![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8273630%2F11ceb3623fdd4dd389da5c1dc9f7ea38%2FLB.png?generation=1718083574863096&alt=media\")
Hello all,
\nI presume many of us had expected a moderate/ high churn in this competition and we indeed witnessed it too! Kindly find below a plot of the overall churn and its inferences-
\n
| Region Label | \nComments | \n
|---|---|
| R1 | \n* Indicates the medal winners staying stable on both the leaderboards- we do witness a high level of local churn (churn within the region), we shall explore this shortly * Considering the extent of the overall churn, staying in the medal zone across the public and private LB is a success in my humble opinion | \n
| R2 | \n* Indicates the set of participants that fell off a medal position in the public leaderboard and reached a modest position in the private leaderboard * This is certainly due to overfitting/ code bugs/ incongruent CV-LB submissions and sometimes, due to sub-optimal submission selection for the private leaderboard. | \n
| R3 | \n* Hearty congratulations for the uplift in the private LB! * These participants went upward in the private LB and beat the churn * Perhaps making a good choice for the final submissions and staying put with the CV score despite the public LB performance must have helped these participants | \n
| R4 | \n* This is surely off a public notebook and indicates a sticky pattern that overlaps a bit with R3 and extends into the non-medal zone * The best public kernel could grant one a silver medal too! * Congratulations to the early submission candidates for the triumph and best regards | \n
| R5 | \n* This region is quite similar to R4, but I am unsure if they too stem from a public kernel * These regions also indicate a sticky cluster away from the medal zones though | \n
The table below tries to analyze the level of churn and attrition across the 3-medal zones
\n| Region Label | \nLast Rank | \nComments | \nStability Ratio | \n
|---|---|---|---|
| Gold medal | \n11 | \n* Stable teams = 7 * Teams entering from below = 4 | \n7 / 11 = 63.63% | \n
| Silver medal | \n50 | \n* Stable teams = 8 * Teams entering from below = 29 * Teams entering from above = 2 | \n8 / 39 = 20.16% | \n
| Bronze medal | \n100 | \n* Stable teams = 0 * Teams entering from below = 50 * Teams entering from above = 0 | \n0/ 50 = 0.00% | \n
The lower regions of silver and the upper regions of bronze medals are entirely occupied by the best public kernel, explaining the churn! This is indeed incredible, as 1 kernel could provide a high rank and medals to more than 50% of the medal winners in the competition!
The submission time was perhaps key in the acquisition of medals in this case, participants submitting earlier than others were awarded a higher position and medals compared to the rest.
\nAs a remark, I may opine that all participants from ranks 32 - 226 have the same score of 0.649998! This is indeed incredible!
\nI extend sincere wishes to the winners of the competition and the forum contributors. Considering the success of public materials in the competition, I extend a sincere thanksgiving to the discussion and kernel contributors who selflessly shared their ideas for others' benefit.
\nI hope the lessons learnt here could be of use in one's real life assignment. Using one's Kaggle knowledge to dispose practical applications outside of Kaggle is one's real life tangible gold medal that could be used for career growth for oneself.
Best of luck for the future and happy learning!
\nRegards,
\nRavi Ramakrishnan
Hello all,
\nI presume many of us had expected a moderate/ high churn in this competition and we indeed witnessed it too! Kindly find below a plot of the overall churn and its inferences-
\n
| Region Label | \nComments | \n
|---|---|
| R1 | \n* Indicates the medal winners staying stable on both the leaderboards- we do witness a high level of local churn (churn within the region), we shall explore this shortly * Considering the extent of the overall churn, staying in the medal zone across the public and private LB is a success in my humble opinion | \n
| R2 | \n* Indicates the set of participants that fell off a medal position in the public leaderboard and reached a modest position in the private leaderboard * This is certainly due to overfitting/ code bugs/ incongruent CV-LB submissions and sometimes, due to sub-optimal submission selection for the private leaderboard. | \n
| R3 | \n* Hearty congratulations for the uplift in the private LB! * These participants went upward in the private LB and beat the churn * Perhaps making a good choice for the final submissions and staying put with the CV score despite the public LB performance must have helped these participants | \n
| R4 | \n* This is surely off a public notebook and indicates a sticky pattern that overlaps a bit with R3 and extends into the non-medal zone * The best public kernel could grant one a silver medal too! * Congratulations to the early submission candidates for the triumph and best regards | \n
| R5 | \n* This region is quite similar to R4, but I am unsure if they too stem from a public kernel * These regions also indicate a sticky cluster away from the medal zones though | \n
The table below tries to analyze the level of churn and attrition across the 3-medal zones
\n| Region Label | \nLast Rank | \nComments | \nStability Ratio | \n
|---|---|---|---|
| Gold medal | \n11 | \n* Stable teams = 7 * Teams entering from below = 4 | \n7 / 11 = 63.63% | \n
| Silver medal | \n50 | \n* Stable teams = 8 * Teams entering from below = 29 * Teams entering from above = 2 | \n8 / 39 = 20.16% | \n
| Bronze medal | \n100 | \n* Stable teams = 0 * Teams entering from below = 50 * Teams entering from above = 0 | \n0/ 50 = 0.00% | \n
The lower regions of silver and the upper regions of bronze medals are entirely occupied by the best public kernel, explaining the churn! This is indeed incredible, as 1 kernel could provide a high rank and medals to more than 50% of the medal winners in the competition!
