Improved logging.

linqs · Jul 12, 2023 · 6d12e64 · 6d12e64
1 parent 2107d70
commit 6d12e64
Show file tree

Hide file tree

Showing 5 changed files with 74 additions and 58 deletions.
diff --git a/...ore/src/main/java/org/linqs/psl/application/learning/weight/gradient/GradientDescent.java b/...ore/src/main/java/org/linqs/psl/application/learning/weight/gradient/GradientDescent.java
@@ -24,6 +24,7 @@
 import org.linqs.psl.database.Database;
 import org.linqs.psl.model.predicate.DeepPredicate;
 import org.linqs.psl.model.predicate.Predicate;
+import org.linqs.psl.model.rule.GroundRule;
 import org.linqs.psl.model.rule.Rule;
 import org.linqs.psl.model.rule.WeightedRule;
 import org.linqs.psl.reasoner.InitialValue;
@@ -229,7 +230,10 @@ protected void doLearn() {
         while (!breakGD) {
             long start = System.currentTimeMillis();
 
-            log.trace("Model: {}", mutableRules);
+            log.trace("Model:");
+            for (WeightedRule weightedRule: mutableRules) {
+                log.trace("{}", weightedRule);
+            }
 
             gradientStep(iteration);
 
@@ -464,6 +468,8 @@ protected float atomGradientStep() {
             deepPredicate.fitDeepPredicate(deepAtomGradient);
             deepPredicateChange += deepPredicate.predictDeepModel(true);
         }
+
+        log.trace("Deep Predicate Change: {}", deepPredicateChange);
         return deepPredicateChange;
     }
 

diff --git a/...src/main/java/org/linqs/psl/application/learning/weight/gradient/minimizer/Minimizer.java b/...src/main/java/org/linqs/psl/application/learning/weight/gradient/minimizer/Minimizer.java
@@ -244,12 +244,12 @@ protected void gradientStep(int iteration) {
         parameterMovement += internalParameterGradientStep(iteration);
         parameterMovement += atomGradientStep();
 
+        // Update the penalty coefficients and tolerance.
+        float totalObjectiveDifference = computeObjectiveDifference();
+
         if ((iteration > 0) && (parameterMovement < parameterMovementTolerance)) {
             outerIteration++;
 
-            // Update the penalty coefficients and tolerance.
-            float totalObjectiveDifference = computeObjectiveDifference();
-
             if (totalObjectiveDifference < constraintTolerance) {
                 if ((totalObjectiveDifference < finalConstraintTolerance) && (parameterMovement < finalParameterMovementTolerance)) {
                     // Learning has converged.
@@ -263,11 +263,10 @@ protected void gradientStep(int iteration) {
                 constraintTolerance = (float)(1.0f / Math.pow(squaredPenaltyCoefficient, 0.1));
                 parameterMovementTolerance = (float)(1.0f / squaredPenaltyCoefficient);
             }
-
-            log.trace("Outer iteration: {}, Objective Difference: {}, Parameter Movement: {}, Squared Penalty Coefficient: {}, Linear Penalty Coefficient: {}, Constraint Tolerance: {}, parameterMovementTolerance: {}.",
-                    outerIteration, totalObjectiveDifference, parameterMovement, squaredPenaltyCoefficient, linearPenaltyCoefficient, constraintTolerance, parameterMovementTolerance);
-
         }
+
+        log.trace("Outer iteration: {}, Objective Difference: {}, Parameter Movement: {}, Squared Penalty Coefficient: {}, Linear Penalty Coefficient: {}, Constraint Tolerance: {}, parameterMovementTolerance: {}.",
+                outerIteration, totalObjectiveDifference, parameterMovement, squaredPenaltyCoefficient, linearPenaltyCoefficient, constraintTolerance, parameterMovementTolerance);
     }
 
     @Override
@@ -286,6 +285,8 @@ protected float internalParameterGradientStep(int iteration) {
             augmentedInferenceAtomValueState[proxRuleObservedAtomIndexes[i]] = newProxRuleObservedAtomsValue;
         }
 
+        log.trace("Proximity Rule Observed Atoms Value Movement: {}.", proxRuleObservedAtomsValueMovement);
+
         return proxRuleObservedAtomsValueMovement;
     }
 

diff --git a/...in/java/org/linqs/psl/application/learning/weight/gradient/optimalvalue/OptimalValue.java b/...in/java/org/linqs/psl/application/learning/weight/gradient/optimalvalue/OptimalValue.java
@@ -20,10 +20,12 @@
 import org.linqs.psl.application.learning.weight.gradient.GradientDescent;
 import org.linqs.psl.database.AtomStore;
 import org.linqs.psl.database.Database;
+import org.linqs.psl.model.atom.GroundAtom;
 import org.linqs.psl.model.atom.ObservedAtom;
 import org.linqs.psl.model.atom.RandomVariableAtom;
 import org.linqs.psl.model.rule.Rule;
 import org.linqs.psl.reasoner.term.TermState;
+import org.linqs.psl.util.Logger;
 
