I'm using h2o package and trying to create a learner using the below given code
install.packages("h2o")
library("h2o")
h2o.learner <- makeLearner("regr.h2o.deeplearning",predict.type = "response")
But I'm getting this error
> h2o.learner <- makeLearner("regr.h2o.deeplearning",predict.type = "response")
Error: could not find function "makeLearner"
Note: Few months back I used this code without any problem.
Any idea what could be possible thing for this error?
The correct code for this is simply
library(mlr)
h2o.learner = makeLearner("regr.h2o.deeplearning")
The makeLearner() is not part of H2O. It appears to be part of the mlr package. It also seems that mlr does have h2o support, so it might be as simple as adding a library(mlr) to the top of your script? (Making sure that the mlr package has been installed, already, of course.)
Related
I am using the MVN package, in R, to find whether each class of the dataset iris is multivariate normal or not.
I used the below code earlier in the day and generated results from it. However, I went to revisit it and now keep getting the following error message:
Error in mvn(data = iris[Species == m[1], 1:4], mvnTest = c("mardia")) :
argument "modelName" is missing, with no default
Can not figure out what this means and how to fix it !
Code:
#Mardia's Test
SM<-mvn(data=iris,subset="Species", mvnTest="mardia")
SM$multivariateNormality
SetosaPlot<-mvn(data=iris, subset="Species", multivariatePlot="qq")
You loaded the mclust package. When you did so you should have seen a warning
The following object is masked from ‘package:MVN’: mvn
So now mvn() is calling mclust::mvn() (i.e. the mvn function in the mclust package) rather than MVN::mvn().
In general you can make sure you get the version from the MVN package by using
MVN::mvn(data=iris, subset="Species", multivariatePlot="qq")
If you want to know where R is finding mvn, try find("mvn")
In general, to resolve these kinds of problems you should start a clean R session, so that you know you're starting with no packages loaded.
(By the way, no real data set is ever truly multivariate normal; you're not testing "whether it is MVN or not", but rather whether it is close enough to MVN that you can't reject the null hypothesis of multivariate normality ...)
I'm using R Studio based on R 3.4.3. However, when I tried to call the forecast.HoltWinters function, R told me that "could not find function "forecast.HoltWinters"". Inspect the installed package (v8.2) told me that it's true, there is no forecast.HoltWinters. But the manual in https://cran.r-project.org/web/packages/forecast/ clearly stated that forecast.HoltWinters is still available.
I have also tried stats::HoldWinters, but it's working wrong. The code run fine on another computer, but it couldn't run at all on mine. Is there any solution?
Here is the code. Book2.csv has enough data to last more than 3 periods.
dltt <- read.csv("book2.csv", header = TRUE)
dltt.ts <- ts(dltt$Total, frequency=12, start=c(2014,4))
dltt.ts.hw <- HoltWinters(dltt.ts)
library(forecast)
dltt.ts.hw.fc <- forecast.HoltWinters(dltt.ts.hw) //Error as soon as I run this line
Fit a HoltWinters model using the HoltWinters function and then use forecast. Its all in the help for HoltWinters and forecast, namely "The function invokes particular _methods_ which depend on the class of the first argument". I'll copy the guts of it here:
m <- HoltWinters(co2)
forecast(m)
Note this will call the non-exported forecast.HoltWinters function, which you should never call directly using triple-colon notation as some may suggest.
I am running into problems when applying recursive feature selection to nnet models with caret::rfe; I get the following error message:
Error in { : task 1 failed - "undefined columns selected"
The actual task is more complex than the following example, but I am confident that this is a similar problem:
library(caret)
rfe(x = iris[,1:3],
y = iris[,4]/max(iris[,4]),
sizes = c(2),
method="nnet",
rfeControl = rfeControl(functions = caretFuncs)
)
I know this error can occur when trying to select more features than there are available in x (e.g. see https://stats.stackexchange.com/questions/18362/odd-error-with-caret-function-rfe), but this does not seem to be the problem here. I also ran very similar calls in earlier versions of caret, without this problem occurring.
I use R 3.3.1 and caret 6.0.71.
Thank you very much for your help.
EDIT: I went through the archived versions of caret and found that the example code is working in caret versions <= 6.0.62.
I went through the archived versions of caret and found that the example code is working in caret versions <= 6.0.62. This also solves the problems my original code had. I reported this issue on the caret github.
EDIT: The problem is now fixed : https://github.com/topepo/caret/issues/485
I'm currently working on preditive models with the 'randomForest' package.
Fitting my model as follow
rf <- foreach(ntree=rep(10, 3), .combine= combine, .packages='randomForest') %dopar% {
randomForest(bou~.,data=train, trees=50, importance=TRUE)}
When using 'confusionMatrix' from the 'caret' package, I've got the results below :
I'd like to know if it's possible to set the positive class to 1 in the model. I searched in the package description but couldn't find anything about it.
Thank you very much.
Edit : I've found it. It's an option in the 'confusionMatrix' command from the 'caret' package. I was lokking at the wrong place. Here an example if needed.
confusionMatrix(predicted,true_values,positive='1')
Should I leave my post or delete it ?
I've found it. It's an option in the confusionMatrix command from the caret package. I was looking at the wrong place. Here an example if needed:
confusionMatrix(predicted,true_values,positive='1')
I use R version 2.15.1 (2012-06-22) and mgcv version 1.7-22
I load the following set of packages in R:
library(sqldf)
library(timeDate)
library(forecast)
library(xts)
library(tseries)
library(MASS)
library(mgcv)
It happens that I can not run a simple model (I omit the code). Even the sample code taken from the help pages:
dat = gamSim(1,n=400,dist="normal",scale=2)
b = gam(y~s(x0)+s(x1)+s(x2)+s(x3),data=dat)
gives an error:
Error in qr.qty(qrc, sm$S[[l]]) :
NA/NaN/Inf in foreign function call (arg 5)
In addition: Warning message:
In smoothCon(split$smooth.spec[[i]], data, knots, absorb.cons, scale.penalty = scale.penalty, :
number of items to replace is not a multiple of replacement length
Note that everything works fine, if I just load the package mgcv and then use the sample code right away. It also works if I just load all the packages and run the sample code. It just does not work if I
load all packages
do some file reading, sqldf statements, ts operations and some models from package forecast.
if I then apply GAM, it does not work anymore.
Apparently the variable definitions in the general environment mess up the functioning of the package.
Are there any known issues? Are there general rules that I have to obey if I load various packages? Can I write code that "disturbed" the package mgcv?
# Richard there are 2 GAM related packages: gam and mgcv. Loading both libraries at the same time usually causes a conflict.
Loading mgcv as the first package solved my problem ... strange but true.