I've just come to this competition a few days ago and I have some questions about the test data, which are probably stupid because I've missed some information.
\nthe test data we have now is not the final test data, so what is the final test data? The training data is from the first 4 weeks, the test data is from week 5, and the final test data is from week 6 or later?
The public lb and private lb are 33% and 67% respectively, how is this divided? Is it correct that the public lb is for week 5 and the private lb is for weeks 6 and 7?
I've just come to this competition a few days ago and I have some questions about the test data, which are probably stupid because I've missed some information.
\nthe test data we have now is not the final test data, so what is the final test data? The training data is from the first 4 weeks, the test data is from week 5, and the final test data is from week 6 or later?
The public lb and private lb are 33% and 67% respectively, how is this divided? Is it correct that the public lb is for week 5 and the private lb is for weeks 6 and 7?
Hey @chenlin1999!
\nThe test data is from the period straight after train. So yes, it is week #5.
\nThe entirety of the test data is from week 5, so the public LB is just randomly sampled 33% and the private LB is the remaining 67%. The test set comes from the same distribution (week 5 of the data with train ending with week 4).
\nHope this helps! 🙂 All the best in the competition!
", "rawMarkdown": "Hey @chenlin1999! \n\nThe test data is from the period straight after train. So yes, it is week #5.\n\nThe entirety of the test data is from week 5, so the public LB is just randomly sampled 33% and the private LB is the remaining 67%. The test set comes from the same distribution (week 5 of the data with train ending with week 4).\n\nHope this helps! 🙂 All the best in the competition!", "votes": null }, { "id": "2090991", "postDate": "01/07/2023 22:18:49", "content": "Thank you for your reply!
\nMy problem was solved. I wasn't sure why we were making predictions, as we all already had test data that we could just write to csv. I did some eda, but found that the test data didn't look much different from the train data, so I didn't understand at first why the test data was incomplete until I saw Chris's reply here. Turns out it's because we only have half of the data from week 5.
\nThanks to you and Chris for providing so much quality information for all!
", "rawMarkdown": "Thank you for your reply!\n\nMy problem was solved. I wasn't sure why we were making predictions, as we all already had test data that we could just write to csv. I did some eda, but found that the test data didn't look much different from the train data, so I didn't understand at first why the test data was incomplete until I saw Chris's reply [here](https://www.kaggle.com/competitions/otto-recommender-system/discussion/367154#2037877). Turns out it's because we only have half of the data from week 5.\n\nThanks to you and Chris for providing so much quality information for all!", "votes": null } ], "comments": [ { "id": 2090146, "author_name": "radek1", "author_url": "", "post_date": "01/07/2023 01:29:59", "content": "Hey @chenlin1999!
\nThe test data is from the period straight after train. So yes, it is week #5.
\nThe entirety of the test data is from week 5, so the public LB is just randomly sampled 33% and the private LB is the remaining 67%. The test set comes from the same distribution (week 5 of the data with train ending with week 4).
\nHope this helps! 🙂 All the best in the competition!
", "votes": null, "replies": [ { "id": 2090991, "author_name": "chenlin1999", "author_url": "", "post_date": "01/07/2023 22:18:49", "content": "Thank you for your reply!
\nMy problem was solved. I wasn't sure why we were making predictions, as we all already had test data that we could just write to csv. I did some eda, but found that the test data didn't look much different from the train data, so I didn't understand at first why the test data was incomplete until I saw Chris's reply here. Turns out it's because we only have half of the data from week 5.
\nThanks to you and Chris for providing so much quality information for all!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2090048": "I've just come to this competition a few days ago and I have some questions about the test data, which are probably stupid because I've missed some information.\n\n1. the test data we have now is not the final test data, so what is the final test data? The training data is from the first 4 weeks, the test data is from week 5, and the final test data is from week 6 or later?\n\n2. The public lb and private lb are 33% and 67% respectively, how is this divided? Is it correct that the public lb is for week 5 and the private lb is for weeks 6 and 7?", "2090146": "Hey @chenlin1999! \n\nThe test data is from the period straight after train. So yes, it is week #5.\n\nThe entirety of the test data is from week 5, so the public LB is just randomly sampled 33% and the private LB is the remaining 67%. The test set comes from the same distribution (week 5 of the data with train ending with week 4).\n\nHope this helps! 🙂 All the best in the competition!", "2090991": "Thank you for your reply!\n\nMy problem was solved. I wasn't sure why we were making predictions, as we all already had test data that we could just write to csv. I did some eda, but found that the test data didn't look much different from the train data, so I didn't understand at first why the test data was incomplete until I saw Chris's reply [here](https://www.kaggle.com/competitions/otto-recommender-system/discussion/367154#2037877). Turns out it's because we only have half of the data from week 5.\n\nThanks to you and Chris for providing so much quality information for all!" }, "source": "meta" }