Hey there!
\nJust like everybody else, I was scared and unsure if I’d make it through the shake-up — but turns out, I did.
\nTo those who survived — congratulations.
\nAnd to those who didn’t — trust me, I’ve been on both sides, and it’s never easy. It hurts — a lot. No sugarcoating that. But one thing’s for sure: we do learn from these moments, and that learning is a win in itself.
I genuinely hope whatever this competition taught us — the highs, the lows, the mistakes — we carry those lessons into the next one.
\nHere’s to all of us — we made it through, in one way or another.
\nCheers! :))
Hey there!
\nJust like everybody else, I was scared and unsure if I’d make it through the shake-up — but turns out, I did.
\nTo those who survived — congratulations.
\nAnd to those who didn’t — trust me, I’ve been on both sides, and it’s never easy. It hurts — a lot. No sugarcoating that. But one thing’s for sure: we do learn from these moments, and that learning is a win in itself.
I genuinely hope whatever this competition taught us — the highs, the lows, the mistakes — we carry those lessons into the next one.
\nHere’s to all of us — we made it through, in one way or another.
\nCheers! :))
These words helped me more than I expected. Really needed this today. Thank you.
", "rawMarkdown": "These words helped me more than I expected. Really needed this today. Thank you.", "votes": null }, { "id": "3218364", "postDate": "06/06/2025 05:41:04", "content": "Hi just so that to ensure the above words are not just words but actual experiences I want to share my stuff.
\nI am Kaggle competition expert I won my first comp medal after some 30 odd comps wherein I jumped from 200+ places to 70th places . In the next comp after that there was some 4k users it was LLM Detect AI Generated text competition , I had a solution which could have ranked 40th out of 4k users and land me in silver zone but I ended up not selecting that submission and not gaining medal . I faced one more shakeup in another competition and the competition after that I got my bronze and became competition expert . Recently I have faced shakeups in 3-4 comps including this one .
\nBut in this comp I got to learn about the audio data and deep learning models and ensembling and optimization.
\nI also felt bad when the leaderboard was announced and I realized fell from 70+ places to 599 places , but after a good nap I am able to get back things in perspective .
You got this! Buckle up! Up and onwards to the next competition! :))
", "rawMarkdown": "You got this! Buckle up! Up and onwards to the next competition! :))", "votes": null }, { "id": "3218381", "postDate": "06/06/2025 05:53:28", "content": "This happens @sayedathar11 - here are some inherent issues with the current practice that could be reconsidered for one to progress and achieve within reasonable time!
\nGetting up and down on the leaderboard is a journey of self improvement and learning. There are some issues with the current system and I hope Kaggle takes cognizance of these-
\nThanks for sharing your journey! @sayedathar11 :))
\nNaps are always helpful! XP
@ravi20076 One of the thing that I feel can improve the system as you said is
\nHaving separate LB for GMs particularly the ones with 10s of competition gold medals , similar to what Codeforces does by separating users into Div1 and Div2
As you said some rewards and recognition should be given to people performing well on public LB as well like some Top 50-100 users
Yes it always helps me and clear my mind😄
", "rawMarkdown": "Yes it always helps me and clear my mind😄", "votes": null }, { "id": "3218398", "postDate": "06/06/2025 06:13:07", "content": "This makes sense as the public LB is also a component of the test set - why not evaluate the model on the entire test set including public + private LB and then rank? This will be fair to all - and may, to an extent, reduce the shakeup also @ahsuna123 @sayedathar11
\nAlso, one may try out flash competitions also. 3-months seems to be too long and unnecessary for a lot of competitions. One may design a different set of competition tiers that could offer medals with 2-3 weeks of effort, perhaps named as sprints - this could reduce time invested and focus on directed efforts rather than a full-blown 3-month exercise followed by such events!
