Re: XGBoost Installation

Thanks,
I did so however in my problem the cost of a False Positive is 5 times more
than a False Negative thus the confusion matrix used is [0.0 1.0; 5.0 0.0].
When implementing the setting you suggested I get an accuracy of 57% still
relatively low.
The highest I was able to get was 67% when using nrounds=1000.
Could it just be that XGBoost does not work as well as knn and rf for this
case or am I missing something?

Kind regards,
Prakash

On Sun, Jun 6, 2021 at 2:13 AM Eibe Frank <[email protected]> wrote:

> Looking at the configuration string I posted, you will see
>
> -learner classif.xgboost -params
> nrounds=100,max_depth=3,subsample=0.5,colsample_bynode=0.5
>
> The
>
> -params
>
> flag introduces the parameters that will be used for the MLR learner that
> is specified. You need to paste those into the corresponding text field in
> the GenericObjectEditor when you configure MLRClassifier.
>
> Yes, you should use the same cost matrix.
>
> Cheers,
> Eibe
>
> On Sun, Jun 6, 2021 at 1:09 PM Prakash Bhagat <[email protected]>
> wrote:
>
>> Thanks so much for this!
>> For your first point, how do I set a minimum number of boosting
>> iterations?
>> I have looked within the MLR classifier setting however see no option for
>> min iterations.
>> If I use a cost matrix in the classifier then under more options for test
>> I should create the same confusion matrix, correct?
>>
>> Kind regards,
>> Prakash
>>
>> On Sat, Jun 5, 2021 at 12:56 PM Eibe Frank <[email protected]> wrote:
>>
>>> The default number of iterations used by XGBoost when applied through
>>> the RPlugin is 1. You definitely need to increase the number of boosting
>>> iterations. I would also switch the CostSensitiveClassifier to use the
>>> minimum-expected cost approach, not the default approach based on
>>> reweighting the training data.
>>>
>>> Of course, you should also specify the same cost matrix for evaluation
>>> under "More options..." in the Classify tab.
>>>
>>> The following configuration worked well in a quick try on the sonar data
>>> from the UCI repository:
>>>
>>> weka.classifiers.meta.CostSensitiveClassifier -cost-matrix "[0.0 5.0;
>>> 1.0 0.0]" -M  -W weka.classifiers.mlr.MLRClassifier -- -learner
>>> classif.xgboost -params
>>> nrounds=100,max_depth=3,subsample=0.5,colsample_bynode=0.5
>>>
>>> You can right-click on the text field containing the classifier
>>> configuration in the Classify panel and paste the above configuration into
>>> the Explorer (assuming you have a dataset with only two classes, it should
>>> work).
>>>
>>> The reason for using the minimum-expected cost mode of the
>>> CostSensitiveClassifier is that it does not change the training data. The
>>> default mode of the CostSensitiveClassifier uses instance weights in the
>>> training data to reflect costs. I don't know how well that works in
>>> conjunction with MLRClassifier.
>>>
>>> Cheers,
>>> Eibe
>>>
>>> On Sat, Jun 5, 2021 at 11:33 PM Prakash Bhagat <
>>> [email protected]> wrote:
>>>
>>>> Thanks very much,
>>>>
>>>> I have managed to get windows access and followed the tutorial on
>>>> youtube for downloading the rplugin. I am using xgboost as a
>>>> classification method and comparing it to random forests. Random forest
>>>> gives an accuracy of 75% whereas XGBoost gives an accuracy of only 52%. I
>>>> was wondering if you could help me as i may have made some errors when
>>>> implementing xgboost.
>>>> I have attached the data used. I filled in missing values for
>>>> revol_util and converted nominal features to binary when using xgboost.
>>>> I then used the costsensitiveclassifier with a 5 to 1 cost for false
>>>> positives and used 5 fold cross-validation. Following this, I used
>>>> mlrclassifier and used classif.xgboost under Rlearner.
>>>> I will do hyper parameter tuning after but was wandering if i had made
>>>> a mistake with xgboost as the accuracy is so low.
