![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1748051%2F91ed08ef5539cfca4d694e33fb7264f4%2Fweight.PNG?generation=1729708791804516&alt=media\")
Hi Kagglers,
\nI was wondering about whether using the scoring weight would be considered valid in this competition as it isn't technically a feature in the data but only used for evaluating the models?
Let me know if you plan on using it to train your model to tailor your predictions according to its evaluation weighting!
", "messages": [ { "id": 3026397, "postDate": "2024-10-23T18:41:30.567Z", "content": "Yes, sure. You can use whatever you like as features. I thought that using weight may be helpful because (presumably) JS is puting higher weight on more liquid instruments, so weight might be a proxy for other characteristics.
\nIn vanilla lgb, I do not find that weight is very helpful though. If you find otherwise, let us know.
\n
This is the shapley beeswarm plot for the weight feature. Since most of the data is on the 0 line, it meanse that weight does not play an important role in any of those predictions. The values away from 0 are all similar color (mostly), meaning that the weight impact in those situations is coming from an interaction, not a direct effect (hence weight is prob not that important).
", "rawMarkdown": "Yes, sure. You can use whatever you like as features. I thought that using weight may be helpful because (presumably) JS is puting higher weight on more liquid instruments, so weight might be a proxy for other characteristics. \n\nIn vanilla lgb, I do not find that weight is very helpful though. If you find otherwise, let us know.\n\n\n\nThis is the shapley beeswarm plot for the weight feature. Since most of the data is on the 0 line, it meanse that weight does not play an important role in any of those predictions. The values away from 0 are all similar color (mostly), meaning that the weight impact in those situations is coming from an interaction, not a direct effect (hence weight is prob not that important).", "votes": 3, "replies": [ { "id": 3026432, "postDate": "2024-10-23T19:30:14.277Z", "content": "Ah, thank you for the clarification. I will look into it and see what I find!
", "rawMarkdown": "Ah, thank you for the clarification. I will look into it and see what I find!" } ] }, { "id": 3026276, "postDate": "2024-10-23T15:32:32.417Z", "content": "Hi Kagglers,
\nI was wondering about whether using the scoring weight would be considered valid in this competition as it isn't technically a feature in the data but only used for evaluating the models?
Let me know if you plan on using it to train your model to tailor your predictions according to its evaluation weighting!
", "rawMarkdown": "Hi Kagglers,\n\nI was wondering about whether using the scoring `weight` would be considered valid in this competition as it isn't technically a feature in the data but only used for evaluating the models?\n\nLet me know if you plan on using it to train your model to tailor your predictions according to its evaluation weighting!", "votes": 1 }, { "id": 3026376, "postDate": "2024-10-23T18:14:24.970Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 3026397, "author_name": "marketneutral", "author_url": "", "post_date": "2024-10-23T18:41:30.567000", "content": "Yes, sure. You can use whatever you like as features. I thought that using weight may be helpful because (presumably) JS is puting higher weight on more liquid instruments, so weight might be a proxy for other characteristics.
\nIn vanilla lgb, I do not find that weight is very helpful though. If you find otherwise, let us know.
\n
This is the shapley beeswarm plot for the weight feature. Since most of the data is on the 0 line, it meanse that weight does not play an important role in any of those predictions. The values away from 0 are all similar color (mostly), meaning that the weight impact in those situations is coming from an interaction, not a direct effect (hence weight is prob not that important).
", "votes": 3, "replies": [ { "id": 3026432, "author_name": "JordanYHChan", "author_url": "", "post_date": "2024-10-23T19:30:14.277000", "content": "Ah, thank you for the clarification. I will look into it and see what I find!
", "votes": 0, "replies": [] } ] }, { "id": 3026376, "author_name": "", "author_url": "", "post_date": "2024-10-23T18:14:24.970000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3026397": "Yes, sure. You can use whatever you like as features. I thought that using weight may be helpful because (presumably) JS is puting higher weight on more liquid instruments, so weight might be a proxy for other characteristics. \n\nIn vanilla lgb, I do not find that weight is very helpful though. If you find otherwise, let us know.\n\n\n\nThis is the shapley beeswarm plot for the weight feature. Since most of the data is on the 0 line, it meanse that weight does not play an important role in any of those predictions. The values away from 0 are all similar color (mostly), meaning that the weight impact in those situations is coming from an interaction, not a direct effect (hence weight is prob not that important).", "3026276": "Hi Kagglers,\n\nI was wondering about whether using the scoring `weight` would be considered valid in this competition as it isn't technically a feature in the data but only used for evaluating the models?\n\nLet me know if you plan on using it to train your model to tailor your predictions according to its evaluation weighting!", "3026376": "" } }