R help - Oct 2008 - How to save/load RWeka models into/from a file?

Hi,

I want to save a RWeka model into a file, in order to retrive it latter 
with a load function.

See this example:

library(RWeka)

NB <- make_Weka_classifier("weka/classifiers/bayes/NaiveBayes")
model<-NB(formula,data=data,...) # does not run but you get the idea

save(model,file="model.dat") # simple save R command
# ...
load("model.dat") # load the model from the previously saved file...
model # should work but I get instead this error:
Error in .jcall(x$classifier, "S", "toString") :
   RcallMethod: attempt to call a method of a NULL object.

What is wrong and how can I solve this problem?

Regards,
-- 
Paulo Alexandre Ribeiro Cortez  (PhD, MSc)
Lecturer (Prof. Auxiliar) at the Department of Information Systems (DSI)
University of Minho, Campus de Azur?m, 4800-058 Guimaraes, Portugal
http://www.dsi.uminho.pt/~pcortez +351253510313 Fax:+351253510300

On Thu, 16 Oct 2008, Paulo Cortez wrote:
> Hi,
>
> I want to save a RWeka model into a file, in order to retrive it latter
with
> a load function.
>
> See this example:
>
> library(RWeka)
>
> NB <-
make_Weka_classifier("weka/classifiers/bayes/NaiveBayes")
> model<-NB(formula,data=data,...) # does not run but you get the idea
>
> save(model,file="model.dat") # simple save R command
> # ...
> load("model.dat") # load the model from the previously saved
file...
> model # should work but I get instead this error:
> Error in .jcall(x$classifier, "S", "toString") :
>  RcallMethod: attempt to call a method of a NULL object.
>
> What is wrong and how can I solve this problem?
The R object is just a reference to the corresponding object on the Java 
side (in Weka). When you close R, the Java/Weka session is also closed and 
the model is gone. Thus, when you load the object again in a new session, 
you only have a reference to a Java object that does not live anymore.
Z
> Regards,
> -- 
> Paulo Alexandre Ribeiro Cortez  (PhD, MSc)
> Lecturer (Prof. Auxiliar) at the Department of Information Systems (DSI)
> University of Minho, Campus de Azur?m, 4800-058 Guimaraes, Portugal
> http://www.dsi.uminho.pt/~pcortez +351253510313 Fax:+351253510300
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
>

Achim Zeileis wrote:
> On Thu, 16 Oct 2008, Paulo Cortez wrote:
> 
>> Hi,
>>
>> I want to save a RWeka model into a file, in order to retrive it 
>> latter with a load function.
>>
>> See this example:
>>
>> library(RWeka)
>>
>> NB <-
make_Weka_classifier("weka/classifiers/bayes/NaiveBayes")
>> model<-NB(formula,data=data,...) # does not run but you get the idea
>>
>> save(model,file="model.dat") # simple save R command
>> # ...
>> load("model.dat") # load the model from the previously saved
file...
>> model # should work but I get instead this error:
>> Error in .jcall(x$classifier, "S", "toString") :
>>  RcallMethod: attempt to call a method of a NULL object.
>>
>> What is wrong and how can I solve this problem?
> 
> The R object is just a reference to the corresponding object on the Java 
> side (in Weka). When you close R, the Java/Weka session is also closed 
> and the model is gone. Thus, when you load the object again in a new 
> session, you only have a reference to a Java object that does not live 
> anymore.
> Z
Thanks, I understand now what is happening.

Yet, I still need to save/load a RWeka model to/from a file.
Thus, how can I do this? Is it impossible in R? Any help?

Regards,
-- 
Paulo Alexandre Ribeiro Cortez  (PhD, MSc)
Lecturer (Prof. Auxiliar) at the Department of Information Systems (DSI)
University of Minho, Campus de Azur?m, 4800-058 Guimaraes, Portugal
http://www.dsi.uminho.pt/~pcortez +351253510313 Fax:+351253510300

>>>>> Paulo Cortez writes:
> Achim Zeileis wrote:
>> On Thu, 16 Oct 2008, Paulo Cortez wrote:
>> 
>>> Hi,
>>> 
>>> I want to save a RWeka model into a file, in order to retrive it 
>>> latter with a load function.
>>> 
>>> See this example:
>>> 
>>> library(RWeka)
>>> 
>>> NB <-
make_Weka_classifier("weka/classifiers/bayes/NaiveBayes")
>>> model<-NB(formula,data=data,...) # does not run but you get the
idea
>>> 
>>> save(model,file="model.dat") # simple save R command
>>> # ...
>>> load("model.dat") # load the model from the previously
saved file...
>>> model # should work but I get instead this error:
>>> Error in .jcall(x$classifier, "S", "toString")
:
>>> RcallMethod: attempt to call a method of a NULL object.
>>> 
>>> What is wrong and how can I solve this problem?
>> 
>> The R object is just a reference to the corresponding object on the
Java
>> side (in Weka). When you close R, the Java/Weka session is also closed 
>> and the model is gone. Thus, when you load the object again in a new 
>> session, you only have a reference to a Java object that does not live 
>> anymore.
>> Z
> Thanks, I understand now what is happening.
> Yet, I still need to save/load a RWeka model to/from a file.
> Thus, how can I do this? Is it impossible in R? Any help?
Using a current version of rJava, you can register the Java side objects
for serialization using .jcache().

E.g.,

  m1 <- J48(Species ~ ., data = iris)
  .jcache(m1$classifier)

then save/load will work as expected.

Eventually, the above may happen automagically.

-k

Kurt Hornik wrote:
>>>>>> Paulo Cortez writes:
> Using a current version of rJava, you can register the Java side objects
> for serialization using .jcache().
> 
> E.g.,
> 
>   m1 <- J48(Species ~ ., data = iris)
>   .jcache(m1$classifier)
> 
> then save/load will work as expected.
> 
> Eventually, the above may happen automagically.
> 
> -k
> 
Thanks, it works quite well in my MacOS.

Yet, in a linux machine (Fedora 9, R version 2.7.2, latest rJava and 
RWeka versions has found in the berkeley server), sometimes the save 
function produces a Segmentation fault. Since I am working with large 
datasets (email filtering), I wonder if there is any 
bug/limitation/problem of rJava/RWeka for large models (e.g. NaiveBayes 
with more than 1000 inputs).

Regards,
-- 
Paulo Alexandre Ribeiro Cortez  (PhD, MSc)
Lecturer (Prof. Auxiliar) at the Department of Information Systems (DSI)
University of Minho, Campus de Azur?m, 4800-058 Guimaraes, Portugal
http://www.dsi.uminho.pt/~pcortez +351253510313 Fax:+351253510300