Different Accuracy Each Time

mirsaty

 For a dataset (but not for different datasets) I get deffirent auto model GLM Learning algorithm accuracy. For my personel PC ı use educational ı get same accuracy 77.5 - for trial automodel different pc ı get different accuray 76 - 39 range. I can post video for proof. I use other pc auto model design with my pc ı get 77.5 again.

rfuentealba

Hello, @mirsaty,

Are you able to post the XML for your solution, so that we can figure it out? I normally get different accuracies for runs with the same dataset when I don't set a random seed that ensures me replicable results. May this be the case?

All the best,

Rod.

varunm1

Hello @mirsaty

Can you please check if the split data operator has local random seed set? This is set by default with 1992 as the seed number. Generally, this helps in producing a reproducible result. I am sure @IngoRM gave it a thought and set the seed by default, I cross checked this. Also, can you inform which Rapidminer versions are in these systems?

To check this, you need to open the auto model process in rapidminer by clicking 


Once you open the process, you can see a split data operator, you can click on that and see parameters in the parameter window as shown below.



I do have one more question. What is the sample size of your data? If this is more than 10000 samples, I recommend you to set a random seed for Sample FE operator as well.

If you still have a huge variation in the performance, please inform here.