I have no clever solutions to share, but I'm thrilled to score my first silver medal in a Kaggle competition, so I'll briefly share my approach, especially because it might be helpful for others who are relatively new to Kaggle (or non-playground competitions). I couldn't devote much time to this competition - in the end, I only had 24 submissions, of which 5 threw errors during submission (mostly related to installation problems), so only 19 valid submissions. Whatever experiments I performed were only possible due to the generosity of fellow Kagglers who shared their solutions via public notebooks. I would like to thank them all, but especially
\n@cdeotte, @yekenot, @eikyou, @albansteff, @kendontcare11, @andreasbis and @omidbaghchehsaraei
\nSince I had some experience with tabular competitions due to being a regular competitor in the Tabular Playground Series on Kaggle for a year, I knew that GBDTs and NNs were likely to combine well, and I focused on building such ensembles. About 2 months ago, I'd narrowly missed my first medal due to trusting the LB over my CV, so I focused on my CV, with a goal of CV >= 0.685 and LB >= 0.69. But until the last day, I couldn't cross CV = 0.683, even though some of the submissions were in the 0.689-0.691 range on the LB. Entering the final day, I was outside the Top 1000, & sorely tempted to submit the top scoring public notebook, which I'd been avoiding because of a large LB-CV gap. Finally, when I started to play around with weights, I managed to push the CV past 0.685 for the first time - with 2 hours to go, I finally re-entered the Top 1000 (at 787), and my final two submissions were ones that finished with less than half an hour to go (CV = 0.686, LB = 0.693; CV = 0.682, LB = 0.690). I deliberately didn't choose my second-highest scoring submission on the LB, because it was a variant of the top scoring one. I also didn't select the submission with the highest CV (CV = 0.687, LB = 0.687), because the LB, while equal to the CV, was too low compared to several other submissions. In the end, the gamble paid off, and I was on the right side of the shakeup, moving up (657 spots) to 130 and earning a silver, my first medal. I'm very aware of how big a role many others (named above) played in this, but I'm still happy with my experiments, and especially final selections.
\nI'd like to end by thanking everyone whose work I built on once again, and congratulating all the winners. Looking forward to more competition, learning and fun - Happy Kaggling, everyone!
", "comments": [ { "id": 3142198, "author_name": "SCRIPTCHEF", "author_url": "", "post_date": "2025-03-06T07:36:09.677000", "content": "Hi @optimistix we meet again! I actually checked if you were competing early on but it looked as if you had stopped submitting! Congrats on your silver - with only 19 subs!
", "votes": 1, "replies": [ { "id": 3142207, "author_name": "Optimistix", "author_url": "", "post_date": "2025-03-06T07:54:28.170000", "content": "Thanks! Yes, lots of things were happening in my life, and I couldn't make time for this competition for the most part. Congratulations on your silver as well!
", "votes": 1, "replies": [] } ] }, { "id": 3142172, "author_name": "Mart Preusse", "author_url": "", "post_date": "2025-03-06T06:46:19.123000", "content": "Well done, congratulations to your first silver! I experimented this time with different folds for each model and this probably backfired a bit, as I could not get a reliable CV score. My cv score was around 0.691! Simpler ensembles with a slightly worse cv score would have done better.
", "votes": 1, "replies": [ { "id": 3142186, "author_name": "Optimistix", "author_url": "", "post_date": "2025-03-06T07:06:10.270000", "content": "Thanks so much, and congrats on your bronze as well!
", "votes": 1, "replies": [] } ] }, { "id": 3142095, "author_name": "Ken Lee", "author_url": "", "post_date": "2025-03-06T04:55:37.253000", "content": "Congratulations!!
", "votes": 1, "replies": [ { "id": 3142142, "author_name": "Optimistix", "author_url": "", "post_date": "2025-03-06T06:03:42.220000", "content": "Thanks so much, and also thanks for sharing your code!
