Use the same neural model for validation.

psl-core/src/main/java/org/linqs/psl/application/learning/weight/WeightLearningApplication.java

@@ -22,6 +22,9 @@
 import org.linqs.psl.config.Options;
 import org.linqs.psl.database.Database;
 import org.linqs.psl.evaluation.EvaluationInstance;
+import org.linqs.psl.model.deep.DeepModelPredicate;
+import org.linqs.psl.model.predicate.DeepPredicate;
+import org.linqs.psl.model.predicate.Predicate;
 import org.linqs.psl.model.rule.Rule;
 import org.linqs.psl.model.rule.WeightedRule;
 import org.linqs.psl.util.Logger;
@@ -54,6 +57,10 @@ public abstract class WeightLearningApplication implements ModelApplication {
     protected Database validationTargetDatabase;
     protected Database validationTruthDatabase;
 
+    protected List<DeepPredicate> deepPredicates;
+    protected List<DeepModelPredicate> deepModelPredicates;
+    protected List<DeepModelPredicate> validationDeepModelPredicates;
+
     protected boolean runValidation;
 
     protected List<Rule> allRules;
@@ -87,6 +94,10 @@ public WeightLearningApplication(List<Rule> rules, Database trainTargetDatabase,
 
         this.runValidation = runValidation;
 
+        deepPredicates = new ArrayList<DeepPredicate>();
+        deepModelPredicates = new ArrayList<DeepModelPredicate>();
+        validationDeepModelPredicates = new ArrayList<DeepModelPredicate>();
+
         allRules = new ArrayList<Rule>();
         mutableRules = new ArrayList<WeightedRule>();
 
@@ -154,11 +165,8 @@ protected void initGroundModel() {
 
         InferenceApplication trainInferenceApplication = InferenceApplication.getInferenceApplication(Options.WLA_INFERENCE.getString(), allRules, trainTargetDatabase);
         trainInferenceApplication.loadDeepPredicates("learning");
-        InferenceApplication validationInferenceApplication = InferenceApplication.getInferenceApplication(Options.WLA_INFERENCE.getString(), allRules, validationTargetDatabase);
 
-        if (runValidation) {
-            validationInferenceApplication.loadDeepPredicates("inference");
-        }
+        InferenceApplication validationInferenceApplication = InferenceApplication.getInferenceApplication(Options.WLA_INFERENCE.getString(), allRules, validationTargetDatabase);
 
         initGroundModel(trainInferenceApplication, validationInferenceApplication);
     }
@@ -195,6 +203,17 @@ public void initGroundModel(InferenceApplication trainInferenceApplication, Trai
             initRandomWeights();
         }
 
+        for (Predicate predicate : Predicate.getAll()) {
+            if (predicate instanceof DeepPredicate) {
+                deepPredicates.add((DeepPredicate)predicate);
+                deepModelPredicates.add(((DeepPredicate)predicate).getDeepModel());
+
+                DeepModelPredicate validationDeepModelPredicate = ((DeepPredicate)predicate).getDeepModel().copy();
+                validationDeepModelPredicate.setAtomStore(validationInferenceApplication.getDatabase().getAtomStore(), true);
+                validationDeepModelPredicates.add(((DeepPredicate)predicate).getDeepModel().copy());
+            }
+        }
+
         postInitGroundModel();
 
         groundModelInit = true;

psl-core/src/main/java/org/linqs/psl/application/learning/weight/gradient/GradientDescent.java

@@ -22,6 +22,7 @@
 import org.linqs.psl.config.Options;
 import org.linqs.psl.database.AtomStore;
 import org.linqs.psl.database.Database;
+import org.linqs.psl.model.deep.DeepModelPredicate;
 import org.linqs.psl.model.predicate.DeepPredicate;
 import org.linqs.psl.model.predicate.Predicate;
 import org.linqs.psl.model.rule.GroundRule;
@@ -67,8 +68,6 @@ public static enum GDExtension {
     protected float[] MAPRVAtomGradient;
     protected float[] MAPDeepAtomGradient;
 
-    protected List<DeepPredicate> deepPredicates;
-
     protected TermState[] trainMAPTermState;
     protected float[] trainMAPAtomValueState;
 
@@ -111,8 +110,6 @@ public GradientDescent(List<Rule> rules, Database trainTargetDatabase, Database
         MAPRVAtomGradient = null;
         MAPDeepAtomGradient = null;
 
-        deepPredicates = new ArrayList<DeepPredicate>();
-
         trainMAPTermState = null;
         trainMAPAtomValueState = null;
 
@@ -176,12 +173,6 @@ protected void postInitGroundModel() {
 
         MAPRVAtomGradient = new float[trainAtomValues.length];
         MAPDeepAtomGradient = new float[trainAtomValues.length];
-
-        for (Predicate predicate : Predicate.getAll()) {
-            if (predicate instanceof DeepPredicate) {
-                deepPredicates.add((DeepPredicate)predicate);
-            }
-        }
     }
 
     protected void initForLearning() {
@@ -238,6 +229,7 @@ protected void doLearn() {
             gradientStep(iteration);
 
             if (log.isTraceEnabled() && (evaluation != null)) {
+                log.trace("Running Inference.");
                 // Compute the MAP state before evaluating so variables have assigned values.
                 computeMAPStateWithWarmStart(trainInferenceApplication, trainMAPTermState, trainMAPAtomValueState);
                 inTrainingMAPState = true;
@@ -257,17 +249,16 @@ protected void doLearn() {
             }
 
             if (runValidation) {
-                for (DeepPredicate deepPredicate : deepPredicates) {
+                for (int i = 0; i < deepPredicates.size(); i++) {
+                    DeepPredicate deepPredicate = deepPredicates.get(i);
+                    deepPredicate.setDeepModel(validationDeepModelPredicates.get(i));
                     deepPredicate.predictDeepModel(false);
                 }
 
+                log.trace("Running Validation Inference.");
                 computeMAPStateWithWarmStart(validationInferenceApplication, validationMAPTermState, validationMAPAtomValueState);
                 inValidationMAPState = true;
 
-                for (DeepPredicate deepPredicate : deepPredicates) {
-                    deepPredicate.predictDeepModel(true);
-                }
-
                 evaluation.compute(validationMap);
                 currentValidationEvaluationMetric = evaluation.getNormalizedRepMetric();
                 log.debug("MAP State Validation Evaluation Metric: {}", currentValidationEvaluationMetric);
@@ -284,6 +275,12 @@ protected void doLearn() {
                 }
 
                 log.debug("MAP State Best Validation Evaluation Metric: {}", bestValidationEvaluationMetric);
+
+                for (int i = 0; i < deepPredicates.size(); i++) {
+                    DeepPredicate deepPredicate = deepPredicates.get(i);
+                    deepPredicate.setDeepModel(deepModelPredicates.get(i));
+                    deepPredicate.predictDeepModel(true);
+                }
             }
 
             computeIterationStatistics();

psl-core/src/main/java/org/linqs/psl/application/learning/weight/gradient/minimizer/Minimizer.java

@@ -362,7 +362,7 @@ protected void computeProxRuleObservedAtomValueGradient() {
      * Compute the incompatibility of the mpe state and the gradient of the energy function at the mpe state.
      */
     private void computeFullInferenceStatistics() {
-        log.trace("Running Full Inference.");
+        log.trace("Running Inference.");
         computeMAPStateWithWarmStart(trainInferenceApplication, trainMAPTermState, trainMAPAtomValueState);
         inTrainingMAPState = true;
 

psl-core/src/main/java/org/linqs/psl/model/deep/DeepModel.java

@@ -46,20 +46,20 @@ public abstract class DeepModel {
     private static int startingPort = -1;
     private static Map<Integer, DeepModel> usedPorts = null;
 
-    private String deepModel;
+    protected String deepModel;
     protected Map<String, String> pythonOptions;
     protected String application;
 
-    private int port;
-    private String pythonModule;
-    private String sharedMemoryPath;
-    private Process pythonServerProcess;
-    private RandomAccessFile sharedFile;
+    protected int port;
+    protected String pythonModule;
+    protected String sharedMemoryPath;
+    protected Process pythonServerProcess;
+    protected RandomAccessFile sharedFile;
     protected MappedByteBuffer sharedBuffer;
-    private Socket socket;
-    private BufferedReader socketInput;
-    private PrintWriter socketOutput;
-    private boolean serverOpen;
+    protected Socket socket;
+    protected BufferedReader socketInput;
+    protected PrintWriter socketOutput;
+    protected boolean serverOpen;
 
     protected DeepModel(String deepModel) {
         this.deepModel = deepModel;
@@ -404,7 +404,7 @@ private static synchronized int getOpenPort(DeepModel model) {
         return port;
     }
 
-    private static synchronized void freePort(int port) {
+    protected static synchronized void freePort(int port) {
         usedPorts.remove(Integer.valueOf(port));
     }
 }

psl-core/src/main/java/org/linqs/psl/model/deep/DeepModelPredicate.java

@@ -68,6 +68,41 @@ public DeepModelPredicate(Predicate predicate) {
         this.validDataIndexes = new ArrayList<Integer>();
     }
 
+    public DeepModelPredicate copy() {
+        DeepModelPredicate copy = new DeepModelPredicate(predicate);
+
+        copy.pythonOptions = pythonOptions;
+
+        copy.application = application;
+
+        freePort(copy.port);
+        copy.port = (this.port);
+
+        copy.pythonModule = pythonModule;
+        copy.sharedMemoryPath = sharedMemoryPath;
+        copy.pythonServerProcess = pythonServerProcess;
+        copy.sharedFile = sharedFile;
+        copy.sharedBuffer = sharedBuffer;
+        copy.socket = socket;
+        copy.socketInput = socketInput;
+        copy.socketOutput = socketOutput;
+        copy.serverOpen = serverOpen;
+
+        copy.atomStore = atomStore;
+
+        copy.classSize = classSize;
+        copy.atomIndexes = atomIndexes;
+        copy.dataIndexes = dataIndexes;
+
+        copy.validAtomIndexes = validAtomIndexes;
+        copy.validDataIndexes = validDataIndexes;
+
+        copy.gradients = gradients;
+        copy.symbolicGradients = symbolicGradients;
+
+        return copy;
+    }
+
     public synchronized int init() {
         log.debug("Initializing deep model predicate: {}", predicate.getName());
 
@@ -160,7 +195,15 @@ public synchronized void close() {
     }
 
     public void setAtomStore(AtomStore atomStore) {
+        setAtomStore(atomStore, false);
+    }
+
+    public void setAtomStore(AtomStore atomStore, boolean init) {
         this.atomStore = atomStore;
+
+        if (init) {
+            init();
+        }
     }
 
     public void setSymbolicGradients(float[] symbolicGradients) {

psl-core/src/main/java/org/linqs/psl/model/predicate/DeepPredicate.java

@@ -51,6 +51,14 @@ public void fitDeepPredicate(float[] symbolicGradients) {
         deepModel.fitDeepModel();
     }
 
+    public DeepModelPredicate getDeepModel() {
+        return deepModel;
+    }
+
+    public void setDeepModel(DeepModelPredicate deepModel) {
+        this.deepModel = deepModel;
+    }
+
     public float predictDeepModel() {
         return deepModel.predictDeepModel(false);
     }