How to check the accuracy of a model without submitting on Kaggle?




I want to check the accuracy of a model on the test set of the data provided on one of the Kaggle competitions. How can I do this without submitting on Kaggle?



Do they give you the reponse variables along with the test/CV set? Unless you have access to those, I don’t think it is possible. (Someone please correct me if I’m wrong.)


I don’t think they do.


Hi Aditya,

It’s not possible to check your model accuracy over test dataset in Kaggle without submitting it on Kaggle, But you can create training(~80%) and testing(~20%) data set from your test dataset itself and can check the model accuracy over it.

Hope this helps

Aayush Agrawal


Hi Aditya,

As @anon and @aayushmnit said, it is not possible to check the accuracy without submitting on Kaggle for most of the contests. However, it depends on contest to contest. For some contests, where you are trying for an absolute goal (e.g. number of hours in which Santa’s helpers can produce all the toys), you can get a sense of where your outcome would end up.

For very large datasets, you can also split the dataset into test and train and create a replica of the quality measure.

Hope this helps