Skip to content

Variable importance bar plot

4 messages · Sarah Goslee, Bede-Fazekas Ákos, Enoch Gyamfi Ampadu

Original
Enoch Gyamfi Ampadu
# 
Dear List,

I hope this finds you well. Please I have done a Land cover classification
using randomForest. I used selected satellite image bands as input
variables. I have four thematic classes. I obtained the general variable
importance using *varImpPlot (my model) . *Now I want to have bar chart
plot showing each thematic class and the variable importance for each
class. That is, how each band is ranked in predicting a thematic class. I
tried using this function:

*importance = varImp(my model, scale= TRUE)*

*plot (importance. *

However it did not work. I tried to put in *geom= "bar*". And I had error.

Please I will be glad to get some assistance.

Thank you.

Best regards,

Enoch


Enoch Gyamfi - Ampadu

email: egampadu at gmail.com
skype: enoch.ampadu

The highest sense of nobility is self control.

A simple act of kindness creates endless ripple
2 days later
Sarah Goslee
# 
It's not entirely clear to me how you expect the values to be
calculated, but you might look at partial dependence plots, for
example as implemented in the pdp package.

Sarah
On Sun, Oct 27, 2019 at 6:00 AM Enoch Gyamfi Ampadu <egampadu at gmail.com> wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Bede-Fazekas Ákos
    

    

      
      #
    

  


  

      
Dear Enoch,

randomForest:varImpPlot() has a parameter called 'class' that is NULL by 
default but can be changed to get the class-specific measure of importance.
'geom= "bar"': do you want to plot the result using ggplot2 instead of 
the base plot produced by varImpPlot()? In this case you should firstly 
assign the invisiby returning result of varImpPlot() (or that of 
importance()) to a variable, and then plot it with ggplot().

HTH,
?kos Bede-Fazekas
Hungarian Academy of Sciences


2019.10.29. 15:51 keltez?ssel, Sarah Goslee ?rta:

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Enoch Gyamfi Ampadu
    

    

      
      #
    

  


  

      
Hi Sarah,

Thank you for getting in touch and providing a suggestion.  I have checked
but that is not what want actually.

 I want to have a vertical bar plot depicting how each variable contrite to
predicting the class. I used 8 bands of landsat image. So let's say if one
of the  land cover class to be Built up, I want to have bar plots in
ranking order of the bands to enable me compare their importance in
predicting the land cover class. I don't know if this explains what I want
better.

Thank you.

Best regards,

Enoch

Enoch Gyamfi - Ampadu

email: egampadu at gmail.com
skype: enoch.ampadu

The highest sense of nobility is self control.

A simple act of kindness creates endless ripple

        
On Tue, 29 Oct 2019, 16:51 Sarah Goslee, <sarah.goslee at gmail.com> wrote: