It is not a very significant problem. But I tought i should mention it might generally be better to encode target / signals as -1, 0, 1 for down (sell) / flat (keep) / up (buy). It make slightly more sense and allow to deal with the problem as a regression problem (after all down is the opposite of up).
", "messages": [ { "id": "2316960", "postDate": "06/25/2023 10:45:38", "content": "It is not a very significant problem. But I tought i should mention it might generally be better to encode target / signals as -1, 0, 1 for down (sell) / flat (keep) / up (buy). It make slightly more sense and allow to deal with the problem as a regression problem (after all down is the opposite of up).
", "rawMarkdown": "It is not a very significant problem. But I tought i should mention it might generally be better to encode target / signals as -1, 0, 1 for down (sell) / flat (keep) / up (buy). It make slightly more sense and allow to deal with the problem as a regression problem (after all down is the opposite of up).", "votes": null }, { "id": "2317262", "postDate": "06/25/2023 14:48:09", "content": "Thank you for being here, Lucas! I've updated test and submission files, also checked them: 0s are inside. Also I changed metrics to macro avg f1, because it gives better look at model behavior
", "rawMarkdown": "Thank you for being here, Lucas! I've updated test and submission files, also checked them: 0s are inside. Also I changed metrics to macro avg f1, because it gives better look at model behavior", "votes": null }, { "id": "2319802", "postDate": "06/27/2023 10:19:33", "content": "Thanks, nice for the update. Maybe you want to do that for the target too.
\nRegarding the change of metric, I am affraid that kaggle hasn't updated the score… thus I am appearing with my old 0.98 score. (I don't know where to go with this but I'll give kaggle a feed back that they should update scores if metric is updated).
Hi Lucas. I feel myself really weird, inviting you every time in my competitions, but here I got a new one - https://www.kaggle.com/competitions/shortlong-prediction
\nYou know, I decided to go smarter and split my previous competition into a small once: maybe first of all, rather than trying to predict 3 classes, try to predict one as the majority and one (which is now two mitorities combined) as minority. And then among them predict which one is short and which one is long.
\nSo, url above leads to this second part task, predicting long or short. Probably you are tired of me already, but again, I am always happy to see you in my competitions.
\nP.S. Second one - https://www.kaggle.com/competitions/anomaly-class-detection
", "rawMarkdown": "Hi Lucas. I feel myself really weird, inviting you every time in my competitions, but here I got a new one - https://www.kaggle.com/competitions/shortlong-prediction \n\nYou know, I decided to go smarter and split my previous competition into a small once: maybe first of all, rather than trying to predict 3 classes, try to predict one as the majority and one (which is now two mitorities combined) as minority. And then among them predict which one is short and which one is long. \n\nSo, url above leads to this second part task, predicting long or short. Probably you are tired of me already, but again, I am always happy to see you in my competitions.\n\nP.S. Second one - https://www.kaggle.com/competitions/anomaly-class-detection", "votes": null }, { "id": "2321061", "postDate": "06/28/2023 08:51:35", "content": "Hey nice; i'll look into them. Dont be ashamed, that's infinitely better than doing nothing.
\nI think you should mention the competition somewhere (better visibility, organisation feedback), maybe in the discussion forum ?
Thank you for being here, Lucas! I've updated test and submission files, also checked them: 0s are inside. Also I changed metrics to macro avg f1, because it gives better look at model behavior
", "votes": null, "replies": [ { "id": 2319802, "author_name": "lucasmorin", "author_url": "", "post_date": "06/27/2023 10:19:33", "content": "Thanks, nice for the update. Maybe you want to do that for the target too.
\nRegarding the change of metric, I am affraid that kaggle hasn't updated the score… thus I am appearing with my old 0.98 score. (I don't know where to go with this but I'll give kaggle a feed back that they should update scores if metric is updated).
Hi Lucas. I feel myself really weird, inviting you every time in my competitions, but here I got a new one - https://www.kaggle.com/competitions/shortlong-prediction
\nYou know, I decided to go smarter and split my previous competition into a small once: maybe first of all, rather than trying to predict 3 classes, try to predict one as the majority and one (which is now two mitorities combined) as minority. And then among them predict which one is short and which one is long.
\nSo, url above leads to this second part task, predicting long or short. Probably you are tired of me already, but again, I am always happy to see you in my competitions.
\nP.S. Second one - https://www.kaggle.com/competitions/anomaly-class-detection
", "votes": null, "replies": [ { "id": 2321061, "author_name": "lucasmorin", "author_url": "", "post_date": "06/28/2023 08:51:35", "content": "Hey nice; i'll look into them. Dont be ashamed, that's infinitely better than doing nothing.
\nI think you should mention the competition somewhere (better visibility, organisation feedback), maybe in the discussion forum ?