I discovered an easy way to cheat and win, and I wanted to share it now that we can no longer make submissions so that the admins can check for those that might have cheated.
\n\nThe data is not very helpful for making predictions, especially during such a small time period. So making an accurate prediction on just a few people that you know will make purchases can greatly increase your score.
\n\nBut here's the key point, most of us are part of the test submission! If you visited the merchandise store, you are one of the IDs that you submitted a prediction for. It is very easy to figure out which ID is your by tracking visit times and so on. If you make a prediction for yourself of $10 and then buy $10 of merchandise between December and January, that would be a big advantage. You could also organize with friends and potentially have 10 or 20 people buy small items. If you wanted to be even more devious, you could purchase tons of really expensive items on January 31st and then cancel your orders/request refunds in early February.
\n\nI did not cheat in this way, but I am pointing it out now that the submission is over in the hopes that the admins will look for it.
", "messages": [ { "id": "431021", "postDate": "12/01/2018 13:05:46", "content": "I discovered an easy way to cheat and win, and I wanted to share it now that we can no longer make submissions so that the admins can check for those that might have cheated.
\n\nThe data is not very helpful for making predictions, especially during such a small time period. So making an accurate prediction on just a few people that you know will make purchases can greatly increase your score.
\n\nBut here's the key point, most of us are part of the test submission! If you visited the merchandise store, you are one of the IDs that you submitted a prediction for. It is very easy to figure out which ID is your by tracking visit times and so on. If you make a prediction for yourself of $10 and then buy $10 of merchandise between December and January, that would be a big advantage. You could also organize with friends and potentially have 10 or 20 people buy small items. If you wanted to be even more devious, you could purchase tons of really expensive items on January 31st and then cancel your orders/request refunds in early February.
\n\nI did not cheat in this way, but I am pointing it out now that the submission is over in the hopes that the admins will look for it.
", "rawMarkdown": "I discovered an easy way to cheat and win, and I wanted to share it now that we can no longer make submissions so that the admins can check for those that might have cheated.\n\nThe data is not very helpful for making predictions, especially during such a small time period. So making an accurate prediction on just a few people that you know will make purchases can greatly increase your score.\n\n**But here's the key point, most of us are part of the test submission!** If you visited the merchandise store, you are one of the IDs that you submitted a prediction for. It is very easy to figure out which ID is your by tracking visit times and so on. If you make a prediction for yourself of $10 and then buy $10 of merchandise between December and January, that would be a big advantage. You could also organize with friends and potentially have 10 or 20 people buy small items. If you wanted to be even more devious, you could purchase tons of really expensive items on January 31st and then cancel your orders/request refunds in early February.\n\nI did not cheat in this way, but I am pointing it out now that the submission is over in the hopes that the admins will look for it.", "votes": null }, { "id": "431036", "postDate": "12/01/2018 13:51:54", "content": "I think this would only work if you first visited by October 15th, but thanks for highlighting this as some people may have done exactly that. Admins please incorporate this in your cheat detection algorithms e.g. by IP address.
", "rawMarkdown": "I think this would only work if you first visited by October 15th, but thanks for highlighting this as some people may have done exactly that. Admins please incorporate this in your cheat detection algorithms e.g. by IP address.", "votes": null }, { "id": "431038", "postDate": "12/01/2018 13:55:05", "content": "Yeah. I think it was understandable since competition was update. Just one more ridiculous thing about it.
", "rawMarkdown": "Yeah. I think it was understandable since competition was update. Just one more ridiculous thing about it.", "votes": null }, { "id": "431041", "postDate": "12/01/2018 13:59:14", "content": "I think this was noted in this discussion, too. ( https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/69592 ) I am not sure how Kaggle will be able to identify these people.
", "rawMarkdown": "I think this was noted in this discussion, too. ( https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/69592 ) I am not sure how Kaggle will be able to identify these people.", "votes": null }, { "id": "431117", "postDate": "12/01/2018 16:22:28", "content": "I think it is quite easy to detect. If someone wants to get reasonable advantage by such cheating, he would have to increase those predictions for himself or his friend quite high. And statistically it will look really suspicious if that high prediction will guess real purchase, especially if there's nothing special about that visitor in history and all other submissions will predict it close to 0.
\n\nClearly not worth spending $10 for unnecessary items just to get banned out at the end... there's many better ways to invest those $10 and your time ;)
", "rawMarkdown": "I think it is quite easy to detect. If someone wants to get reasonable advantage by such cheating, he would have to increase those predictions for himself or his friend quite high. And statistically it will look really suspicious if that high prediction will guess real purchase, especially if there's nothing special about that visitor in history and all other submissions will predict it close to 0. \n\nClearly not worth spending $10 for unnecessary items just to get banned out at the end... there's many better ways to invest those $10 and your time ;)", "votes": null }, { "id": "432144", "postDate": "12/03/2018 13:11:54", "content": "I'm getting a 404 error when I click your link.
