist: normalize(this.hist)});
 };
 
 module.exports = Histogram;