>>>> Any help would be much appreciated!
>>>>
>>>> Kind regards,
>>>> Prakash
>>>>
>>>> On Sat, Jun 5, 2021 at 4:16 AM Eibe Frank <[email protected]> wrote:
>>>>
>>>>> You cannot set environment variables, etc., in WEKA's built-in CLI,
>>>>> which is really quite primitive and only pretends to be a proper OS
>>>>> terminal! You need to run the command in the macOS terminal. However, as I
>>>>> said, you should not set R_HOME explicitly unless you have installed R in a
>>>>> non-standard location. WEKA's RPlugin will find R on the Mac if it is
>>>>> installed in the standard place.
>>>>>
>>>>> Cheers,
>>>>> Eibe
>>>>>
>>>>> On Sat, Jun 5, 2021 at 1:43 PM Prakash Bhagat <
>>>>> [email protected]> wrote:
>>>>>
>>>>>> Yes, I have followed this guide:
>>>>>> https://riptutorial.com/weka/topic/7916/how-to-use-r-in-weka
>>>>>> I pasted this in the command line in WEKA: export
>>>>>> R_HOME=/Library/Frameworks/R.framework/Resources
>>>>>> java -Xss10M -Xmx4096M -cp .:weka.jar weka.gui.GUIChooser
>>>>>>
>>>>>> And this is the error from weka:
>>>>>>
>>>>>> Kind regards,
>>>>>> Prakash
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 04 Jun,2021, at 13:52, Eibe Frank <[email protected]>
>>>>>> wrote:
>>>>>>
>>>>>> Are you still running WEKA with R_HOME set to some value? It is
>>>>>> normally best to let WEKA set the value of R_HOME itself.
>>>>>>
>>>>>> If there are any error messages shown in the terminal from which you
>>>>>> run WEKA, it would be useful to share those.
>>>>>>
>>>>>> Cheers,
>>>>>> Eibe
>>>>>>
>>>>>> On Fri, Jun 4, 2021 at 11:43 PM Prakash Bhagat <
>>>>>> [email protected]> wrote:
>>>>>>
>>>>>>> https://github.com/SigDelta/weka-xgboost/releases - This is the
>>>>>>> link I used to download xgboost.
>>>>>>> Using this downloaded I installed xgboost as an unofficial package
>>>>>>> in tools. It then comes under trees however does not run.
>>>>>>>
>>>>>>> With the RPlugin the MLR classifier is unavailable to use. I have
>>>>>>> downloaded a new version of R and within R I also downloaded the MLR
>>>>>>> package. This still however does not run.
>>>>>>> I have installed rJava within R as well.
>>>>>>> <Screenshot 2021-06-04 at 11.51.31.png>
>>>>>>>
>>>>>>> On 04 Jun,2021, at 11:07, Eibe Frank <[email protected]> wrote:
>>>>>>>
>>>>>>> Which unofficial version do you mean? Both, RPlugin and wekaPython,
>>>>>>> are official packages, and both enable you to use XGBoost. With the
>>>>>>> RPlugin, you need to use XGBoost through the MLRClassifier and select
>>>>>>> XGBoost as the base learner (the first time you select this base learner,
>>>>>>> it will try to install it in R, which may take a while). Similarly, with
>>>>>>> wekaPython, you also need to use the corresponding ScikitLearnClassifier in
>>>>>>> WEKA with XGBoost as the base learner (and you need to have XGBoost
>>>>>>> installed in your relevant Python environment, which you may need to do
>>>>>>> manually).
>>>>>>>
>>>>>>> The easiest way to use XGBoost in WEKA is probably through the
>>>>>>> RPlugin because the only manual work you (normally) need to do is to
>>>>>>> install R.
>>>>>>>
>>>>>>> There might be an unofficial WEKA package for XGBoost out there, but
>>>>>>> it may not be maintained.
>>>>>>>
>>>>>>> Cheers,
>>>>>>> Eibe
>>>>>>>
>>>>>>>
>>>>>>> On Fri, Jun 4, 2021 at 4:49 PM <[email protected]> wrote:
>>>>>>>
>>>>>>>> Dear All,
>>>>>>>>
>>>>>>>> I have a mac with the latest version of R and python installed. I
>>>>>>>> am using WEKA for an assignment and need to use XGBoost.
>>>>>>>> I have tried downloading the unofficial version as a package
>>>>>>>> however that does not work.
>>>>>>>> I then tried the R Plugin version downloading RPlugin and pasting
>>>>>>>> this in the command line in WEKA:
>>>>>>>> export R_HOME=/Library/Frameworks/R.framework/Resources
>>>>>>>> java -Xss10M -Xmx4096M -cp .:weka.jar weka.gui.GUIChoose
>>>>>>>>
>>>>>>>> This however did not work either and R does not run.
