When I woke up and looked at Leaderboard. I found that a lot of people got 96.19 with a few submissions.\nsomebody knows what happened?
", "messages": [ { "id": "961142", "postDate": "08/07/2020 00:11:18", "content": "When I woke up and looked at Leaderboard. I found that a lot of people got 96.19 with a few submissions.\nsomebody knows what happened?
", "rawMarkdown": "When I woke up and looked at Leaderboard. I found that a lot of people got 96.19 with a few submissions.\nsomebody knows what happened?", "votes": null }, { "id": "961144", "postDate": "08/07/2020 00:12:06", "content": "This happened: https://www.kaggle.com/paklau9/minmax-highest-public-lb-9619
", "rawMarkdown": "This happened: https://www.kaggle.com/paklau9/minmax-highest-public-lb-9619", "votes": null }, { "id": "961146", "postDate": "08/07/2020 00:18:31", "content": "Oh I See thank you!
", "rawMarkdown": "Oh I See thank you!", "votes": null }, { "id": "961161", "postDate": "08/07/2020 00:48:13", "content": "Kaggle as usual :)
", "rawMarkdown": "Kaggle as usual :)", "votes": null }, { "id": "961192", "postDate": "08/07/2020 01:42:48", "content": "😄 I learned a lot from it!
", "rawMarkdown": "😄 I learned a lot from it!", "votes": null }, { "id": "961259", "postDate": "08/07/2020 03:47:37", "content": "Yep, public LB went crazy, where will we stay at the end? At 0.97? And huge shakeup after 😈
", "rawMarkdown": "Yep, public LB went crazy, where will we stay at the end? At 0.97? And huge shakeup after 😈", "votes": null }, { "id": "961469", "postDate": "08/07/2020 07:50:22", "content": "As a beginner I find these activites so discouraging. I am literally trying months to get to 0.91+ and someone just submitted some notebooks submission.csv and reach to 96.19.
", "rawMarkdown": "As a beginner I find these activites so discouraging. I am literally trying months to get to 0.91+ and someone just submitted some notebooks submission.csv and reach to 96.19.", "votes": null }, { "id": "961672", "postDate": "08/07/2020 11:38:53", "content": "Don't be discouraged. You can always blend your stuff with public notebooks. If you did something valuable, even if your public LB score is lower than the public notebook, then blending your work with it will give you an edge over those who only used the public notebook.
\nThere is also the very probable possibility that the public notebook overfits to public test data. In that case all those who used it will sink in final ranking. A great example of this was Porto Seguro competition. A top 20 public notebook sent over 1000 participants below the 600th rank.
\nLast, and this is new, it maybe that the public notebook does not meet the licensing requirements set by this competition rules. Indeed, we must be able to reproduce our submission using code we license with MIT license. Tracking all the piece of code that led to the best public score notebook is going to be a tedious task.
", "rawMarkdown": "Don't be discouraged. You can always blend your stuff with public notebooks. If you did something valuable, even if your public LB score is lower than the public notebook, then blending your work with it will give you an edge over those who only used the public notebook.\n\nThere is also the very probable possibility that the public notebook overfits to public test data. In that case all those who used it will sink in final ranking. A great example of this was Porto Seguro competition. A top 20 public notebook sent over 1000 participants below the 600th rank.\n\nLast, and this is new, it maybe that the public notebook does not meet the licensing requirements set by this competition rules. Indeed, we must be able to reproduce our submission using code we license with MIT license. Tracking all the piece of code that led to the best public score notebook is going to be a tedious task.", "votes": null }, { "id": "961680", "postDate": "08/07/2020 11:56:11", "content": "This topic appears in every popular Kaggle competition.
\nIf your results are worse than public results it means that you should not receive medal for them. Medals should not be awarded for time spent on competion but for results.
As beginner i'm impressive for the times that i've see this topic. Why matters so much? i'm not an expert but a see a lot of overfitting in high score blendings. I bet that the LB it's gonna change a lot in the private leaderboard for all the people who is just overfitting. Relax if your work it's good at the end you are gonna see it.
", "rawMarkdown": "As beginner i'm impressive for the times that i've see this topic. Why matters so much? i'm not an expert but a see a lot of overfitting in high score blendings. I bet that the LB it's gonna change a lot in the **private leaderboard** for all the people who is just overfitting. Relax if your work it's good at the end you are gonna see it.", "votes": null }, { "id": "961910", "postDate": "08/07/2020 15:48:37", "content": "If you look at the current best blender you will see some hyperparameters for a very simple min-max ensemble method. How do you think these were chosen ? By checking which one gave best LB result.
\nDoesn't sound like good science to me :D
", "rawMarkdown": "If you look at the current best blender you will see some hyperparameters for a very simple min-max ensemble method. How do you think these were chosen ? By checking which one gave best LB result. \n\nDoesn't sound like good science to me :D", "votes": null }, { "id": "961979", "postDate": "08/07/2020 17:19:54", "content": "Just dont worry about it , trust your LB , I have been victim of high LB , and difference of rank was of 500 between public and private LB.
\nSo just trust your model CV .
