GH-15809: enable loglikelihood and AIC calculation for multinomial fa…

…mily

h2o-core/src/main/java/hex/ModelMetricsRegression.java

@@ -13,21 +13,28 @@
 
 public class ModelMetricsRegression extends ModelMetricsSupervised {
   public final double _mean_residual_deviance;
+  public final double _AIC;
+  public final double _loglikelihood;
   /**
    * @return {@link #mean_residual_deviance()} for all algos except GLM, for which it means "total residual deviance".
    **/
   public double residual_deviance() { return _mean_residual_deviance; }
+  public double loglikelihood() { return _loglikelihood; }
+  public double aic() { return _AIC; }
   @SuppressWarnings("unused")
   public double mean_residual_deviance() { return _mean_residual_deviance; }
   public final double _mean_absolute_error;
   public double mae() { return _mean_absolute_error; }
   public final double _root_mean_squared_log_error;
   public double rmsle() { return _root_mean_squared_log_error; }
-  public ModelMetricsRegression(Model model, Frame frame, long nobs, double mse, double sigma, double mae,double rmsle, double meanResidualDeviance, CustomMetric customMetric) {
+  public ModelMetricsRegression(Model model, Frame frame, long nobs, double mse, double sigma, double mae,double rmsle, 
+                                double meanResidualDeviance, CustomMetric customMetric, double loglikelihood, double aic) {
     super(model, frame, nobs, mse, null, sigma, customMetric);
     _mean_residual_deviance = meanResidualDeviance;
     _mean_absolute_error = mae;
     _root_mean_squared_log_error = rmsle;
+    _loglikelihood = loglikelihood;
+    _AIC = aic;
   }
 
   public static ModelMetricsRegression getFromDKV(Model model, Frame frame) {
@@ -51,6 +58,8 @@ public String toString() {
     }
     sb.append(" mean absolute error: " + (float)_mean_absolute_error + "\n");
     sb.append(" root mean squared log error: " + (float)_root_mean_squared_log_error + "\n");
+    sb.append(" loglikelihood: " + (float)_loglikelihood + "\n");
+    sb.append(" AIC: " + (float)_AIC + "\n");
     return sb.toString();
   }
 
@@ -117,6 +126,7 @@ public static class MetricBuilderRegression<T extends MetricBuilderRegression<T>
     Distribution _dist;
     double _abserror;
     double _rmslerror;
+    protected double _loglikelihood;
     public MetricBuilderRegression() {
       super(1,null); //this will make _work = new float[2];
     }
@@ -147,6 +157,10 @@ public MetricBuilderRegression(Distribution dist) {
           _sumdeviance += _dist.deviance(w, yact[0], ds[0]);
         }
       }
+
+      if(m.getClass().toString().contains("Generic")) {
+        _loglikelihood += m.likelihood(w, yact[0], ds);
+      }
 
       _count++;
       _wcount += w;
@@ -160,6 +174,7 @@ public MetricBuilderRegression(Distribution dist) {
       _sumdeviance += mb._sumdeviance;
       _abserror += mb._abserror;
       _rmslerror += mb._rmslerror;
+      _loglikelihood += mb._loglikelihood;
     }
 
     // Having computed a MetricBuilder, this method fills in a ModelMetrics
@@ -173,6 +188,8 @@ ModelMetricsRegression computeModelMetrics(Model m, Frame f, Frame adaptedFrame,
       double mse = _sumsqe / _wcount;
       double mae = _abserror/_wcount; //Mean Absolute Error
       double rmsle = Math.sqrt(_rmslerror/_wcount); //Root Mean Squared Log Error
+      double loglikelihood = Double.NaN;
+      double aic = Double.NaN;
       if (adaptedFrame ==null) adaptedFrame = f;
       double meanResDeviance = 0;
       if (m != null && m.isDistributionHuber()){
@@ -195,7 +212,12 @@ ModelMetricsRegression computeModelMetrics(Model m, Frame f, Frame adaptedFrame,
       } else {
         meanResDeviance = Double.NaN;
       }
-      ModelMetricsRegression mm = new ModelMetricsRegression(m, f, _count, mse, weightedSigma(), mae, rmsle, meanResDeviance, _customMetric);
+      if(m.getClass().toString().contains("Generic")) {
+        loglikelihood = -1 * _loglikelihood ; // get likelihood from negative loglikelihood
+        aic = m.aic(loglikelihood);
+      }
+      ModelMetricsRegression mm = new ModelMetricsRegression(m, f, _count, mse, weightedSigma(), mae, rmsle, 
+              meanResDeviance, _customMetric, loglikelihood, aic);
       return mm;
     }
   }

