Skip to content

Commit

Permalink
why examples
Browse files Browse the repository at this point in the history
  • Loading branch information
@rcabanasdepaz
rcabanasdepaz committed Sep 21, 2021
1 parent 20dc973 commit 626daf3
Show file tree
Hide file tree
Showing 253 changed files with 93,896 additions and 2,748 deletions.
4 changes: 2 additions & 2 deletions papers/21why/examples/PearlExample.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -136,8 +136,8 @@ public static void main(String[] args) throws InterruptedException, ExecutionCon

CredalCausalVE inf = new CredalCausalVE(m_eqless, empData.values());

IO.write(m_eqless, folder+"consPearl.uai");
DataUtil.toCSV(folder+"dataPearl.csv", data);
IO.write(m_eqless, folder+ "examples/consPearl.uai");
DataUtil.toCSV(folder+ "examples/dataPearl.csv", data);


VertexFactor pn = (VertexFactor) inf.probNecessity(X,Y,1,0);
Expand Down
178 changes: 178 additions & 0 deletions papers/21why/examples/PearlIncompatibleUxUy.java
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,178 @@
package examples;

import ch.idsia.credici.IO;
import ch.idsia.credici.inference.CausalInference;
import ch.idsia.credici.inference.CausalMultiVE;
import ch.idsia.credici.model.StructuralCausalModel;
import ch.idsia.credici.model.builder.EMCredalBuilder;
import ch.idsia.credici.utility.DataUtil;
import ch.idsia.credici.utility.FactorUtil;
import ch.idsia.credici.utility.Probability;
import ch.idsia.crema.factor.bayesian.BayesianFactor;
import ch.idsia.crema.factor.credal.vertex.VertexFactor;
import ch.idsia.crema.model.graphical.SparseModel;
import com.opencsv.exceptions.CsvException;
import gnu.trove.map.TIntIntMap;
import jdk.jshell.spi.ExecutionControl;

import java.io.IOException;
import java.util.HashMap;

public class PearlIncompatibleUxUy {
public static void main(String[] args) throws InterruptedException, ExecutionControl.NotImplementedException, IOException, CsvException {

String prj_folder = "/Users/rcabanas/GoogleDrive/IDSIA/causality/dev/credici/";


String folder = prj_folder+"papers/21why/";

TIntIntMap[] data = (TIntIntMap[])DataUtil.fromCSV(folder+ "examples/dataPearl.csv");
StructuralCausalModel mcons = (StructuralCausalModel)IO.readUAI(folder+ "examples/consPearl.uai");


// states x, y, z = True = 0
int x=1, y=1, z=1;

// states x', y', z' = False = 1
int x_=0, y_=0, z_=0;

int X=0, Y=1, Z=2;
int Ux = mcons.getExogenousParents(X)[0]; //3
int Uy = mcons.getExogenousParents(Y)[0]; //4
int Uz = mcons.getExogenousParents(Z)[0]; //5



HashMap empData = DataUtil.getEmpiricalMap(mcons, data);
empData = FactorUtil.fixEmpiricalMap(empData,6);
System.out.println(empData);


StructuralCausalModel mxy = mcons.copy();


// Alter the Ux
mxy.removeVariable(Ux);
mxy.addVariable(Ux, 3, true);


BayesianFactor f = new BayesianFactor(mxy.getDomain(Ux,X,Z));

f = f.reorderDomain(Ux,X,Z);
f.setData(new double[]{0,0,1, 1,1,0, 1,0,0, 0,1,1});

f.filter(Z,0).filter(X,0);
f.filter(Z,0).filter(X,1);
f.filter(Z,1).filter(X,0);
f.filter(Z,1).filter(X,1);

f = f.reorderDomain(mcons.getFactor(X).getDomain().getVariables());

mxy.setFactor(X, f);

////// Alter fy


f = mxy.getFactor(Y);

f = f.reorderDomain(Uy,Y,X,Z);

f.getData();

f.setData(new double[]{
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0,
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, //
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, //
0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, //
1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, //
1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0,
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0
});

f.filter(X,0).filter(Z,1).filter(Y,0);
f.filter(X,0).filter(Z,1).filter(Y,1);
f.filter(X,1).filter(Z,0).filter(Y,0);
f.filter(X,1).filter(Z,0).filter(Y,1);

f = f.reorderDomain(mcons.getFactor(Y).getDomain().getVariables());

mxy.setFactor(Y, f);



//f.reorderDomain(mcons.getFactor(X).getDomain().getVariables())


SparseModel vmcons = mcons.toVCredal(empData.values());

// not feasible as expected
//mxy.toVCredal(empData.values());



EMCredalBuilder builder = new EMCredalBuilder(mxy, data, empData)
.setMaxEMIter(200)
.setNumTrajectories(20)
//.setNumDecimalsRound(-1)
.build();


CausalMultiVE inf2 = new CausalMultiVE(builder.getSelectedPoints());


