vb runnable from trainer

git-svn-id: https://ws10smt.googlecode.com/svn/trunk@380 ec762483-ff6d-05da-a07a-a48fb63a330f

2 files changed, 92 insertions, 48 deletions

diff --git a/gi/posterior-regularisation/prjava/src/phrase/Trainer.java b/gi/posterior-regularisation/prjava/src/phrase/Trainer.java
index b51db919..cea6a20a 100644
--- a/gi/posterior-regularisation/prjava/src/phrase/Trainer.java
+++ b/gi/posterior-regularisation/prjava/src/phrase/Trainer.java
@@ -18,7 +18,7 @@ public class Trainer
 {
 	public static void main(String[] args) 
 	{
-
+		
         OptionParser parser = new OptionParser();
         parser.accepts("help");
         parser.accepts("in").withRequiredArg().ofType(File.class);
@@ -107,6 +107,7 @@ public class Trainer
  		PhraseCluster cluster = null;
  		Agree2Sides agree2sides = null;
  		Agree agree= null;
+ 		VB vbModel=null;
  		if (options.has("agree-language"))
  			agree2sides = new Agree2Sides(tags, corpus,corpus1);
  		else if (options.has("agree-direction"))
@@ -115,7 +116,11 @@ public class Trainer
  		{
  			cluster = new PhraseCluster(tags, corpus);
  			if (threads > 0) cluster.useThreadPool(threads);
- 			if (vb)	cluster.initialiseVB(alphaEmit, alphaPi);
+ 			
+ 			if (vb)	{
+ 				//cluster.initialiseVB(alphaEmit, alphaPi);
+ 				vbModel=new VB(tags,corpus);
+ 			}
  			if (options.has("no-parameter-cache")) 
  				cluster.cacheLambda = false;
  			if (options.has("start"))
@@ -149,7 +154,7 @@ public class Trainer
 					if (!vb)
 						o = cluster.EM((i < skip) ? i+1 : 0);
 					else
-						o = cluster.VBEM(alphaEmit, alphaPi);	
+						o = vbModel.EM();	
 				}
 				else
 					o = cluster.PREM(scale_phrase, scale_context, (i < skip) ? i+1 : 0);
@@ -166,10 +171,8 @@ public class Trainer
 			last = o;
 		}
 		
-		if (cluster == null && agree != null)
+		if (cluster == null)
 			cluster = agree.model1;
-		else if (cluster == null && agree2sides != null)
-			cluster = agree2sides.model1;
 
 		double pl1lmax = cluster.phrase_l1lmax();
 		double cl1lmax = cluster.context_l1lmax();
@@ -180,26 +183,20 @@ public class Trainer
 			File outfile = (File) options.valueOf("out");
 			try {
 				PrintStream ps = FileUtil.printstream(outfile);
-				List<Edge> test = corpus.getEdges();
-				if (options.has("test")) // just use the training
+				List<Edge> test;
+				if (!options.has("test")) // just use the training
+					test = corpus.getEdges();
+				else
 				{	// if --test supplied, load up the file
-					if (agree2sides == null)
-					{
-						infile = (File) options.valueOf("test");
-						System.out.println("Reading testing concordance from " + infile);
-						test = corpus.readEdges(FileUtil.reader(infile));
-					}
-					else
-						System.err.println("Can't run bilingual agreement model on different test data cf training (yet); --test ignored.");
+					infile = (File) options.valueOf("test");
+					System.out.println("Reading testing concordance from " + infile);
+					test = corpus.readEdges(FileUtil.reader(infile));
 				}
-				
-				if (agree != null)
-					agree.displayPosterior(ps, test);
-				else if (agree2sides != null)
-					agree2sides.displayPosterior(ps);
-				else
+				if(vb){
+					vbModel.displayPosterior(ps);
+				}else{
 					cluster.displayPosterior(ps, test);
-					
+				}
 				ps.close();
 			} catch (IOException e) {
 				System.err.println("Failed to open either testing file or output file");
diff --git a/gi/posterior-regularisation/prjava/src/phrase/VB.java b/gi/posterior-regularisation/prjava/src/phrase/VB.java
index cc1c1c96..a858c883 100644
--- a/gi/posterior-regularisation/prjava/src/phrase/VB.java
+++ b/gi/posterior-regularisation/prjava/src/phrase/VB.java
@@ -16,7 +16,7 @@ import phrase.Corpus.Edge;
 

 public class VB {

 

-	public static int MAX_ITER=40;

+	public static int MAX_ITER=400;

 	

 	/**@brief

 	 * hyper param for beta

@@ -28,11 +28,13 @@ public class VB {
 	 * hyper param for theta

 	 * where theta is dirichlet for z

 	 */

-	public double alpha=0.000001;

+	public double alpha=0.0001;

 	/**@brief

 	 * variational param for beta

 	 */

 	private double rho[][][];

+	private double digamma_rho[][][];

+	private double rho_sum[][];

 	/**@brief

 	 * variational param for z

 	 */

@@ -41,8 +43,7 @@ public class VB {
 	 * variational param for theta

 	 */

 	private double gamma[];

-	

-	private static double VAL_DIFF_RATIO=0.001;

+	private static double VAL_DIFF_RATIO=0.005;

 	

 	/**@brief

 	 * objective for a single document

@@ -55,8 +56,8 @@ public class VB {
 	

 	private Corpus c;

 	public static void main(String[] args) {

-		String in="../pdata/canned.con";

-		//String in="../pdata/btec.con";

+	//	String in="../pdata/canned.con";

+		String in="../pdata/btec.con";

 		String out="../pdata/vb.out";

 		int numCluster=25;

 		Corpus corpus = null;