", "rawMarkdown": "I'm getting a 404 error when I click your link.", "votes": null }, { "id": "432147", "postDate": "12/03/2018 13:13:47", "content": "I agree it should be easy to catch as long as they look for it, I am just hoping they look for it.
", "rawMarkdown": "I agree it should be easy to catch as long as they look for it, I am just hoping they look for it.", "votes": null }, { "id": "432816", "postDate": "12/04/2018 11:00:45", "content": "A great line of thought indeed!
", "rawMarkdown": "A great line of thought indeed!", "votes": null } ], "comments": [ { "id": 431036, "author_name": "datahobbit", "author_url": "", "post_date": "12/01/2018 13:51:54", "content": "I think this would only work if you first visited by October 15th, but thanks for highlighting this as some people may have done exactly that. Admins please incorporate this in your cheat detection algorithms e.g. by IP address.
", "votes": null, "replies": [] }, { "id": 431038, "author_name": "artgor", "author_url": "", "post_date": "12/01/2018 13:55:05", "content": "Yeah. I think it was understandable since competition was update. Just one more ridiculous thing about it.
", "votes": null, "replies": [] }, { "id": 431041, "author_name": "returnofsputnik", "author_url": "", "post_date": "12/01/2018 13:59:14", "content": "I think this was noted in this discussion, too. ( https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/69592 ) I am not sure how Kaggle will be able to identify these people.
", "votes": null, "replies": [ { "id": 432144, "author_name": "afgonczol", "author_url": "", "post_date": "12/03/2018 13:11:54", "content": "I'm getting a 404 error when I click your link.
", "votes": null, "replies": [] } ] }, { "id": 431117, "author_name": "alijs1", "author_url": "", "post_date": "12/01/2018 16:22:28", "content": "I think it is quite easy to detect. If someone wants to get reasonable advantage by such cheating, he would have to increase those predictions for himself or his friend quite high. And statistically it will look really suspicious if that high prediction will guess real purchase, especially if there's nothing special about that visitor in history and all other submissions will predict it close to 0.
\n\nClearly not worth spending $10 for unnecessary items just to get banned out at the end... there's many better ways to invest those $10 and your time ;)
", "votes": null, "replies": [ { "id": 432147, "author_name": "afgonczol", "author_url": "", "post_date": "12/03/2018 13:13:47", "content": "I agree it should be easy to catch as long as they look for it, I am just hoping they look for it.
", "votes": null, "replies": [] } ] }, { "id": 432816, "author_name": "shrutimechlearn", "author_url": "", "post_date": "12/04/2018 11:00:45", "content": "A great line of thought indeed!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "431021": "I discovered an easy way to cheat and win, and I wanted to share it now that we can no longer make submissions so that the admins can check for those that might have cheated.\n\nThe data is not very helpful for making predictions, especially during such a small time period. So making an accurate prediction on just a few people that you know will make purchases can greatly increase your score.\n\n**But here's the key point, most of us are part of the test submission!** If you visited the merchandise store, you are one of the IDs that you submitted a prediction for. It is very easy to figure out which ID is your by tracking visit times and so on. If you make a prediction for yourself of $10 and then buy $10 of merchandise between December and January, that would be a big advantage. You could also organize with friends and potentially have 10 or 20 people buy small items. If you wanted to be even more devious, you could purchase tons of really expensive items on January 31st and then cancel your orders/request refunds in early February.\n\nI did not cheat in this way, but I am pointing it out now that the submission is over in the hopes that the admins will look for it.", "431036": "I think this would only work if you first visited by October 15th, but thanks for highlighting this as some people may have done exactly that. Admins please incorporate this in your cheat detection algorithms e.g. by IP address.", "431038": "Yeah. I think it was understandable since competition was update. Just one more ridiculous thing about it.", "431041": "I think this was noted in this discussion, too. ( https://www.kaggle.com/c/ga-customer-revenue-prediction/discussion/69592 ) I am not sure how Kaggle will be able to identify these people.", "431117": "I think it is quite easy to detect. If someone wants to get reasonable advantage by such cheating, he would have to increase those predictions for himself or his friend quite high. And statistically it will look really suspicious if that high prediction will guess real purchase, especially if there's nothing special about that visitor in history and all other submissions will predict it close to 0. \n\nClearly not worth spending $10 for unnecessary items just to get banned out at the end... there's many better ways to invest those $10 and your time ;)", "432144": "I'm getting a 404 error when I click your link.", "432147": "I agree it should be easy to catch as long as they look for it, I am just hoping they look for it.", "432816": "A great line of thought indeed!" }, "source": "meta" }