Merge branch 'reducedmemory'

vertica · Oct 14, 2015 · b4d5c33 · b4d5c33
2 parents e0120d5 + f1cae11
commit b4d5c33
Show file tree

Hide file tree

Showing 3 changed files with 60 additions and 52 deletions.
diff --git a/algorithms/HPdclassifier/R/hpdRF_parallelTree.R b/algorithms/HPdclassifier/R/hpdRF_parallelTree.R
@@ -418,8 +418,6 @@ hpdrandomForest <- hpdRF_parallelTree <- function(formula, data,
 		model$type = "regression"
 		if(completeModel)
 		{
-			tryCatch(
-			{
 			model$mse = oob_predictions$mse
 			model$rsq = oob_predictions$rsq
 
@@ -439,10 +437,6 @@ hpdrandomForest <- hpdRF_parallelTree <- function(formula, data,
 					       model$test$predicted,
 					       na.rm = TRUE)
 			}
-			},
-			error = function(e){
-			      stop(paste("could not compute additional statistics due to error:", e))
-			})			
 		}
 	}
 	else
@@ -451,9 +445,7 @@ hpdrandomForest <- hpdRF_parallelTree <- function(formula, data,
 
 		if(completeModel)
 		{
-			tryCatch(
-			{
-			confusion = confusionMatrix(true_responses, 
+			confusion = HPdutility::confusionMatrix(true_responses, 
 				    	model$predicted)
 			model$err.rate = oob_predictions$err.rate
 			classErr <- model$err.rate[nrow(model$err.rate),-1]
@@ -486,7 +478,7 @@ hpdrandomForest <- hpdRF_parallelTree <- function(formula, data,
 				{
 					stop("Categories of 'ytest' are not the same as categories of response variable trained in model")
 				}
-				confusionTest = confusionMatrix(ytest,
+				confusionTest = HPdutility::confusionMatrix(ytest,
 			       		model$test$predicted)
 				model$test$err.rate = errorRate(ytest,
 					model$test$predicted)
@@ -497,11 +489,6 @@ hpdrandomForest <- hpdRF_parallelTree <- function(formula, data,
             			model$test$confusion  <-cbind(confusionTest, 
 						      classErrTest)
 			}
-			},
-			error = function(e){
-			      stop(paste("could not compute additional statistics due to error:", e))
-			})			
-
 		}
 	}
 

diff --git a/algorithms/HPdclassifier/R/hpdRFdistributed.R b/algorithms/HPdclassifier/R/hpdRFdistributed.R
@@ -587,7 +587,6 @@
 		tree_ids = tree_ids[-1]
 		forest = .Call("unserializeForest",forest, 
 		       	    PACKAGE = "HPdclassifier")  
-
 		forestparam = .Call("getForestParameters", forest,  
 			    PACKAGE = "HPdclassifier")
 
@@ -609,35 +608,16 @@
 				lapply(categorical_variables,
 				function(var)
 				observations[,var]+1)
-			
-		predictions = matrix(as.numeric(NA), 
+
+		predictions = matrix(as.double(NA), 
 			      	ncol = nrow(observations),
 				length(tree_ids))
+
 		tree_ids = tree_ids - 1
-		temp_predictions = lapply(1:length(tree_ids),
-		function(tree_id_index)
-		{
-			tree_id = tree_ids[tree_id_index]
-			tree_oob_indices = oob_indices[[tree_id]]
-			tree_predictions = sapply(tree_oob_indices,
-				function(obs)
-				.Call("specificTreePredictObservation", 
-				forest, as.integer(tree_id),
-				observations, 
-				as.integer(obs),
-				PACKAGE = "HPdclassifier"))
-			return(cbind(rep(tree_id_index,
-				length(tree_oob_indices)),
-				tree_oob_indices,as.numeric(tree_predictions)))
-		})
-		if(length(temp_predictions)>0)
-		{
-		for(i in 1:length(temp_predictions))
-		{
-		      predictions[temp_predictions[[i]][,c(1,2)]] = 
-		      		temp_predictions[[i]][,3]
-		}
-		}
+		oob_indices = lapply(oob_indices, as.integer)
+		.Call("forestPredictOOB", forest, predictions, observations,
+					  oob_indices, as.integer(tree_ids))
+
 		.Call("garbageCollectForest",forest)
 		update(predictions)
 	},progress = FALSE)
@@ -1097,16 +1077,12 @@
 		       	    PACKAGE = "HPdclassifier") 
 
 			forestparam=.Call("getForestParameters", forest)
-			response_cardinality = forestparam[[2]]
-			ntree = forestparam[[7]]
-			predictions = sapply(1:nrow(new_observations), function(index)
-			       sapply(tree_ids, function(tree_id)
-			            as.numeric(.Call("specificTreePredictObservation",
-					forest, as.integer(tree_id), 
-			      	      	new_observations, 
-			      		as.integer(index),
-		     			PACKAGE = "HPdclassifier"))))
-			predictions = matrix(predictions,ncol=nrow(new_observations))
+			predictions = matrix(as.double(NA), 
+			      	ncol = nrow(new_observations),
+				length(tree_ids))
+			.Call("forestPredictObservations", forest, predictions, 
+					new_observations, as.integer(tree_ids))
+
 			update(predictions)
 			.Call("garbageCollectForest",forest)
 			rm(list = ls())

diff --git a/algorithms/HPdclassifier/src/hpdRFpredict.cpp b/algorithms/HPdclassifier/src/hpdRFpredict.cpp
@@ -490,4 +490,49 @@ extern "C"
     return R_new_errors;
   }
 
+
+  void forestPredictOOB(SEXP R_forest, SEXP R_predictions, SEXP R_observations,
+	       SEXP R_indices, SEXP R_tree_ids)
+  {
+    hpdRFforest *forest = (hpdRFforest *) R_ExternalPtrAddr(R_forest);
+    double *predictions = REAL(R_predictions);
+    for(int tree_index = 0; tree_index < length(R_tree_ids); tree_index++)
+      {
+	int tree_id = INTEGER(R_tree_ids)[tree_index]-1;
+	SEXP indices;
+	if(length(R_indices) > 1)
+	  indices = VECTOR_ELT(R_indices,tree_id);
+	else
+	  indices = R_indices;
+	for(int i = 0; i < length(indices); i++)
+	  {
+	    int obs_index = INTEGER(indices)[i]-1;
+	    predictions[obs_index*length(R_tree_ids)+tree_index] =
+	      treePredictObservation(forest->trees[tree_id], 
+				     R_observations, 
+				     forest->features_cardinality, 
+				     obs_index);
+	  }
+      }
+  }
+  void forestPredictObservations(SEXP R_forest, SEXP R_predictions, 
+				 SEXP R_observations, SEXP R_tree_ids)
+  {
+    hpdRFforest *forest = (hpdRFforest *) R_ExternalPtrAddr(R_forest);
+    double *predictions = REAL(R_predictions);
+    int num_predictions = length(VECTOR_ELT(R_observations,0));
+    for(int tree_index = 0; tree_index < length(R_tree_ids); tree_index++)
+      {
+	int tree_id = INTEGER(R_tree_ids)[tree_index]-1;
+	for(int obs_index = 0; obs_index < num_predictions; obs_index++)
+	  {
+	    predictions[obs_index*length(R_tree_ids)+tree_index] =
+	      treePredictObservation(forest->trees[tree_id], 
+				     R_observations, 
+				     forest->features_cardinality, 
+				     obs_index);
+	  }
+      }
+  }
+
 }