@@ -118,6 +119,7 @@ public class VB {
 				}

 			}

 		}

+		

 	}

 	

 	private void inference(int phraseID){

@@ -128,26 +130,21 @@ public class VB {
 				phi[i][j]=1.0/K;

 			}

 		}

-		gamma = new double[K];

-		double digamma_gamma[]=new double[K];

-		for(int i=0;i<gamma.length;i++){

-			gamma[i] = alpha + 1.0/K;

+		if(gamma==null){

+			gamma=new double[K];

 		}

+		Arrays.fill(gamma,alpha+1.0/K);

 		

-		double rho_sum[][]=new double [K][n_positions];

-		for(int i=0;i<K;i++){

-			for(int pos=0;pos<n_positions;pos++){

-				rho_sum[i][pos]=Gamma.digamma(arr.F.l1norm(rho[i][pos]));

-			}

-		}

-		double gamma_sum=Gamma.digamma(arr.F.l1norm(gamma));

+		double digamma_gamma[]=new double[K];

+		

+		double gamma_sum=digamma(arr.F.l1norm(gamma));

 		for(int i=0;i<K;i++){

-			digamma_gamma[i]=Gamma.digamma(gamma[i]);

+			digamma_gamma[i]=digamma(gamma[i]);

 		}

 		double gammaSum[]=new double [K];

-		

 		double prev_val=0;

 		obj=0;

+		

 		for(int iter=0;iter<MAX_ITER;iter++){

 			prev_val=obj;

 			obj=0;

@@ -159,7 +156,7 @@ public class VB {
 					double sum=0;

 					for(int pos=0;pos<n_positions;pos++){

 						int word=context.get(pos);

-						sum+=Gamma.digamma(rho[i][pos][word])-rho_sum[i][pos];

+						sum+=digamma_rho[i][pos][word]-rho_sum[i][pos];

 					}

 					sum+= digamma_gamma[i]-gamma_sum;

 					phi[n][i]=sum;

@@ -183,11 +180,12 @@ public class VB {
 			for(int i=0;i<K;i++){

 				gamma[i]=alpha+gammaSum[i];

 			}

-			gamma_sum=Gamma.digamma(arr.F.l1norm(gamma));

+			gamma_sum=digamma(arr.F.l1norm(gamma));

 			for(int i=0;i<K;i++){

-				digamma_gamma[i]=Gamma.digamma(gamma[i]);

+				digamma_gamma[i]=digamma(gamma[i]);

 			}

 			//compute objective for reporting

+

 			obj=0;

 			

 			for(int i=0;i<K;i++){

@@ -209,13 +207,13 @@ public class VB {
 					double beta_sum=0;

 					for(int pos=0;pos<n_positions;pos++){

 						int word=context.get(pos);

-						beta_sum+=(Gamma.digamma(rho[i][pos][word])-rho_sum[i][pos]);

+						beta_sum+=(digamma(rho[i][pos][word])-rho_sum[i][pos]);

 					}

 					obj+=phi[n][i]*beta_sum;

 				}

 			}

 			

-			obj-=Gamma.logGamma(arr.F.l1norm(gamma));

+			obj-=log_gamma(arr.F.l1norm(gamma));

 			for(int i=0;i<K;i++){

 				obj+=Gamma.logGamma(gamma[i]);

 				obj-=(gamma[i]-1)*(digamma_gamma[i]-gamma_sum);

@@ -233,6 +231,26 @@ public class VB {
 	 */

 	public double EM(){

 		double emObj=0;

+		if(digamma_rho==null){

+			digamma_rho=new double[K][n_positions][n_words];

+		}

+		for(int i=0;i<K;i++){

+			for (int pos=0;pos<n_positions;pos++){

+				for(int j=0;j<n_words;j++){

+					digamma_rho[i][pos][j]= digamma(rho[i][pos][j]);

+				}

+			}

+		}

+		

+		if(rho_sum==null){

+			rho_sum=new double [K][n_positions];

+		}

+		for(int i=0;i<K;i++){

+			for(int pos=0;pos<n_positions;pos++){

+				rho_sum[i][pos]=digamma(arr.F.l1norm(rho[i][pos]));

+			}

+		}

+

 		

 		//E

 		double exp_rho[][][]=new double[K][n_positions][n_words];

@@ -248,7 +266,13 @@ public class VB {
 					}

 				}

 			}

-			

+/*			if(d!=0 && d%100==0){

+				System.out.print(".");

+			}

+			if(d!=0 && d%1000==0){

+				System.out.println(d);

+			}

+*/

 			emObj+=obj;

 		}

 		

@@ -313,5 +337,28 @@ public class VB {
 	  }

 	  return(v);

 	}

+		

+	double digamma(double x)

+	{

+	    double p;

+	    x=x+6;

+	    p=1/(x*x);

+	    p=(((0.004166666666667*p-0.003968253986254)*p+

+		0.008333333333333)*p-0.083333333333333)*p;

+	    p=p+Math.log(x)-0.5/x-1/(x-1)-1/(x-2)-1/(x-3)-1/(x-4)-1/(x-5)-1/(x-6);

+	    return p;

+	}

+	

+	double log_gamma(double x)

+	{

+	     double z=1/(x*x);

+

+	    x=x+6;

+	    z=(((-0.000595238095238*z+0.000793650793651)

+		*z-0.002777777777778)*z+0.083333333333333)/x;

+	    z=(x-0.5)*Math.log(x)-x+0.918938533204673+z-Math.log(x-1)-

+	    Math.log(x-2)-Math.log(x-3)-Math.log(x-4)-Math.log(x-5)-Math.log(x-6);

+	    return z;

+	}

 	

 }//End of  class