This happened: https://www.kaggle.com/paklau9/minmax-highest-public-lb-9619
", "votes": null, "replies": [ { "id": 961146, "author_name": "deepkim", "author_url": "", "post_date": "08/07/2020 00:18:31", "content": "Oh I See thank you!
", "votes": null, "replies": [] }, { "id": 961161, "author_name": "jacekpoplawski", "author_url": "", "post_date": "08/07/2020 00:48:13", "content": "Kaggle as usual :)
", "votes": null, "replies": [] }, { "id": 961192, "author_name": "deepkim", "author_url": "", "post_date": "08/07/2020 01:42:48", "content": "😄 I learned a lot from it!
", "votes": null, "replies": [] } ] }, { "id": 961259, "author_name": "aybatov", "author_url": "", "post_date": "08/07/2020 03:47:37", "content": "Yep, public LB went crazy, where will we stay at the end? At 0.97? And huge shakeup after 😈
", "votes": null, "replies": [] }, { "id": 961788, "author_name": "hiramcho", "author_url": "", "post_date": "08/07/2020 13:58:19", "content": "As beginner i'm impressive for the times that i've see this topic. Why matters so much? i'm not an expert but a see a lot of overfitting in high score blendings. I bet that the LB it's gonna change a lot in the private leaderboard for all the people who is just overfitting. Relax if your work it's good at the end you are gonna see it.
", "votes": null, "replies": [] }, { "id": 961979, "author_name": "rajnishe", "author_url": "", "post_date": "08/07/2020 17:19:54", "content": "Just dont worry about it , trust your LB , I have been victim of high LB , and difference of rank was of 500 between public and private LB.
\nSo just trust your model CV .
As a beginner I find these activites so discouraging. I am literally trying months to get to 0.91+ and someone just submitted some notebooks submission.csv and reach to 96.19.
", "votes": null, "replies": [ { "id": 961672, "author_name": "cpmpml", "author_url": "", "post_date": "08/07/2020 11:38:53", "content": "Don't be discouraged. You can always blend your stuff with public notebooks. If you did something valuable, even if your public LB score is lower than the public notebook, then blending your work with it will give you an edge over those who only used the public notebook.
\nThere is also the very probable possibility that the public notebook overfits to public test data. In that case all those who used it will sink in final ranking. A great example of this was Porto Seguro competition. A top 20 public notebook sent over 1000 participants below the 600th rank.
\nLast, and this is new, it maybe that the public notebook does not meet the licensing requirements set by this competition rules. Indeed, we must be able to reproduce our submission using code we license with MIT license. Tracking all the piece of code that led to the best public score notebook is going to be a tedious task.
", "votes": null, "replies": [] }, { "id": 961680, "author_name": "jacekpoplawski", "author_url": "", "post_date": "08/07/2020 11:56:11", "content": "This topic appears in every popular Kaggle competition.
\nIf your results are worse than public results it means that you should not receive medal for them. Medals should not be awarded for time spent on competion but for results.
If you look at the current best blender you will see some hyperparameters for a very simple min-max ensemble method. How do you think these were chosen ? By checking which one gave best LB result.
\nDoesn't sound like good science to me :D
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "961142": "When I woke up and looked at Leaderboard. I found that a lot of people got 96.19 with a few submissions.\nsomebody knows what happened?", "961144": "This happened: https://www.kaggle.com/paklau9/minmax-highest-public-lb-9619", "961146": "Oh I See thank you!", "961161": "Kaggle as usual :)", "961192": "😄 I learned a lot from it!", "961259": "Yep, public LB went crazy, where will we stay at the end? At 0.97? And huge shakeup after 😈", "961469": "As a beginner I find these activites so discouraging. I am literally trying months to get to 0.91+ and someone just submitted some notebooks submission.csv and reach to 96.19.", "961672": "Don't be discouraged. You can always blend your stuff with public notebooks. If you did something valuable, even if your public LB score is lower than the public notebook, then blending your work with it will give you an edge over those who only used the public notebook.\n\nThere is also the very probable possibility that the public notebook overfits to public test data. In that case all those who used it will sink in final ranking. A great example of this was Porto Seguro competition. A top 20 public notebook sent over 1000 participants below the 600th rank.\n\nLast, and this is new, it maybe that the public notebook does not meet the licensing requirements set by this competition rules. Indeed, we must be able to reproduce our submission using code we license with MIT license. Tracking all the piece of code that led to the best public score notebook is going to be a tedious task.", "961680": "This topic appears in every popular Kaggle competition.\nIf your results are worse than public results it means that you should not receive medal for them. Medals should not be awarded for time spent on competion but for results.", "961788": "As beginner i'm impressive for the times that i've see this topic. Why matters so much? i'm not an expert but a see a lot of overfitting in high score blendings. I bet that the LB it's gonna change a lot in the **private leaderboard** for all the people who is just overfitting. Relax if your work it's good at the end you are gonna see it.", "961910": "If you look at the current best blender you will see some hyperparameters for a very simple min-max ensemble method. How do you think these were chosen ? By checking which one gave best LB result. \n\nDoesn't sound like good science to me :D", "961979": "Just dont worry about it , trust your LB , I have been victim of high LB , and difference of rank was of 500 between public and private LB.\nSo just trust your model CV ." }, "source": "meta" }