", "rawMarkdown": "This makes sense as the public LB is also a component of the test set - why not evaluate the model on the entire test set including public + private LB and then rank? This will be fair to all - and may, to an extent, reduce the shakeup also @ahsuna123 @sayedathar11 \n\nAlso, one may try out flash competitions also. 3-months seems to be too long and unnecessary for a lot of competitions. One may design a different set of competition tiers that could offer medals with 2-3 weeks of effort, perhaps named as **sprints** - this could reduce time invested and focus on directed efforts rather than a full-blown 3-month exercise followed by such events!", "votes": null }, { "id": "3218488", "postDate": "06/06/2025 08:39:00", "content": "Most impressive jumps:
\nIt's not uncommon to see shakeups — I'm sure they both must have had more stable models that could generalize well on unseen datasets. Also, if you look at @tomohiroh profile. ToMo has consistently performed well and grabbed medals (specifically past BirdClef competitions) - as I said in the post as well, learnings - they count! I am sure they've applied the prior learnings in this competition. Big congratulations to them! :))🥳
", "rawMarkdown": "It's not uncommon to see shakeups — I'm sure they both must have had more stable models that could generalize well on unseen datasets. Also, if you look at @tomohiroh profile. ToMo has consistently performed well and grabbed medals (specifically past BirdClef competitions) - as I said in the post as well, learnings - they count! I am sure they've applied the prior learnings in this competition. Big congratulations to them! :))🥳", "votes": null } ], "comments": [ { "id": 3218338, "author_name": "maxipulus", "author_url": "", "post_date": "06/06/2025 05:06:20", "content": "These words helped me more than I expected. Really needed this today. Thank you.
", "votes": null, "replies": [ { "id": 3218380, "author_name": "ahsuna123", "author_url": "", "post_date": "06/06/2025 05:53:12", "content": "You got this! Buckle up! Up and onwards to the next competition! :))
", "votes": null, "replies": [] } ] }, { "id": 3218364, "author_name": "sayedathar11", "author_url": "", "post_date": "06/06/2025 05:41:04", "content": "Hi just so that to ensure the above words are not just words but actual experiences I want to share my stuff.
\nI am Kaggle competition expert I won my first comp medal after some 30 odd comps wherein I jumped from 200+ places to 70th places . In the next comp after that there was some 4k users it was LLM Detect AI Generated text competition , I had a solution which could have ranked 40th out of 4k users and land me in silver zone but I ended up not selecting that submission and not gaining medal . I faced one more shakeup in another competition and the competition after that I got my bronze and became competition expert . Recently I have faced shakeups in 3-4 comps including this one .
\nBut in this comp I got to learn about the audio data and deep learning models and ensembling and optimization.
\nI also felt bad when the leaderboard was announced and I realized fell from 70+ places to 599 places , but after a good nap I am able to get back things in perspective .
This happens @sayedathar11 - here are some inherent issues with the current practice that could be reconsidered for one to progress and achieve within reasonable time!
\nGetting up and down on the leaderboard is a journey of self improvement and learning. There are some issues with the current system and I hope Kaggle takes cognizance of these-
\n@ravi20076 One of the thing that I feel can improve the system as you said is
\nHaving separate LB for GMs particularly the ones with 10s of competition gold medals , similar to what Codeforces does by separating users into Div1 and Div2
As you said some rewards and recognition should be given to people performing well on public LB as well like some Top 50-100 users
This makes sense as the public LB is also a component of the test set - why not evaluate the model on the entire test set including public + private LB and then rank? This will be fair to all - and may, to an extent, reduce the shakeup also @ahsuna123 @sayedathar11
\nAlso, one may try out flash competitions also. 3-months seems to be too long and unnecessary for a lot of competitions. One may design a different set of competition tiers that could offer medals with 2-3 weeks of effort, perhaps named as sprints - this could reduce time invested and focus on directed efforts rather than a full-blown 3-month exercise followed by such events!