VertexFactor pns2 = (VertexFactor) inf2.probNecessityAndSufficiency(X,Y,x, x_);


System.out.println("EM method: Ux ternary incompatible and Uy 16 states incompatible too\n");
System.out.println("====================");

System.out.println("pns="+pns2);

for(StructuralCausalModel m : builder.getSelectedPoints())
System.out.println(m.getEmpiricalMap());


for(StructuralCausalModel m : builder.getSelectedPoints())
System.out.println(m.getEmpiricalMap());



for(CausalInference i : inf2.getInferenceList()){
BayesianFactor pns = (BayesianFactor) i.probNecessityAndSufficiency(X,Y,x,x_);
System.out.println(pns);
}


StructuralCausalModel m = builder.getSelectedPoints().get(0);


System.out.println("P(U)");
for(int u : m.getExogenousVars())
FactorUtil.print(m.getFactor(u));

HashMap empModel = m.getEmpiricalMap();

System.out.println("Empiricals: ");
for(BayesianFactor p : m.getEmpiricalMap().values())
FactorUtil.print(p);


BayesianFactor pns = (BayesianFactor) inf2.getInferenceList().get(0).probNecessityAndSufficiency(X,Y,x, x_);

double maxll = Probability.logLikelihood(empData, empData, 1);
double mll = Probability.logLikelihood(empModel, empData, 1);
double ratio = maxll / mll;

System.out.println("PNS by this model = "+pns);
System.out.println("Max LL by this data = "+maxll);
System.out.println("LL by this model and data = "+mll);
System.out.println("Ratio = "+ratio);






}
}
200 changes: 200 additions & 0 deletions papers/21why/examples/PearlRestrictedVU.java
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,200 @@
package examples;

import ch.idsia.credici.IO;
import ch.idsia.credici.factor.EquationBuilder;
import ch.idsia.credici.inference.CausalMultiVE;
import ch.idsia.credici.inference.CausalVE;
import ch.idsia.credici.inference.CredalCausalVE;
import ch.idsia.credici.model.StructuralCausalModel;
import ch.idsia.credici.model.builder.CausalBuilder;
import ch.idsia.credici.model.builder.EMCredalBuilder;
import ch.idsia.credici.utility.DataUtil;
import ch.idsia.credici.utility.FactorUtil;
import ch.idsia.credici.utility.Probability;
import ch.idsia.crema.factor.bayesian.BayesianFactor;
import ch.idsia.crema.factor.credal.linear.SeparateHalfspaceFactor;
import ch.idsia.crema.factor.credal.vertex.VertexFactor;
import ch.idsia.crema.model.graphical.SparseModel;
import ch.idsia.crema.model.graphical.specialized.BayesianNetwork;
import gnu.trove.map.TIntIntMap;
import jdk.jshell.spi.ExecutionControl;
import org.apache.commons.lang3.ArrayUtils;
import org.junit.Assert;

import java.io.IOException;
import java.nio.file.Path;
import java.util.HashMap;
import java.util.stream.IntStream;

public class PearlRestrictedVU {
public static void main(String[] args) throws IOException, InterruptedException, ExecutionControl.NotImplementedException {

String wdir = "/Users/rcabanas/GoogleDrive/IDSIA/causality/dev/credici";
String folder = Path.of(wdir, "papers/21why/examples/").toString();

BayesianNetwork bnet = new BayesianNetwork();

int X = bnet.addVariable(2);
int Y = bnet.addVariable(2);
int Z = bnet.addVariable(2);

bnet.addParents(X, Z);
bnet.addParents(Y, X, Z);


// states x, y, z = True = 1
int x=1, y=1, z=1;

// states x', y', z' = False = 0
int x_=0, y_=0, z_=0;


// Set data counts
BayesianFactor counts = new BayesianFactor(bnet.getDomain(Z,X,Y));
counts.setValue(2, z_, x_, y_);
counts.setValue(114, z_, x_, y);
counts.setValue(41, z_, x, y_);
counts.setValue(313, z_, x, y);

counts.setValue(107, z, x_, y_);
counts.setValue(13, z, x_, y);
counts.setValue(109, z, x, y_);
counts.setValue(1, z, x, y);

FactorUtil.print(counts);


int N = (int) counts.marginalize(X, Y, Z).getValueAt(0);
Assert.assertEquals(700, N, 0);

// Compute the CPTs of the endogenous model
BayesianFactor nz = counts.marginalize(X, Y);
BayesianFactor nxz = counts.marginalize(Y);
BayesianFactor nxy = counts.marginalize(Z);
BayesianFactor px_z = nxz.divide(nz);
BayesianFactor py_xz = counts.reorderDomain(0,1,2).divide(nxz.reorderDomain(X,Z));
BayesianFactor pz = nz.scalarMultiply(1.0/N);
BayesianFactor py_x = nxy.divide(nxy.marginalize(Y));
BayesianFactor pxy = nxy.scalarMultiply(1.0/N);