The submission time was perhaps key in the acquisition of medals in this case, participants submitting earlier than others were awarded a higher position and medals compared to the rest.
\nAs a remark, I may opine that all participants from ranks 32 - 226 have the same score of 0.649998! This is indeed incredible!
\nI extend sincere wishes to the winners of the competition and the forum contributors. Considering the success of public materials in the competition, I extend a sincere thanksgiving to the discussion and kernel contributors who selflessly shared their ideas for others' benefit.
\nI hope the lessons learnt here could be of use in one's real life assignment. Using one's Kaggle knowledge to dispose practical applications outside of Kaggle is one's real life tangible gold medal that could be used for career growth for oneself.
Best of luck for the future and happy learning!
\nRegards,
\nRavi Ramakrishnan
Thanks for sharing a nice plot. The graph you have shared indeed makes this competition very interesting.
", "rawMarkdown": "Thanks for sharing a nice plot. The graph you have shared indeed makes this competition very interesting.", "votes": null }, { "id": "2868274", "postDate": "06/12/2024 09:30:33", "content": "Group R2 is not that far from R1 as the graph suggest. R2 were just unlucky to be worse than the R4 notebook. Even change by 0.00001 of score could make ranking shift of ~200 places
", "rawMarkdown": "Group R2 is not that far from R1 as the graph suggest. R2 were just unlucky to be worse than the R4 notebook. Even change by 0.00001 of score could make ranking shift of ~200 places", "votes": null }, { "id": "2868627", "postDate": "06/12/2024 14:18:25", "content": "Congratulations for your silver medal @crsuthikshnkumar
", "rawMarkdown": "Congratulations for your silver medal @crsuthikshnkumar", "votes": null }, { "id": "2868629", "postDate": "06/12/2024 14:19:07", "content": "Very true, a lot of otherwise good solutions got churned downward in the R2 region @dankal
", "rawMarkdown": "Very true, a lot of otherwise good solutions got churned downward in the R2 region @dankal", "votes": null } ], "comments": [ { "id": 2867973, "author_name": "crsuthikshnkumar", "author_url": "", "post_date": "06/12/2024 06:32:18", "content": "Thanks for sharing a nice plot. The graph you have shared indeed makes this competition very interesting.
", "votes": null, "replies": [ { "id": 2868627, "author_name": "ravi20076", "author_url": "", "post_date": "06/12/2024 14:18:25", "content": "Congratulations for your silver medal @crsuthikshnkumar
", "votes": null, "replies": [] } ] }, { "id": 2868274, "author_name": "dankal", "author_url": "", "post_date": "06/12/2024 09:30:33", "content": "Group R2 is not that far from R1 as the graph suggest. R2 were just unlucky to be worse than the R4 notebook. Even change by 0.00001 of score could make ranking shift of ~200 places
", "votes": null, "replies": [ { "id": 2868629, "author_name": "ravi20076", "author_url": "", "post_date": "06/12/2024 14:19:07", "content": "Very true, a lot of otherwise good solutions got churned downward in the R2 region @dankal
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2866035": "Hello all,\n\nI presume many of us had expected a moderate/ high churn in this competition and we indeed witnessed it too! Kindly find below a plot of the overall churn and its inferences-\n\n\n\n### Inferences\n\n| Region Label | Comments |\n| --- | --- |\n| R1 | * Indicates the medal winners staying stable on both the leaderboards- we do witness a high level of local churn (churn within the region), we shall explore this shortly