", "votes": null, "replies": [] } ] } ] }, { "id": 3218382, "author_name": "ahsuna123", "author_url": "", "post_date": "06/06/2025 05:54:28", "content": "Thanks for sharing your journey! @sayedathar11 :))
\nNaps are always helpful! XP
Yes it always helps me and clear my mind😄
", "votes": null, "replies": [] } ] } ] }, { "id": 3218488, "author_name": "sapr3s", "author_url": "", "post_date": "06/06/2025 08:39:00", "content": "Most impressive jumps:
\nIt's not uncommon to see shakeups — I'm sure they both must have had more stable models that could generalize well on unseen datasets. Also, if you look at @tomohiroh profile. ToMo has consistently performed well and grabbed medals (specifically past BirdClef competitions) - as I said in the post as well, learnings - they count! I am sure they've applied the prior learnings in this competition. Big congratulations to them! :))🥳
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3218289": "Hey there!\n\nJust like everybody else, I was scared and unsure if I’d make it through the shake-up — but turns out, I did.\n\nTo those who survived — congratulations.\nAnd to those who didn’t — trust me, I’ve been on both sides, and it’s never easy. It hurts — a lot. No sugarcoating that. But one thing’s for sure: we do learn from these moments, and that learning is a win in itself.\n\nI genuinely hope whatever this competition taught us — the highs, the lows, the mistakes — we carry those lessons into the next one.\n\nHere’s to all of us — we made it through, in one way or another.\nCheers! :))", "3218338": "These words helped me more than I expected. Really needed this today. Thank you.", "3218364": "Hi just so that to ensure the above words are not just words but actual experiences I want to share my stuff.\nI am Kaggle competition expert I won my first comp medal after some 30 odd comps wherein I jumped from 200+ places to 70th places . In the next comp after that there was some 4k users it was LLM Detect AI Generated text competition , I had a solution which could have ranked 40th out of 4k users and land me in silver zone but I ended up not selecting that submission and not gaining medal . I faced one more shakeup in another competition and the competition after that I got my bronze and became competition expert . Recently I have faced shakeups in 3-4 comps including this one .\nBut in this comp I got to learn about the audio data and deep learning models and ensembling and optimization.\nI also felt bad when the leaderboard was announced and I realized fell from 70+ places to 599 places , but after a good nap I am able to get back things in perspective .", "3218380": "You got this! Buckle up! Up and onwards to the next competition! :))", "3218381": "This happens @sayedathar11 - here are some inherent issues with the current practice that could be reconsidered for one to progress and achieve within reasonable time!\n\nGetting up and down on the leaderboard is a journey of self improvement and learning. There are some issues with the current system and I hope Kaggle takes cognizance of these- \n\n- Waiting for 3 months and working hard on a dataset and coming out of it with no medal due to a shakeup needs to be considered and possibly changed. One should have some consolation for people doing well on the public LB also\n- Do we really need 5 gold medals in competitions to become a Grandmaster? Why not have a combination of golds + silver / participation based attribution? Grandmaster status is not a degree but just a Kaggle level designation. \n- The idea of ranking participants based on the metric alone could be reconsidered.\n- Veteran GM participants could be placed on a separate LB outside the ranking system and more slots could be opened for aspirants to secure gold medals for their progression\n- Solo gold medal condition should certainly be reconsidered. What is the motivation of working solo and in a silo when the said individual will never do this in the industry? Data science is about collaboration and is never accomplished in silos, so why have this condition?", "3218382": "Thanks for sharing your journey! @sayedathar11 :))\nNaps are always helpful! XP", "3218389": "ravi20076 One of the thing that I feel can improve the system as you said is \n1. Having separate LB for GMs particularly the ones with 10s of competition gold medals , similar to what Codeforces does by separating users into Div1 and Div2 \n\n2. As you said some rewards and recognition should be given to people performing well on public LB as well like some Top 50-100 users", "3218390": "Yes it always helps me and clear my mind😄", "3218398": "This makes sense as the public LB is also a component of the test set - why not evaluate the model on the entire test set including public + private LB and then rank? This will be fair to all - and may, to an extent, reduce the shakeup also @ahsuna123 @sayedathar11 \n\nAlso, one may try out flash competitions also. 3-months seems to be too long and unnecessary for a lot of competitions. One may design a different set of competition tiers that could offer medals with 2-3 weeks of effort, perhaps named as **sprints** - this could reduce time invested and focus on directed efforts rather than a full-blown 3-month exercise followed by such events!", "3218488": "Most impressive jumps:\n- ToMo +467 -> 46\n- Danjiri +611 -> 62\nInteresting — I wonder why?", "3218521": "It's not uncommon to see shakeups — I'm sure they both must have had more stable models that could generalize well on unseen datasets. Also, if you look at @tomohiroh profile. ToMo has consistently performed well and grabbed medals (specifically past BirdClef competitions) - as I said in the post as well, learnings - they count! I am sure they've applied the prior learnings in this competition. Big congratulations to them! :))🥳" }, "source": "meta" }