How can I use Auto-model generated Model for test scoring?

hhaider5

Can anybody please give me an example.

Regards.

varunm1

Hello @hhaider5

Sorry for the delayed response as I am occupied with some other tasks. In the below figure you can observe, a naive bayes model process after automodel execution. I attached a store operator after the model and provided a repository location to store it. Once it is stored in the location after running the process, you can open new process and drag and drop it to apply on other data.


Hope this helps

Thanks
Varun

varunm1

Hello @hhaider5

I checked it, I see that they are also looking for individual predictions of each subject survival(yes or No). One of the best methods to get all the predictions is using Cross Validation technique as this trains and tests (predictions) on all samples of the dataset. I recommend you try cross-validation in auto model process, it need to be done manually and a bit challenging as there are many operators that need to be removed or modified in automodel to apply this strategy.

If you are new and need to know about cross-validation, here is the best post from our @sgenzer about it.

https://community.rapidminer.com/discussion/55112/cross-validation-and-its-outputs-in-rm-studio#latest

Try it and see how it goes.

Thanks
Varun

varunm1

Hi @hhaider5

You can open the automodel process once execution is done and observe the process. From there you can store the trained model as well using store operator. Hoe to open process after automodel is given below.

https://community.rapidminer.com/discussion/comment/58971#Comment_58971

Inform here if you need more info.

hhaider5

Hey there @varunm1
     I want to know how can I store the trained model using store operator?
    Can you give me a video tutorial or anything?
It would be really helpful.
I actually want to use the trained model to predict survival in the Titanic set?
How can I do it?
    I would really appreciate your help.
   Thanks.

hhaider5

Got it.
It was easy.

hhaider5

Hey there @varunm1
   Yeah: I got it. Tyvm
   I want to know how can I increase the accuracy of my model? For although the auto-model shows the accuracy of the model to be 83%, I only got 77% in the competition.
What could be the reason for this discrepancy?


Regards.

varunm1

Hello @hhaider5

If you closely observe the auto model, it is splitting the data into 60:40 train : test data. There is some chance that the data might be randomly good during testing and give better results, to mitigate this the auto model have multiple hold out sets that are tested as well. 

I have a question, are you testing on data that is completely different from the Titanic data present in RapidMiner or a part of it?

hhaider5

Hey there @varunm1
    Let me give you the link to the competition. This may help you understand my Q.

Link:   https://www.kaggle.com/c/titanic

navigate to the "data" section, please.