", "votes": 0, "replies": [] } ] }, { "id": 3142021, "author_name": "klog", "author_url": "", "post_date": "2025-03-06T03:26:57.537000", "content": "Congratulations! š It looks like you're making great use of your experience from Playground, which is fantastic. Playground seems like a great way to study table competitions, so Iād love to participate more myself moving forward!
", "votes": 1, "replies": [ { "id": 3142141, "author_name": "Optimistix", "author_url": "", "post_date": "2025-03-06T06:03:04.377000", "content": "Thanks! Yes, TPS is great for familiarizing oneself with Tabular ML. It came about because Tabular competitions were becoming rare on Kaggle, but it's good to see an uptick in them of late.
", "votes": 1, "replies": [] } ] }, { "id": 3141903, "author_name": "Ravi Ramakrishnan", "author_url": "", "post_date": "2025-03-06T01:17:47.830000", "content": "All the best @optimistix and congratulations! I think simplicity was essence here and simple solutions worked well!
", "votes": 1, "replies": [ { "id": 3141913, "author_name": "Optimistix", "author_url": "", "post_date": "2025-03-06T01:23:26.473000", "content": "Thanks! That, and judicious selection of the final two submissions (though there's always some luck involved).
", "votes": 1, "replies": [] } ] }, { "id": 3141888, "author_name": "Joe", "author_url": "", "post_date": "2025-03-06T01:08:48.503000", "content": "Congratulations @optimistix, quite the accomplishment! And a fun story there at the end on the last day. I've noticed H2O appears to do well on this site, perhaps the universe is telling us something . . .
", "votes": 1, "replies": [ { "id": 3141909, "author_name": "Optimistix", "author_url": "", "post_date": "2025-03-06T01:21:22.733000", "content": "Thanks so much! Interestingly, Autogluon (AG) also did well in this competition, and in the AutoML Grand Prix last year, AG, FLAML and LightAutoML were usually the best performing AutoML packages - perhaps H2O wasn't explored enough?
", "votes": 0, "replies": [] } ] }, { "id": 3141880, "author_name": "Sheikh Muhammad Abdullah", "author_url": "", "post_date": "2025-03-06T01:02:38.520000", "content": "congratulations for First Silver @optimistix. Wishing you good luck for Future Competitions
", "votes": 1, "replies": [ { "id": 3141911, "author_name": "Optimistix", "author_url": "", "post_date": "2025-03-06T01:21:48.893000", "content": "Thanks! Congratulations to you as well.
", "votes": 1, "replies": [] } ] }, { "id": 3142126, "author_name": "Mahdi Ravaghi", "author_url": "", "post_date": "2025-03-06T05:40:57.523000", "content": "Congratulations @optimistix! It's good to see you succeed outside TPS competitions as well.
", "votes": 2, "replies": [ { "id": 3142144, "author_name": "Optimistix", "author_url": "", "post_date": "2025-03-06T06:05:39.520000", "content": "Thanks so much - Kaggle helped by organizing some tabular ML competitions! Also, good to see you - you haven't been seen in the TPS competitions of late. Hope you're learning interesting stuff, and shall be back soon.
", "votes": 2, "replies": [ { "id": 3142160, "author_name": "Mahdi Ravaghi", "author_url": "", "post_date": "2025-03-06T06:30:53.793000", "content": "I learned what I could from TPS and reached my goal of winning 1st place once, so I decided to move on and focus on competitions with real data. I might be back though when I'm in between competitions.
", "votes": 1, "replies": [] }, { "id": 3142165, "author_name": "Optimistix", "author_url": "", "post_date": "2025-03-06T06:39:19.060000", "content": "Sensible. All the best for all your endeavors outside TPS!
