weighted quantile runit nopass

h2o-r/tests/testdir_misc/runit_NOPASS_quantile.R

@@ -0,0 +1,37 @@
+#This tests quantile and weighted quantile on synthetic data by comparing with R
+
+test.quantile <- function(conn){
+N = 1000
+
+x = rgamma(N, shape=0.067,  scale = 0.008) 
+aa = as.h2o(x)
+r_q = quantile(x,  probs = c(0.1, 0.5, 1, 2, 5, 10, 50,88.83,99,90)/100,na.rm=T)
+h_q = h2o.quantile(aa,probs = c(0.1, 0.5, 1, 2, 5, 10, 50,88.83,99,90 )/100,na.rm=T)
+expect_equal(r_q,h_q )
+
+x = rlnorm(N,meanlog = 12,sdlog = 132)
+aa = as.h2o(x)
+r_q = quantile(x,  probs = seq(0,1,.05),na.rm=T)
+h_q = h2o.quantile(aa,probs = seq(0,1,.05),na.rm=T)
+expect_equal(r_q,h_q )
+
+x = rexp(N, rate = 12.3) 
+ss = sample(1:N,size = N/10,replace = F)
+ x[ss]=NA
+aa = as.h2o(x)
+r_q = quantile(x,  probs = seq(0,1,.05),na.rm=T)
+h_q = h2o.quantile(aa,probs = seq(0,1,.05),na.rm=T)
+expect_equal(r_q,h_q )
+
+#weighted quantiles
+#library(Hmisc)
+x = runif(N)
+aa = as.h2o(x)
+wts = sample(1:6, N, TRUE)
+h_wts = as.h2o(wts)
+#r_q = wtd.quantile(x, wts, probs = seq(0,1,.05))
+#h_q = h2o.quantile(aa,probs = seq(0,1,.05),weight_column = h_wts)
+#expect_equal(r_q,h_q )
+
+}
+doTest("Test quantile",test.quantile )

h2o-r/tests/testdir_misc/runit_NOPASS_weighted_quantile.R

@@ -0,0 +1,54 @@
+# This tests weighted quantile
+# by comparing results with R's wtd.quntile function and sanity checking by ignoring rows with zero weight
+# dataset - http://mlr.cs.umass.edu/ml/datasets/Bank+Marketing
+
+test.wtd.quantile <- function(conn){
+
+	a= h2o.importFile(locate("smalldata/gbm_test/bank-full.csv.zip"),destination_frame = "bank_UCI")
+	dim(a)
+	myX = 1:16
+	myY = 17
+
+	rowss =45211
+	#Sample rows for 2-fold xval
+	ss = sample(1:rowss,size = 22000)
+	ww = rep(1,rowss)
+	ww[ss]=2
+
+	#Parse fold column to h2O
+	wei = as.h2o(ww,destination_frame = "weight")
+	colnames(wei)
+
+	#Cbind fold column to the original dataset
+	a = h2o.assign(h2o.cbind(a,wei),key = "bank")
+	dim(a)
+
+	#Build gbm by specifying the fold column
+	gg = h2o.gbm(x = myX,y = myY,training_frame = a,ntrees = 5,fold_column = "x",keep_cross_validation_predictions = T,model_id = "cv_gbm")
+
+	#Define and use weights column 
+	ww[ss]=0
+	wi = as.h2o(ww,destination_frame = "weight_col")
+
+	#Predict
+	pr = h2o.predict(gg,a)
+	pred = as.data.frame(pr[,3])
+
+	# weighted h2o quantile
+	#hq = as.numeric(h2o.quantile(pr[,3],probs = seq(0,.95,.05),weight_column = wi))
+
+	# weighted R quantile
+	#library(Hmisc)
+	#wq = as.numeric(wtd.quantile(pred[,1],ww,probs = seq(0,.95,.05)))
+
+	#expect_equal(wq,hq,tolerance = 1e-5)
+
+	#Sanity check with just nonzero weighted rows
+	#pp=pred[which(ww==1),]
+	#qq = as.numeric(quantile(pp,probs = seq(0,.95,.05)))
+
+	#expect_equal(wq,qq,tolerance = 3e-4)
+	#expect_equal(hq,qq,tolerance = 3e-4)
+}
+doTest("Test weighted quantile",test.wtd.quntile )
+