none
Increase the samples in a "Evaluate model" RRS feed

  • Question

  • My dataset have more than 10.000 records but when i make "evaluate model" with some classificats only use 60 samples, why? i want make a validation with split data partition but doesn't works. If change to two class decision jungle make the validation with thousands of records, but GBT, Neural Networks don´t, Any advice?



    Wednesday, August 21, 2019 7:45 PM

All replies

  • Hello Andres,

    Evaluate model basically measures the accuracy of a trained model. You provide a dataset containing scores generated from a model, and the Evaluate Model module computes a set of industry-standard evaluation metrics. These metrics may vary depending on the type of model you are evaluating. In your case it looks like you are using classification models.

    For the issue you are facing is it a case where only 60 records of your dataset are used for your respective score models from the split dataset?

    -Rohit

    Thursday, August 22, 2019 10:18 AM
    Moderator
  • Thanks for your reply if you see some models use a big quantity of records (3000), but other only 60 o o 70 i can'ttrust in this evaluation i need evaluate with thousands. How can I configure this parameter. 

    Thursday, August 22, 2019 1:25 PM
  • Hi, Could you please share the details/screenshot that you have set in properties of 'split data' and please follow the below screenshot steps and share the details/screenshot for number of rows present in split data of 1 and 2 Result data set. 



    Friday, August 23, 2019 10:20 AM
  •  if you see the architecture of the experiment you will see both graphs is from the same dataset but different model.


    Saturday, August 24, 2019 5:53 PM
  • Hi,

    Can you please add your project in https://gallery.azure.ai/ and share the link. So that I can reproduce the issue.

    Thanks

    Wednesday, August 28, 2019 9:23 AM
  • Hi Andres,

    Do you have any update for this case? Hope everything is good now.

    Regards,

    Yutong

    Tuesday, September 3, 2019 6:51 PM
    Moderator