Merge branch 'dev'

Author: rcabanasdepaz

CHANGELOG.md

@@ -1,3 +1,20 @@
+0.1.6
+===========
+
+- fixed #44, #40.
+- New class hierarchy for Models:
+    * the main interface is `Model`,
+    * the graphical models now implements the `GraphicalModel` interface,
+    * basic _Directed Acyclic Graph_ is `DAGModel`,
+    * the `BayesianNetwork` class is now a specialized case of `DAGModel`,
+- Added *Join Tree* and *Belief Propagation* algorithms for BayesianFactor-based models. 
+- Added algorithms for entropy of Bayesian and Credal networks. 
+- General code cleanup (mainly code style).
+- Removed main methods from the library and moved to separate unit test classes.
+- Updated tests and removed old experiments.
+- Removed examples relative to Adaptive project.
+- Updated tutorials.
+
 0.1.5
 ===========
 

README.md

@@ -12,9 +12,9 @@ An example of exact inference in a credal network is given below.
 ```java
 import ch.idsia.crema.factor.credal.vertex.VertexFactor;
 import ch.idsia.crema.inference.ve.CredalVariableElimination;
-import ch.idsia.crema.model.ObservationBuilder;
-import ch.idsia.crema.model.Strides;
-import ch.idsia.crema.model.graphical.SparseModel;
+import ch.idsia.crema.core.ObservationBuilder;
+import ch.idsia.crema.core.Strides;
+import ch.idsia.crema.model.graphical.GraphicalModel;
 
 public class Starting {
     public static void main(String[] args) {
@@ -24,7 +24,7 @@ public class Starting {
         /*  CN defined with vertex Factor  */
 
         // Define the model (with vertex factors)
-        SparseModel<VertexFactor> model = new SparseModel<>();
+        GraphicalModel<VertexFactor> model = new DAGModel<>();
               int A = model.addVariable(3);
         int B = model.addVariable(2);
 
@@ -37,7 +37,6 @@ public class Starting {
 
         model.setFactor(A,fu);
 
-
         // Define the credal set of the child
         VertexFactor fx = new VertexFactor(model.getDomain(B), model.getDomain(A));
         fx.addVertex(new double[]{1., 0.,}, 0);
@@ -69,7 +68,7 @@ Add the following code in the  pom.xml of your project:
         <dependency>
             <groupId>ch.idsia</groupId>
             <artifactId>crema</artifactId>
-            <version>0.1.5</version>
+            <version>0.1.6</version>
             <scope>compile</scope>
         </dependency>
     </dependencies>

docs/index.rst

@@ -49,13 +49,19 @@ Crema (CREdal Models Algorithms) is an Java library for inference in credal netw
 
    notes/modeldef
    notes/guideinference
+   notes/bayesianinference
    JavaDoc <https://idsia.github.io/crema/javadoc/>
 
 
+.. toctree::
+   :includehidden:
+   :maxdepth: 1
+   :caption: Tutorials
 
-
-
-
+   tutorials/factors.rst
+   tutorials/domains.rst
+   tutorials/networks.rst
+   tutorials/bayesian_inference.rst
 
 
 .. toctree::

docs/notes/bayesianinference.rst

@@ -0,0 +1,39 @@
+Bayesian Inference
+==============================
+
+Crema provides useful algorithm for precise inference on Bayesian networks.
+
+Belief Propagation
+------------------------------
+
+The ``BeliefPropagation`` inference algorithm works on the ``BayesianFactors`` of a ``BayesainNetwork``.
+
+First instantiate the inference algorithm object using the model. The inference engine will build an internal
+``JunctionTree`` that will be used for the following queries. Then remember to call ``fullPropagation()`` to update
+the model. This will return the posterior of a variable considered the root of the internal ``JunctionTree``.
+
+.. literalinclude:: ../../examples/docs/BeliefPropagation.java
+   :language: java
+   :lines: 42-44
+
+
+To perform an inference on a variable, as an example if you want the marginal of ``P(A)``, use the ``query()`` method as
+in the example below:
+
+.. literalinclude:: ../../examples/docs/BeliefPropagation.java
+   :language: java
+   :lines: 55-60
+
+
+If you want to use evidence, you need to create first a ``TIntIntHashMap`` that will include the state of the various