//Generate a dataset with the require counts
TIntIntMap[] data = DataUtil.dataFromCounts(counts);

// Empirical network
bnet.setFactor(Z, pz);
bnet.setFactor(Y, py_xz);
bnet.setFactor(X, px_z);


// Conservative model
StructuralCausalModel m_eqless = CausalBuilder.of(bnet).build();
BayesianFactor fx = EquationBuilder.of(m_eqless).fromVector(X, 1,0, 0,0, 0,1, 1,1);
m_eqless.setFactor(X, fx);
m_eqless.fillExogenousWithRandomFactors(3);

m_eqless.getFactor(X).filter(X, x_).filter(Z, z_);
m_eqless.getFactor(X).filter(X, x).filter(Z, z_);
m_eqless.getFactor(X).filter(X, x_).filter(Z, z);
m_eqless.getFactor(X).filter(X, x).filter(Z, z);

// Exogenous variables
int U = m_eqless.getExogenousParents(X)[0];
int V = m_eqless.getExogenousParents(Y)[0];
int W = m_eqless.getExogenousParents(Z)[0];

HashMap empData = DataUtil.getEmpiricalMap(m_eqless, data);
empData = FactorUtil.fixEmpiricalMap(empData,6);

System.out.println(empData);

SparseModel vmodelPGM = m_eqless.toVCredal(empData.values());
System.out.println(m_eqless);
System.out.println("vmodel PGM:");
System.out.println(vmodelPGM);


SparseModel hmodelPGM = m_eqless.toHCredal(empData.values());
System.out.println("hmodel PGM:");
System.out.println(hmodelPGM);
((SeparateHalfspaceFactor)hmodelPGM.getFactor(U)).printLinearProblem();


CredalCausalVE inf = new CredalCausalVE(m_eqless, empData.values());
IO.write(m_eqless, folder+"consPearl.uai");
DataUtil.toCSV(folder+"dataPearl.csv", data);


VertexFactor pn = (VertexFactor) inf.probNecessity(X,Y,1,0);
VertexFactor ps = (VertexFactor) inf.probSufficiency(X,Y,1,0);
VertexFactor pns = (VertexFactor) inf.probNecessityAndSufficiency(X,Y,1,0);


System.out.println("PGM exact method");
System.out.println("====================");
System.out.println("pn="+pn);
System.out.println("ps="+ps);
System.out.println("pns="+pns);



// EM
for(int u: m_eqless.getExogenousVars())
System.out.println(u+": "+m_eqless.getFactor(u));

EMCredalBuilder builder = EMCredalBuilder.of(m_eqless, data)
.setMaxEMIter(200)
.setNumTrajectories(20)
//.setNumDecimalsRound(-1)
.build();


CausalMultiVE inf2 = new CausalMultiVE(builder.getSelectedPoints());
VertexFactor pns2 = (VertexFactor) inf2.probNecessityAndSufficiency(X,Y,1,0);
System.out.println("pns2="+pns2);



//Start from the non-conservative model
StructuralCausalModel m_reduced = m_eqless.copy();
// Modify the SEs and exogenous domains
m_reduced.removeVariable(U);
m_reduced.removeVariable(V);
m_reduced.addVariable(U,3,true);
m_reduced.addVariable(V,3,true);
m_reduced.addParents(X,U);
m_reduced.addParents(Y,V);

fx = EquationBuilder.of(m_reduced).fromVector(X, 1,0, 0,0, 0,1);
m_reduced.setFactor(X, fx);
BayesianFactor fy = EquationBuilder.of(m_reduced).fromVector(Y, 0,0,1,0, 0,0,0,0, 1,1,0,1);
m_reduced.setFactor(Y, fy);

// Set P(U)
m_reduced.setFactor(W, new BayesianFactor(m_reduced.getDomain(W), new double[]{0.6714289999999999, 0.328571}));
m_reduced.setFactor(V, new BayesianFactor(m_reduced.getDomain(V), new double[]{0.091,0.448,0.462}));
m_reduced.setFactor(U, new BayesianFactor(m_reduced.getDomain(U), new double[]{0.677,0.000,0.323}));

m_reduced.getFactor(X);

CausalVE cve = new CausalVE(m_reduced);
double pns_nocompatible = cve.probNecessityAndSufficiency(X,Y, x,x_).getValueAt(0);



double maxll = Probability.logLikelihood(empData, empData, 1);
double mll = Probability.logLikelihood(m_reduced.getEmpiricalMap(), empData, 1);
double ratio = maxll / mll;



System.out.println("PNS in non-compatible model = "+pns_nocompatible);
System.out.println("PNS intervals = "+pns );

System.out.println("Max LL by this data = "+maxll);
System.out.println("LL by this model and data = "+mll);
System.out.println("Ratio = "+ratio);


}
}
Loading

0 comments on commit 626daf3

Please sign in to comment.