diff --git a/R/ensemble/README.md b/R/ensemble/README.md
index 49fd51cb7c..b0046542dc 100644
--- a/R/ensemble/README.md
+++ b/R/ensemble/README.md
@@ -28,6 +28,7 @@ R CMD INSTALL h2oEnsemble-package
 
 
 ## Known Bugs
+- This package is incompatible with R 3.0.0-3.1.0 due to a [parser bug](https://bugs.r-project.org/bugzilla3/show_bug.cgi?id=15753) in R.  Upgrade to R 3.1.1 or greater to resolve the issue.  It may work on earlier versions of R but has not been tested.
 - Sometimes while executing `h2o.ensemble`, the code hangs due to a communication issue with H2O.  You may see something like this.  To fix, restart R.
 ```
 GET /Cloud.json HTTP/1.1
diff --git a/R/ensemble/h2oEnsemble-package/R/ensemble.R b/R/ensemble/h2oEnsemble-package/R/ensemble.R
index ef63ae207a..2a87e46c1d 100644
--- a/R/ensemble/h2oEnsemble-package/R/ensemble.R
+++ b/R/ensemble/h2oEnsemble-package/R/ensemble.R
@@ -61,7 +61,7 @@ function(x, y, data, family = "binomial",
   if (grepl("^SL.", metalearner)) {
     # this is very hacky and should be used only for testing until we get the h2o metalearner functions sorted out...
     familyFun <- get(family, mode = "function", envir = parent.frame())
-    Ztmp <- subset(Z, select=-c(fold_id, Class))
+    Ztmp <- Z[, -which(names(Z) %in% c("fold_id", y))]
     runtime$metalearning <- system.time(metafit <- match.fun(metalearner)(Y=as.data.frame(data[,c(y)])[,1], X=Ztmp, newX=Ztmp, 
                                                                           family=familyFun, id=seq(N), obsWeights=rep(1,N)), gcFirst=FALSE)
   } else {
@@ -119,9 +119,11 @@ function(x, y, data, family = "binomial",
   print(sprintf("Cross-validating learner %s: fold %s", idxs$l[i], idxs$v[i]))
   if (is.numeric(seed)) set.seed(seed)  #If seed is specified, set seed prior to next step
   fit <- match.fun(learner[idxs$l[i]])(y=y, x=xcols, data=data[data$fold_id!=idxs$v[i]], family=family)
-  # Regarding preds assignment below: This is hardcoded for binary outcome (ie. we are grabbing the X1 column)
-  # Probably need to modify this line so that it also works for regression
-  preds <- as.data.frame(h2o.predict(fit, data[data$fold_id==idxs$v[i]]))$X1
+  if (family == "binomial") {
+    preds <- as.data.frame(h2o.predict(fit, data[data$fold_id==idxs$v[i]]))$X1
+  } else {
+    preds <- as.data.frame(h2o.predict(fit, data[data$fold_id==idxs$v[i]]))$predict
+  }
   # Note: column subsetting not supported yet in H2OParsedData object however, 
   # if we can enable that, then it is probably better to insert the preds into 
   # a H2OParsedData object instead of returning 'preds' and bringing into R memory.
@@ -212,9 +214,13 @@ predict.h2o.ensemble <-
     L <- length(object$basefits)
     basepreddf <- as.data.frame(matrix(NA, nrow = nrow(newdata), ncol = L))
     for (l in seq(L)) {
-      # This is hardcoded ($X1) for binary classification, should change this
-      basepreddf[, l] <- as.data.frame(do.call('h2o.predict', list(object = object$basefits[[l]],
-                                                                   newdata = newdata)))$X1
+      if (object$family == "binomial") {
+        basepreddf[, l] <- as.data.frame(do.call('h2o.predict', list(object = object$basefits[[l]],
+                                                                     newdata = newdata)))$X1 
+      } else {
+        basepreddf[, l] <- as.data.frame(do.call('h2o.predict', list(object = object$basefits[[l]],
+                                                                     newdata = newdata)))$predict
+      }
     }
     names(basepreddf) <- names(object$basefits)
     basepreddf[basepreddf < object$ylim[1]] <- object$ylim[1]  #Enforce bounds
diff --git a/R/ensemble/h2oEnsemble-package/R/wrappers.R b/R/ensemble/h2oEnsemble-package/R/wrappers.R
index 4dc5f2fe18..112a4ebd8c 100644
--- a/R/ensemble/h2oEnsemble-package/R/wrappers.R
+++ b/R/ensemble/h2oEnsemble-package/R/wrappers.R
@@ -5,7 +5,7 @@
 # like "x, y, data, family", so that the ensemble code can be written more cleanly.
 
 
-h2o.glm.wrapper <- function(x, y, data, key = "", family = "binomial", link = "logit", nfolds = 0, alpha = 0.5, nlambda = -1, 
+h2o.glm.wrapper <- function(x, y, data, key = "", family = "binomial", link, nfolds = 0, alpha = 0.5, nlambda = -1, 
                                     lambda.min.ratio = -1, lambda = 1e-5, epsilon = 1e-4, standardize = TRUE, 
                                     prior, variable_importances = FALSE, use_all_factor_levels = FALSE, 
                                     tweedie.p = ifelse(family == 'tweedie', 1.5, as.numeric(NA)), iter.max = 100, 
@@ -44,7 +44,8 @@ h2o.randomForest.wrapper <- function(x, y, data, key = "", family = "binomial",
                    ntree = ntree, depth = depth, sample.rate = sample.rate, nbins = nbins, seed = seed, 
                    importance = importance, nfolds = nfolds, validation = validation, nodesize = nodesize, 
                    balance.classes = balance.classes, max.after.balance.size = max.after.balance.size,
-                   doGrpSplit = doGrpSplit, verbose = verbose, oobee = oobee, stat.type = stat.type, type = type)
+                   doGrpSplit = doGrpSplit, verbose = verbose, oobee = oobee, stat.type = stat.type, 
+                   type = ifelse(family=="binomial", type, "BigData"))
 }