+variables, in the belo case  we query for ``P(A | B=0)``:
+
+.. literalinclude:: ../../examples/docs/BeliefPropagation.java
+   :language: java
+   :lines: 62-69
+
+
+Full example:
+
+.. literalinclude:: ../../examples/docs/BeliefPropagation.java
+   :language: java

docs/notes/installation.rst

@@ -16,7 +16,7 @@ Crema can be easily included at any maven project. For this, add the following c
         <dependency>
             <groupId>ch.idsia</groupId>
             <artifactId>crema</artifactId>
-            <version>0.1.4</version>
+            <version>0.1.6</version>
             <scope>compile</scope>
         </dependency>
     </dependencies>

examples/BeliefPropagation.java

@@ -0,0 +1,71 @@
+package ch.idsia.crema.inference.bp;
+
+import ch.idsia.crema.factor.bayesian.BayesianFactor;
+import ch.idsia.crema.model.graphical.BayesianNetwork;
+import gnu.trove.map.hash.TIntIntHashMap;
+
+public class BeliefPropagationTest {
+
+	public static void main(String[] args) {
+		/* Define your Bayesian Network model */
+
+		BayesianNetwork model = new BayesianNetwork();
+		A = model.addVariable(2);
+		B = model.addVariable(2);
+		C = model.addVariable(2);
+
+		model.addParent(B, A);
+		model.addParent(C, A);
+
+		// define the Bayesian Factors
+		factors = new BayesianFactor[3];
+
+		factors[A] = new BayesianFactor(model.getDomain(A));
+		factors[B] = new BayesianFactor(model.getDomain(A, B));
+		factors[C] = new BayesianFactor(model.getDomain(A, C));
+
+		factors[A].setData(new int[]{A}, new double[]{.4, .6});
+		factors[B].setData(new int[]{B, A}, new double[]{.3, .7, .7, .3});
+		factors[C].setData(new int[]{C, A}, new double[]{.2, .8, .8, .2});
+
+		// Assign factors to model
+		model.setFactors(factors);
+
+		/* Instantiate the inference algorithm over BayesianFactors using the model */
+
+		BeliefPropagation<BayesianFactor> bp = new BeliefPropagation<>(model);
+
+		// perform a full update
+		BayesianFactor factor = bp.fullPropagation();
+
+		// perform the distribution step
+		bp.distributingEvidence();
+
+		// perform the collection step
+		BayesianFactor factor = bp.collectingEvidence();
+
+		/* Simple Inference */
+
+		// P(A)
+		bp.clearEvidence(); //  this will clear previous evidence on the model
+
+		BayesianFactor pA = bp.query(A);
+
+		/* Inference with evidence */
+
+		// P(A | B=0)
+		TIntIntHashMap evidence = new TIntIntHashMap();
+		evidence.put(B, 0);
+		bp.setEvidence(evidence); // this will overwrite previous evidence
+
+		BayesianFactor pAB0 = bp.query(A);
+
+		// P(A | B=0, C=1)
+		evidence = new TIntIntHashMap();
+		evidence.put(B, 0);
+		evidence.put(C, 1);
+		bp.setEvidence(evidence);
+
+		BayesianFactor bp.query(A);
+	}
+}

examples/Constraints.java

@@ -2,7 +2,6 @@
 import ch.idsia.crema.factor.credal.linear.SeparateHalfspaceFactor;
 import ch.idsia.crema.factor.credal.vertex.VertexFactor;
 import ch.idsia.crema.model.Strides;
-import ch.idsia.crema.model.graphical.SparseModel;
 import org.apache.commons.math3.optim.linear.Relationship;
 
 public class Constraints {

examples/CredalInferenceExample.java

@@ -8,7 +8,6 @@
 import ch.idsia.crema.inference.ve.FactorVariableElimination;
 import ch.idsia.crema.inference.ve.VariableElimination;
 import ch.idsia.crema.model.Strides;
-import ch.idsia.crema.model.graphical.SparseModel;
 import org.apache.commons.math3.optim.linear.Relationship;
 
 import java.util.Arrays;

examples/LoadFromFile.java

@@ -1,6 +1,5 @@
 import ch.idsia.crema.IO;
 import ch.idsia.crema.factor.credal.vertex.VertexFactor;
-import ch.idsia.crema.model.graphical.SparseModel;
 import ch.idsia.crema.model.graphical.specialized.BayesianNetwork;
 
 import java.io.IOException;

examples/PGMpaper.java

@@ -5,7 +5,6 @@
 import ch.idsia.crema.inference.ve.CredalVariableElimination;
 import ch.idsia.crema.model.ObservationBuilder;
 import ch.idsia.crema.model.Strides;
-import ch.idsia.crema.model.graphical.SparseModel;
 import ch.idsia.crema.user.credal.Vertex;
 
 import java.io.IOException;