", "votes": 1, "replies": [] } ] } ] }, { "id": 3217268, "author_name": "PACESETTER55", "author_url": "", "post_date": "2025-06-04T18:42:20.517000", "content": "Congratulations!!! can you share your highest scoring notebook so that we can also learn
", "votes": 0, "replies": [] } ], "comment_count": 19, "messages": [ { "id": 3141876, "postDate": "2025-03-06T00:54:50.843Z", "content": "I have no clever solutions to share, but I'm thrilled to score my first silver medal in a Kaggle competition, so I'll briefly share my approach, especially because it might be helpful for others who are relatively new to Kaggle (or non-playground competitions). I couldn't devote much time to this competition - in the end, I only had 24 submissions, of which 5 threw errors during submission (mostly related to installation problems), so only 19 valid submissions. Whatever experiments I performed were only possible due to the generosity of fellow Kagglers who shared their solutions via public notebooks. I would like to thank them all, but especially
\n@cdeotte, @yekenot, @eikyou, @albansteff, @kendontcare11, @andreasbis and @omidbaghchehsaraei
\nSince I had some experience with tabular competitions due to being a regular competitor in the Tabular Playground Series on Kaggle for a year, I knew that GBDTs and NNs were likely to combine well, and I focused on building such ensembles. About 2 months ago, I'd narrowly missed my first medal due to trusting the LB over my CV, so I focused on my CV, with a goal of CV >= 0.685 and LB >= 0.69. But until the last day, I couldn't cross CV = 0.683, even though some of the submissions were in the 0.689-0.691 range on the LB. Entering the final day, I was outside the Top 1000, & sorely tempted to submit the top scoring public notebook, which I'd been avoiding because of a large LB-CV gap. Finally, when I started to play around with weights, I managed to push the CV past 0.685 for the first time - with 2 hours to go, I finally re-entered the Top 1000 (at 787), and my final two submissions were ones that finished with less than half an hour to go (CV = 0.686, LB = 0.693; CV = 0.682, LB = 0.690). I deliberately didn't choose my second-highest scoring submission on the LB, because it was a variant of the top scoring one. I also didn't select the submission with the highest CV (CV = 0.687, LB = 0.687), because the LB, while equal to the CV, was too low compared to several other submissions. In the end, the gamble paid off, and I was on the right side of the shakeup, moving up (657 spots) to 130 and earning a silver, my first medal. I'm very aware of how big a role many others (named above) played in this, but I'm still happy with my experiments, and especially final selections.
\nI'd like to end by thanking everyone whose work I built on once again, and congratulating all the winners. Looking forward to more competition, learning and fun - Happy Kaggling, everyone!
", "rawMarkdown": "I have no clever solutions to share, but I'm thrilled to score my first silver medal in a Kaggle competition, so I'll briefly share my approach, especially because it might be helpful for others who are relatively new to Kaggle (or non-playground competitions). I couldn't devote much time to this competition - in the end, I only had 24 submissions, of which 5 threw errors during submission (mostly related to installation problems), so only 19 valid submissions. Whatever experiments I performed were only possible due to the generosity of fellow Kagglers who shared their solutions via public notebooks. I would like to thank them all, but especially\n\n@cdeotte, @yekenot, @eikyou, @albansteff, @kendontcare11, @andreasbis and @omidbaghchehsaraei \n\nSince I had some experience with tabular competitions due to being a regular competitor in the Tabular Playground Series on Kaggle for a year, I knew that GBDTs and NNs were likely to combine well, and I focused on building such ensembles. About 2 months ago, I'd narrowly missed my first medal due to trusting the LB over my CV, so I focused on my CV, with a goal of CV >= 0.685 and LB >= 0.69. But until the last day, I couldn't cross CV = 0.683, even though some of the submissions were in the 0.689-0.691 range on the LB. Entering the final day, I was outside the Top 1000, & sorely tempted to submit the top scoring public notebook, which I'd been avoiding because of a large LB-CV gap. Finally, when I started to play around with weights, I managed to push the CV past 0.685 for the first time - with 2 hours to go, I finally re-entered the Top 1000 (at 787), and my final two submissions were ones that finished with less than half an hour to go (CV = 0.686, LB = 0.693; CV = 0.682, LB = 0.690). I deliberately didn't choose my second-highest scoring submission on the LB, because it was a variant of the top scoring one. I also didn't select the submission with the highest CV (CV = 0.687, LB = 0.687), because the LB, while equal to the CV, was too low compared to several other submissions. In the end, the gamble paid off, and I was on the right side of the shakeup, moving up (657 spots) to 130 and earning a silver, my first medal. I'm very aware of how big a role many others (named above) played in this, but I'm still happy with my experiments, and especially final selections.\n\nI'd like to end by thanking everyone whose work I built on once again, and congratulating all the winners. Looking forward to more competition, learning and fun - Happy Kaggling, everyone!\n\n", "votes": 21 }, { "id": 3142198, "postDate": "2025-03-06T07:36:09.677Z", "content": "Hi @optimistix we meet again! I actually checked if you were competing early on but it looked as if you had stopped submitting! Congrats on your silver - with only 19 subs!
