Timeline for Similarity between Train and Test data sets
Current License: CC BY-SA 4.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 18, 2018 at 13:47 | comment | added | Ambarish | It's the Former, my train and test are different files. | |
| Aug 18, 2018 at 13:44 | comment | added | Frans Rodenburg | I don't understand that. Do you mean to say the train and test set are different files, or that you generate them from different files? In case of the former, that just seems like a technical issue, not a reason to not make a new split. In case of the latter, you'll have to give more details on how and why you're doing this. | |
| Aug 18, 2018 at 12:28 | comment | added | Ambarish | I have two separate files, one each for test and train for each class i.e, for example, class 1 train and class 1 test. I am sampling the validation set from the available test set. That's why I think the test and train data could be from different distribution? I wanted to know if it is possible to somehow figure this out? The results are still poor even if I perform an independent evaluation of the model using the available test set. To make things more clear, I want to know if class 1 train and class 1 test could come from different distributions? Could there be a "covariate shift" ? | |
| Aug 18, 2018 at 12:01 | comment | added | Frans Rodenburg | Have you tried splitting stratified? Or just a different split? | |
| Aug 18, 2018 at 11:28 | history | asked | Ambarish | CC BY-SA 4.0 |