 import java.util.Arrays;
 import java.util.List;
@@ -39,6 +41,8 @@
  * computing the incompatibility of the latent variable inference problem solution are provided in this class.
  */
 public abstract class OptimalValue extends GradientDescent {
+    private static final Logger log = Logger.getLogger(GradientDescent.class);
+
     protected float[] latentInferenceIncompatibility;
     protected TermState[] latentInferenceTermState;
     protected float[] latentInferenceAtomValueState;
@@ -81,6 +85,10 @@ protected void computeLatentInferenceIncompatibility() {
         computeCurrentIncompatibility(latentInferenceIncompatibility);
         trainInferenceApplication.getReasoner().computeOptimalValueGradient(trainInferenceApplication.getTermStore(), rvLatentAtomGradient, deepLatentAtomGradient);
 
+        for (int i = 0; i < mutableRules.size(); i++) {
+            log.trace("Rule: {} , Latent inference incompatibility: {}", mutableRules.get(i), latentInferenceIncompatibility[i]);
+        }
+
         unfixLabeledRandomVariables();
     }
 

diff --git a/psl-core/src/main/java/org/linqs/psl/reasoner/sgd/term/SGDObjectiveTerm.java b/psl-core/src/main/java/org/linqs/psl/reasoner/sgd/term/SGDObjectiveTerm.java
@@ -87,54 +87,4 @@ public void read(ByteBuffer fixedBuffer) {
             atomIndexes[i] = fixedBuffer.getInt();
         }
     }
-
-    @Override
-    public String toString() {
-        return toString(null);
-    }
-
-    public String toString(AtomStore atomStore) {
-        // weight * [max(coeffs^T * x - constant, 0.0)]^2
-
-        StringBuilder builder = new StringBuilder();
-
-        builder.append(getWeight());
-        builder.append(" * ");
-
-        if (hinge) {
-            builder.append("max(0.0, ");
-        } else {
-            builder.append("(");
-        }
-
-        for (int i = 0; i < size; i++) {
-            builder.append("(");
-            builder.append(coefficients[i]);
-
-            if (atomStore == null) {
-                builder.append(" * <index:");
-                builder.append(atomIndexes[i]);
-                builder.append(">)");
-            } else {
-                builder.append(" * ");
-                builder.append(atomStore.getAtomValue(atomIndexes[i]));
-                builder.append(")");
-            }
-
-            if (i != size - 1) {
-                builder.append(" + ");
-            }
-        }
-
-        builder.append(" - ");
-        builder.append(constant);
-
-        builder.append(")");
-
-        if (squared) {
-            builder.append(" ^2");
-        }
-
-        return builder.toString();
-    }
 }
diff --git a/psl-core/src/main/java/org/linqs/psl/reasoner/term/ReasonerTerm.java b/psl-core/src/main/java/org/linqs/psl/reasoner/term/ReasonerTerm.java
@@ -17,6 +17,7 @@
  */
 package org.linqs.psl.reasoner.term;
 
+import org.linqs.psl.database.AtomStore;
 import org.linqs.psl.model.atom.GroundAtom;
 import org.linqs.psl.model.rule.Rule;
 import org.linqs.psl.model.rule.WeightedRule;
@@ -329,4 +330,54 @@ public TermState saveState() {
     public void saveState(TermState termState) {
         // Pass.
     }
+
+    @Override
+    public String toString() {
+        return toString(null);
+    }
+
+    public String toString(AtomStore atomStore) {
+        // weight * [max(coeffs^T * x - constant, 0.0)]^2
+
+        StringBuilder builder = new StringBuilder();
+
+        builder.append(getWeight());
+        builder.append(" * ");
+
+        if (hinge) {
+            builder.append("max(0.0, ");
+        } else {
+            builder.append("(");
+        }
+
+        for (int i = 0; i < size; i++) {
+            builder.append("(");
+            builder.append(coefficients[i]);
+
+            if (atomStore == null) {
+                builder.append(" * <index:");
+                builder.append(atomIndexes[i]);
+                builder.append(">)");
+            } else {
+                builder.append(" * ");
+                builder.append(atomStore.getAtomValue(atomIndexes[i]));
+                builder.append(")");
+            }
+
+            if (i != size - 1) {
+                builder.append(" + ");
+            }
+        }
+
+        builder.append(" - ");
+        builder.append(constant);
+
+        builder.append(")");
+
+        if (squared) {
+            builder.append(" ^2");
+        }
+
+        return builder.toString();
+    }
 }