I am sorry for the basic question but I am new to this. What is the meaning of "The train and test dataset are split based on time, and the public/private leaderboard in the test data are split randomly."
", "messages": [ { "id": "119762", "postDate": "05/12/2016 17:10:12", "content": "I am sorry for the basic question but I am new to this. What is the meaning of "The train and test dataset are split based on time, and the public/private leaderboard in the test data are split randomly."
", "rawMarkdown": "I am sorry for the basic question but I am new to this. What is the meaning of \"The train and test dataset are split based on time, and the public/private leaderboard in the test data are split randomly.\"", "votes": null }, { "id": "119868", "postDate": "05/13/2016 11:15:28", "content": "The train data contains all records before a cut-off date and the test data contains all records after the cut-off date. This replicates a real-world scenario better than randomly choosing a holdout sample as we are using data to make predictions about future, unseen events.
\n\nThe test data is then randomly split further into two parts. Our predictions on the first part are evaluated and used to give the Public leaderboard scores (ie the scores/rankings we can see now), whilst our predictions on the second part will be used at the end to calculate the final scores.
\n\nThe scores on the private leaderboard part of the test data remain unseen to us until after the competition closes, which prevents someone from being able to win by using the leaderboard scores to overfit to the test data.
", "rawMarkdown": "The train data contains all records before a cut-off date and the test data contains all records after the cut-off date. This replicates a real-world scenario better than randomly choosing a holdout sample as we are using data to make predictions about future, unseen events.\r\n\r\nThe test data is then randomly split further into two parts. Our predictions on the first part are evaluated and used to give the Public leaderboard scores (ie the scores/rankings we can see now), whilst our predictions on the second part will be used at the end to calculate the final scores.\r\n\r\nThe scores on the private leaderboard part of the test data remain unseen to us until after the competition closes, which prevents someone from being able to win by using the leaderboard scores to overfit to the test data.", "votes": null }, { "id": "119877", "postDate": "05/13/2016 13:10:27", "content": "Great answer Thank you, this is all what I need it to know to win this thing.
", "rawMarkdown": "Great answer Thank you, this is all what I need it to know to win this thing.", "votes": null } ], "comments": [ { "id": 119868, "author_name": "peatle", "author_url": "", "post_date": "05/13/2016 11:15:28", "content": "The train data contains all records before a cut-off date and the test data contains all records after the cut-off date. This replicates a real-world scenario better than randomly choosing a holdout sample as we are using data to make predictions about future, unseen events.
\n\nThe test data is then randomly split further into two parts. Our predictions on the first part are evaluated and used to give the Public leaderboard scores (ie the scores/rankings we can see now), whilst our predictions on the second part will be used at the end to calculate the final scores.
\n\nThe scores on the private leaderboard part of the test data remain unseen to us until after the competition closes, which prevents someone from being able to win by using the leaderboard scores to overfit to the test data.
", "votes": null, "replies": [] }, { "id": 119877, "author_name": "carlosdelamora", "author_url": "", "post_date": "05/13/2016 13:10:27", "content": "Great answer Thank you, this is all what I need it to know to win this thing.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "119762": "I am sorry for the basic question but I am new to this. What is the meaning of \"The train and test dataset are split based on time, and the public/private leaderboard in the test data are split randomly.\"", "119868": "The train data contains all records before a cut-off date and the test data contains all records after the cut-off date. This replicates a real-world scenario better than randomly choosing a holdout sample as we are using data to make predictions about future, unseen events.\r\n\r\nThe test data is then randomly split further into two parts. Our predictions on the first part are evaluated and used to give the Public leaderboard scores (ie the scores/rankings we can see now), whilst our predictions on the second part will be used at the end to calculate the final scores.\r\n\r\nThe scores on the private leaderboard part of the test data remain unseen to us until after the competition closes, which prevents someone from being able to win by using the leaderboard scores to overfit to the test data.", "119877": "Great answer Thank you, this is all what I need it to know to win this thing." }, "source": "meta" }