Apart from the weird evaluation metric, the Public LB targets are publicly available and are provided as part of the competition dataset. This kinda makes the LB useless right ? This is the first time I have seen more than 100 people have perfect scores. What is the reason for this ?
\nI'd love to get some other perspectives here.
", "messages": [ { "id": "1622833", "postDate": "12/19/2021 07:35:01", "content": "Apart from the weird evaluation metric, the Public LB targets are publicly available and are provided as part of the competition dataset. This kinda makes the LB useless right ? This is the first time I have seen more than 100 people have perfect scores. What is the reason for this ?
\nI'd love to get some other perspectives here.
", "rawMarkdown": "Apart from the weird evaluation metric, the Public LB targets are publicly available and are provided as part of the competition dataset. This kinda makes the LB useless right ? This is the first time I have seen more than 100 people have perfect scores. What is the reason for this ?\n\nI'd love to get some other perspectives here.", "votes": null }, { "id": "1622986", "postDate": "12/19/2021 11:13:18", "content": "@skaarface I'd believe you will have read through the Competition overview once again.
\nYes, the LB is \"useless\" as it's the data which is present in the training set. But we can use it the right way by not training our model on this portion of the dataset to assess our model's score.
", "rawMarkdown": "skaarface I'd believe you will have read through the Competition overview once again. \n\nYes, the LB is \"useless\" as it's the data which is present in the training set. But we can use it the right way by not training our model on this portion of the dataset to assess our model's score.", "votes": null }, { "id": "1623401", "postDate": "12/19/2021 20:25:47", "content": "Hello @skaarface, there is already plenty of explanations of the discussion forums; take a look; there are also some guidances to avoid using the information after a specific time to have a more meaningful score.
", "rawMarkdown": "Hello @skaarface, there is already plenty of explanations of the discussion forums; take a look; there are also some guidances to avoid using the information after a specific time to have a more meaningful score.", "votes": null }, { "id": "1625094", "postDate": "12/21/2021 14:13:08", "content": "Oh okay, so the leaderboard top scores are purely overfitting ones then right ?
", "rawMarkdown": "Oh okay, so the leaderboard top scores are purely overfitting ones then right ?", "votes": null }, { "id": "1628019", "postDate": "12/24/2021 12:34:21", "content": "I think so. A bit of overfitting and a lot of looking into the future :)
", "rawMarkdown": "I think so. A bit of overfitting and a lot of looking into the future :)", "votes": null }, { "id": "1628431", "postDate": "12/24/2021 21:19:36", "content": "Thanks. I appreciate the help :)
", "rawMarkdown": "Thanks. I appreciate the help :)", "votes": null } ], "comments": [ { "id": 1622986, "author_name": "nitishmalagi", "author_url": "", "post_date": "12/19/2021 11:13:18", "content": "@skaarface I'd believe you will have read through the Competition overview once again.
\nYes, the LB is \"useless\" as it's the data which is present in the training set. But we can use it the right way by not training our model on this portion of the dataset to assess our model's score.
", "votes": null, "replies": [ { "id": 1625094, "author_name": "skaarface", "author_url": "", "post_date": "12/21/2021 14:13:08", "content": "Oh okay, so the leaderboard top scores are purely overfitting ones then right ?
", "votes": null, "replies": [] }, { "id": 1628019, "author_name": "rugert", "author_url": "", "post_date": "12/24/2021 12:34:21", "content": "I think so. A bit of overfitting and a lot of looking into the future :)
", "votes": null, "replies": [] }, { "id": 1628431, "author_name": "skaarface", "author_url": "", "post_date": "12/24/2021 21:19:36", "content": "Thanks. I appreciate the help :)
", "votes": null, "replies": [] } ] }, { "id": 1623401, "author_name": "cv13j0", "author_url": "", "post_date": "12/19/2021 20:25:47", "content": "Hello @skaarface, there is already plenty of explanations of the discussion forums; take a look; there are also some guidances to avoid using the information after a specific time to have a more meaningful score.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1622833": "Apart from the weird evaluation metric, the Public LB targets are publicly available and are provided as part of the competition dataset. This kinda makes the LB useless right ? This is the first time I have seen more than 100 people have perfect scores. What is the reason for this ?\n\nI'd love to get some other perspectives here.", "1622986": "skaarface I'd believe you will have read through the Competition overview once again. \n\nYes, the LB is \"useless\" as it's the data which is present in the training set. But we can use it the right way by not training our model on this portion of the dataset to assess our model's score.", "1623401": "Hello @skaarface, there is already plenty of explanations of the discussion forums; take a look; there are also some guidances to avoid using the information after a specific time to have a more meaningful score.", "1625094": "Oh okay, so the leaderboard top scores are purely overfitting ones then right ?", "1628019": "I think so. A bit of overfitting and a lot of looking into the future :)", "1628431": "Thanks. I appreciate the help :)" }, "source": "meta" }