Has anyone managed to come up with an idea for this yet? The minimum value I'm predicting for my test set is around -15, so these particular rows in the training set could just potentially be outliers which the Elo people encoded with a certain unrealistically low value?
", "messages": [ { "id": "432622", "postDate": "12/04/2018 05:32:09", "content": "Has anyone managed to come up with an idea for this yet? The minimum value I'm predicting for my test set is around -15, so these particular rows in the training set could just potentially be outliers which the Elo people encoded with a certain unrealistically low value?
", "rawMarkdown": "Has anyone managed to come up with an idea for this yet? The minimum value I'm predicting for my test set is around -15, so these particular rows in the training set could just potentially be outliers which the Elo people encoded with a certain unrealistically low value?", "votes": null }, { "id": "432703", "postDate": "12/04/2018 08:16:38", "content": "The entire dataset is supposed to be simulated, so your explanation is quite probable.
", "rawMarkdown": "The entire dataset is supposed to be simulated, so your explanation is quite probable.", "votes": null }, { "id": "434308", "postDate": "12/06/2018 07:19:05", "content": "Since most of the rmse score is apparently due to these outliers I tried to classify them first and use that information for the regression, but not working yet.l
", "rawMarkdown": "Since most of the rmse score is apparently due to these outliers I tried to classify them first and use that information for the regression, but not working yet.l", "votes": null }, { "id": "434406", "postDate": "12/06/2018 10:53:00", "content": "Predicting the outliers seems very important in my model.
", "rawMarkdown": "Predicting the outliers seems very important in my model.", "votes": null }, { "id": "435978", "postDate": "12/09/2018 07:33:44", "content": "I've found that -33.21928095 (all scores < 30 are equal to that value) is a -10*log2(10), so it looks like some synthetical value. \nIt may be a result of some value overflow or it was added as a synthetic noise to train data?
", "rawMarkdown": "I've found that -33.21928095 (all scores < 30 are equal to that value) is a -10*log2(10), so it looks like some synthetical value. \nIt may be a result of some value overflow or it was added as a synthetic noise to train data?", "votes": null }, { "id": "435980", "postDate": "12/09/2018 07:44:56", "content": "My guess would be the latter, but in that case, why give it such a very formulaic value?
", "rawMarkdown": "My guess would be the latter, but in that case, why give it such a very formulaic value?", "votes": null }, { "id": "445951", "postDate": "12/27/2018 09:05:18", "content": "has anyone tried removing the outliers from data and then train it
", "rawMarkdown": "has anyone tried removing the outliers from data and then train it", "votes": null }, { "id": "446188", "postDate": "12/27/2018 17:06:50", "content": "At this point I'm currently focusing all my efforts in the Outliers Classification, I tried so many feature combinations with the Regression part and none of them seem to have effect improving RMSE, I even tried trend features over time.
", "rawMarkdown": "At this point I'm currently focusing all my efforts in the Outliers Classification, I tried so many feature combinations with the Regression part and none of them seem to have effect improving RMSE, I even tried trend features over time.", "votes": null }, { "id": "446470", "postDate": "12/28/2018 06:01:04", "content": "Nice observation!
", "rawMarkdown": "Nice observation!", "votes": null }, { "id": "466858", "postDate": "02/06/2019 04:50:09", "content": "spot on! Thx, I was wondering why that number!
", "rawMarkdown": "spot on! Thx, I was wondering why that number!", "votes": null } ], "comments": [ { "id": 432703, "author_name": "konradb", "author_url": "", "post_date": "12/04/2018 08:16:38", "content": "The entire dataset is supposed to be simulated, so your explanation is quite probable.
", "votes": null, "replies": [] }, { "id": 434308, "author_name": "guntherthepenguin", "author_url": "", "post_date": "12/06/2018 07:19:05", "content": "Since most of the rmse score is apparently due to these outliers I tried to classify them first and use that information for the regression, but not working yet.l
", "votes": null, "replies": [] }, { "id": 434406, "author_name": "jfr311", "author_url": "", "post_date": "12/06/2018 10:53:00", "content": "Predicting the outliers seems very important in my model.
", "votes": null, "replies": [] }, { "id": 435978, "author_name": "andreylat", "author_url": "", "post_date": "12/09/2018 07:33:44", "content": "I've found that -33.21928095 (all scores < 30 are equal to that value) is a -10*log2(10), so it looks like some synthetical value. \nIt may be a result of some value overflow or it was added as a synthetic noise to train data?
", "votes": null, "replies": [ { "id": 435980, "author_name": "delayedkarma", "author_url": "", "post_date": "12/09/2018 07:44:56", "content": "My guess would be the latter, but in that case, why give it such a very formulaic value?
", "votes": null, "replies": [] }, { "id": 446470, "author_name": "salamatburj", "author_url": "", "post_date": "12/28/2018 06:01:04", "content": "Nice observation!
", "votes": null, "replies": [] }, { "id": 466858, "author_name": "liwenhuang", "author_url": "", "post_date": "02/06/2019 04:50:09", "content": "spot on! Thx, I was wondering why that number!
", "votes": null, "replies": [] } ] }, { "id": 445951, "author_name": "abhinav97dutt", "author_url": "", "post_date": "12/27/2018 09:05:18", "content": "has anyone tried removing the outliers from data and then train it
", "votes": null, "replies": [] }, { "id": 446188, "author_name": "cv13j0", "author_url": "", "post_date": "12/27/2018 17:06:50", "content": "At this point I'm currently focusing all my efforts in the Outliers Classification, I tried so many feature combinations with the Regression part and none of them seem to have effect improving RMSE, I even tried trend features over time.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "432622": "Has anyone managed to come up with an idea for this yet? The minimum value I'm predicting for my test set is around -15, so these particular rows in the training set could just potentially be outliers which the Elo people encoded with a certain unrealistically low value?", "432703": "The entire dataset is supposed to be simulated, so your explanation is quite probable.", "434308": "Since most of the rmse score is apparently due to these outliers I tried to classify them first and use that information for the regression, but not working yet.l", "434406": "Predicting the outliers seems very important in my model.", "435978": "I've found that -33.21928095 (all scores < 30 are equal to that value) is a -10*log2(10), so it looks like some synthetical value. \nIt may be a result of some value overflow or it was added as a synthetic noise to train data?", "435980": "My guess would be the latter, but in that case, why give it such a very formulaic value?", "445951": "has anyone tried removing the outliers from data and then train it", "446188": "At this point I'm currently focusing all my efforts in the Outliers Classification, I tried so many feature combinations with the Regression part and none of them seem to have effect improving RMSE, I even tried trend features over time.", "446470": "Nice observation!", "466858": "spot on! Thx, I was wondering why that number!" }, "source": "meta" }