>>>>>>>>
>>>>>>>> If anyone has a very specific step-by-step guide to getting XGBoost
>>>>>>>> on macOS please let me know!
>>>>>>>>
>>>>>>>> Kind regards,
>>>>>>>> PYB
>>>>>>>> _______________________________________________
>>>>>>>> Wekalist mailing list -- [email protected]
>>>>>>>> Send posts to [email protected]
>>>>>>>> To unsubscribe send an email to [email protected]
>>>>>>>> To subscribe, unsubscribe, etc., visit
>>>>>>>> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
>>>>>>>> List etiquette:
>>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Wekalist mailing list -- [email protected]
>>>>>>> Send posts to [email protected]
>>>>>>> To unsubscribe send an email to [email protected]
>>>>>>> To subscribe, unsubscribe, etc., visit
>>>>>>> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
>>>>>>> List etiquette:
>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Wekalist mailing list -- [email protected]
>>>>>>> Send posts to [email protected]
>>>>>>> To unsubscribe send an email to [email protected]
>>>>>>> To subscribe, unsubscribe, etc., visit
>>>>>>> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
>>>>>>> List etiquette:
>>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>>
>>>>>> _______________________________________________
>>>>>> Wekalist mailing list -- [email protected]
>>>>>> Send posts to [email protected]
>>>>>> To unsubscribe send an email to [email protected]
>>>>>> To subscribe, unsubscribe, etc., visit
>>>>>> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
>>>>>> List etiquette:
>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> Wekalist mailing list -- [email protected]
>>>>>> Send posts to [email protected]
>>>>>> To unsubscribe send an email to [email protected]
>>>>>> To subscribe, unsubscribe, etc., visit
>>>>>> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
>>>>>> List etiquette:
>>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>>
>>>>> _______________________________________________
>>>>> Wekalist mailing list -- [email protected]
>>>>> Send posts to [email protected]
>>>>> To unsubscribe send an email to [email protected]
>>>>> To subscribe, unsubscribe, etc., visit
>>>>> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
>>>>> List etiquette:
>>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>>
>>>> _______________________________________________
>>>> Wekalist mailing list -- [email protected]
>>>> Send posts to [email protected]
>>>> To unsubscribe send an email to [email protected]
>>>> To subscribe, unsubscribe, etc., visit
>>>> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
>>>> List etiquette:
>>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>>
>>> _______________________________________________
>>> Wekalist mailing list -- [email protected]
>>> Send posts to [email protected]
>>> To unsubscribe send an email to [email protected]
>>> To subscribe, unsubscribe, etc., visit
>>> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
>>> List etiquette:
>>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>>
>> _______________________________________________
>> Wekalist mailing list -- [email protected]
>> Send posts to [email protected]
>> To unsubscribe send an email to [email protected]
>> To subscribe, unsubscribe, etc., visit
>> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
>> List etiquette:
>> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>>
> _______________________________________________
> Wekalist mailing list -- [email protected]
> Send posts to [email protected]
> To unsubscribe send an email to [email protected]
> To subscribe, unsubscribe, etc., visit
> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>

_______________________________________________
Wekalist mailing list -- [email protected]
Send posts to [email protected]
To unsubscribe send an email to [email protected]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html