This is what I would think when the straight-gold kaggler(guess who?) is not in the gold region for a while; this either means overfitting or incoming shakeup? Let me know your thoughts?
", "messages": [ { "id": "698057", "postDate": "12/18/2019 18:41:57", "content": "This is what I would think when the straight-gold kaggler(guess who?) is not in the gold region for a while; this either means overfitting or incoming shakeup? Let me know your thoughts?
", "rawMarkdown": "This is what I would think when the straight-gold kaggler(guess who?) is not in the gold region for a while; this either means overfitting or incoming shakeup? Let me know your thoughts?", "votes": null }, { "id": "698157", "postDate": "12/18/2019 21:39:52", "content": "Do you mean the rank 1, bestfitting?\nNo, I don't think its overfitting, but rather a combination of good model architecture + a good way of encoding train data.
\n\nWhat is that way, I'm still trying to figure out :)
", "rawMarkdown": "Do you mean the rank 1, bestfitting?\nNo, I don't think its overfitting, but rather a combination of good model architecture + a good way of encoding train data.\n\nWhat is that way, I'm still trying to figure out :)", "votes": null }, { "id": "698240", "postDate": "12/19/2019 01:30:45", "content": "I don't think the top teams are overfitting either. With good model architecture, I think their score would be quite reasonable.
", "rawMarkdown": "I don't think the top teams are overfitting either. With good model architecture, I think their score would be quite reasonable.", "votes": null }, { "id": "698269", "postDate": "12/19/2019 02:05:53", "content": "Since we don't have enough information about metrics, we can not make reliable local validation and LB is the best compass (even Public LB has 200 images) for now.
\n\nSo I think we can't avoid overfitting without more metrics details. :(
", "rawMarkdown": "Since we don't have enough information about metrics, we can not make reliable local validation and LB is the best compass (even Public LB has 200 images) for now.\n\nSo I think we can't avoid overfitting without more metrics details. :(", "votes": null }, { "id": "698271", "postDate": "12/19/2019 02:11:05", "content": "Yes, I also hope the organizer can clear some information here.. Or this is the part of competition? lol
", "rawMarkdown": "Yes, I also hope the organizer can clear some information here.. Or this is the part of competition? lol", "votes": null }, { "id": "698799", "postDate": "12/19/2019 17:48:17", "content": "@bibek good to see you in this competition..\nhave you joined in ?
", "rawMarkdown": "bibek good to see you in this competition..\nhave you joined in ?", "votes": null }, { "id": "699187", "postDate": "12/20/2019 06:12:10", "content": "You can't really overfit a metric that you don't know. So PB really is our best bet, as @its7171 said
", "rawMarkdown": "You can't really overfit a metric that you don't know. So PB really is our best bet, as @its7171 said", "votes": null } ], "comments": [ { "id": 698157, "author_name": "chroteus", "author_url": "", "post_date": "12/18/2019 21:39:52", "content": "Do you mean the rank 1, bestfitting?\nNo, I don't think its overfitting, but rather a combination of good model architecture + a good way of encoding train data.
\n\nWhat is that way, I'm still trying to figure out :)
", "votes": null, "replies": [] }, { "id": 698240, "author_name": "xiejialun", "author_url": "", "post_date": "12/19/2019 01:30:45", "content": "I don't think the top teams are overfitting either. With good model architecture, I think their score would be quite reasonable.
", "votes": null, "replies": [] }, { "id": 698269, "author_name": "its7171", "author_url": "", "post_date": "12/19/2019 02:05:53", "content": "Since we don't have enough information about metrics, we can not make reliable local validation and LB is the best compass (even Public LB has 200 images) for now.
\n\nSo I think we can't avoid overfitting without more metrics details. :(
", "votes": null, "replies": [ { "id": 698271, "author_name": "xiejialun", "author_url": "", "post_date": "12/19/2019 02:11:05", "content": "Yes, I also hope the organizer can clear some information here.. Or this is the part of competition? lol
", "votes": null, "replies": [] } ] }, { "id": 698799, "author_name": "jaideepvalani", "author_url": "", "post_date": "12/19/2019 17:48:17", "content": "@bibek good to see you in this competition..\nhave you joined in ?
", "votes": null, "replies": [] }, { "id": 699187, "author_name": "ilu000", "author_url": "", "post_date": "12/20/2019 06:12:10", "content": "You can't really overfit a metric that you don't know. So PB really is our best bet, as @its7171 said
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "698057": "This is what I would think when the straight-gold kaggler(guess who?) is not in the gold region for a while; this either means overfitting or incoming shakeup? Let me know your thoughts?", "698157": "Do you mean the rank 1, bestfitting?\nNo, I don't think its overfitting, but rather a combination of good model architecture + a good way of encoding train data.\n\nWhat is that way, I'm still trying to figure out :)", "698240": "I don't think the top teams are overfitting either. With good model architecture, I think their score would be quite reasonable.", "698269": "Since we don't have enough information about metrics, we can not make reliable local validation and LB is the best compass (even Public LB has 200 images) for now.\n\nSo I think we can't avoid overfitting without more metrics details. :(", "698271": "Yes, I also hope the organizer can clear some information here.. Or this is the part of competition? lol", "698799": "bibek good to see you in this competition..\nhave you joined in ?", "699187": "You can't really overfit a metric that you don't know. So PB really is our best bet, as @its7171 said" }, "source": "meta" }