h2o-core/src/main/java/hex/ModelMetricsRegressionCoxPH.java

@@ -34,7 +34,7 @@ public class ModelMetricsRegressionCoxPH extends ModelMetricsRegression {
   public ModelMetricsRegressionCoxPH(Model model, Frame frame, long nobs, double mse, double sigma, double mae,
                                      double rmsle, double meanResidualDeviance, CustomMetric customMetric,
                                      double concordance, long concordant, long discordant, long tied_y) {
-    super(model, frame, nobs, mse, sigma, mae, rmsle, meanResidualDeviance, customMetric);
+    super(model, frame, nobs, mse, sigma, mae, rmsle, meanResidualDeviance, customMetric, 0, 0);
 
     this._concordance = concordance;
     this._concordant = concordant;

h2o-core/src/main/java/hex/ModelMetricsRegressionGLM.java

@@ -10,20 +10,17 @@ public class ModelMetricsRegressionGLM extends ModelMetricsRegression implements
   public final long _residualDegressOfFreedom;
   public final double _resDev;
   public final double _nullDev;
-  public final double _AIC;
-  public final double _loglikelihood;
+
 
   public ModelMetricsRegressionGLM(Model model, Frame frame, long nobs, double mse, double sigma,
                                    double mae, double rmsle, double resDev, double meanResDev,
                                    double nullDev, double aic, long nDof, long rDof,
                                    CustomMetric customMetric, double loglikelihood) {
-    super(model, frame, nobs, mse, sigma, mae, rmsle, meanResDev, customMetric);
+    super(model, frame, nobs, mse, sigma, mae, rmsle, meanResDev, customMetric, loglikelihood, aic);
     _resDev = resDev;
     _nullDev = nullDev;
-    _AIC = aic;
     _nullDegressOfFreedom = nDof;
     _residualDegressOfFreedom = rDof;
-    _loglikelihood = loglikelihood;
   }
 
   @Override

h2o-core/src/main/java/hex/ModelMetricsRegressionGeneric.java

@@ -6,7 +6,7 @@ public class ModelMetricsRegressionGeneric extends ModelMetricsRegression {
 
   public ModelMetricsRegressionGeneric(Model model, Frame frame, long nobs, double mse, double sigma, double mae, double rmsle,
                                        double meanResidualDeviance, CustomMetric customMetric, String description) {
-    super(model, frame, nobs, mse, sigma, mae, rmsle, meanResidualDeviance, customMetric);
+    super(model, frame, nobs, mse, sigma, mae, rmsle, meanResidualDeviance, customMetric, 0, 0);
     _description = description;
   }
 }

h2o-core/src/main/java/water/api/schemas3/ModelMetricsRegressionGenericV3.java

@@ -14,12 +14,21 @@ public class ModelMetricsRegressionGenericV3<I extends ModelMetricsRegressionGen
   @API(help="The root mean squared log error for this scoring run.", direction=API.Direction.OUTPUT)
   public double rmsle;
 
+  @API(help="The negative logarithmic likelihood for this scoring run.", direction=API.Direction.OUTPUT)
+  public double loglikelihood;
+
+  @API(help="The AIC for this scoring run.", direction=API.Direction.OUTPUT)
+  public double AIC;
+
+
   @Override
   public S fillFromImpl(I modelMetrics) {
     super.fillFromImpl(modelMetrics);
     mae = modelMetrics._mean_absolute_error;
     rmsle = modelMetrics._root_mean_squared_log_error;
     mean_residual_deviance = modelMetrics._mean_residual_deviance;
+    loglikelihood = modelMetrics.loglikelihood();
+    AIC = modelMetrics.aic();
     return (S) this;
   }
 }

h2o-core/src/main/java/water/api/schemas3/ModelMetricsRegressionV3.java

@@ -17,13 +17,22 @@ public class ModelMetricsRegressionV3<I extends ModelMetricsRegression, S extend
   @API(help="The root mean squared log error for this scoring run.", direction=API.Direction.OUTPUT)
   public double rmsle;
 
+  @API(help="The negative logarithmic likelihood for this scoring run.", direction=API.Direction.OUTPUT)
+  public double loglikelihood;
+
+  @API(help="The AIC for this scoring run.", direction=API.Direction.OUTPUT)
+  public double AIC;
+
+
   @Override
   public S fillFromImpl(I modelMetrics) {
     super.fillFromImpl(modelMetrics);
     r2 = modelMetrics.r2();
     mae = modelMetrics._mean_absolute_error;
     rmsle = modelMetrics._root_mean_squared_log_error;
     mean_residual_deviance = modelMetrics._mean_residual_deviance;
+    loglikelihood = modelMetrics.loglikelihood();
+    AIC = modelMetrics.aic();
     return (S) this;
   }
 }