", "rawMarkdown": "Hi @optimistix we meet again! I actually checked if you were competing early on but it looked as if you had stopped submitting! Congrats on your silver - with only 19 subs!", "votes": 1, "replies": [ { "id": 3142207, "postDate": "2025-03-06T07:54:28.170Z", "content": "Thanks! Yes, lots of things were happening in my life, and I couldn't make time for this competition for the most part. Congratulations on your silver as well!
", "rawMarkdown": "Thanks! Yes, lots of things were happening in my life, and I couldn't make time for this competition for the most part. Congratulations on your silver as well!", "votes": 1 } ] }, { "id": 3142172, "postDate": "2025-03-06T06:46:19.123Z", "content": "Well done, congratulations to your first silver! I experimented this time with different folds for each model and this probably backfired a bit, as I could not get a reliable CV score. My cv score was around 0.691! Simpler ensembles with a slightly worse cv score would have done better.
", "rawMarkdown": "Well done, congratulations to your first silver! I experimented this time with different folds for each model and this probably backfired a bit, as I could not get a reliable CV score. My cv score was around 0.691! Simpler ensembles with a slightly worse cv score would have done better.", "votes": 1, "replies": [ { "id": 3142186, "postDate": "2025-03-06T07:06:10.270Z", "content": "Thanks so much, and congrats on your bronze as well!
", "rawMarkdown": "Thanks so much, and congrats on your bronze as well!", "votes": 1 } ] }, { "id": 3142095, "postDate": "2025-03-06T04:55:37.253Z", "content": "Congratulations!!
", "rawMarkdown": "Congratulations!!", "votes": 1, "replies": [ { "id": 3142142, "postDate": "2025-03-06T06:03:42.220Z", "content": "Thanks so much, and also thanks for sharing your code!
", "rawMarkdown": "Thanks so much, and also thanks for sharing your code!" } ] }, { "id": 3142021, "postDate": "2025-03-06T03:26:57.537Z", "content": "Congratulations! š It looks like you're making great use of your experience from Playground, which is fantastic. Playground seems like a great way to study table competitions, so Iād love to participate more myself moving forward!
", "rawMarkdown": "Congratulations! š It looks like you're making great use of your experience from Playground, which is fantastic. Playground seems like a great way to study table competitions, so Iād love to participate more myself moving forward! ", "votes": 1, "replies": [ { "id": 3142141, "postDate": "2025-03-06T06:03:04.377Z", "content": "Thanks! Yes, TPS is great for familiarizing oneself with Tabular ML. It came about because Tabular competitions were becoming rare on Kaggle, but it's good to see an uptick in them of late.
", "rawMarkdown": "Thanks! Yes, TPS is great for familiarizing oneself with Tabular ML. It came about because Tabular competitions were becoming rare on Kaggle, but it's good to see an uptick in them of late.", "votes": 1 } ] }, { "id": 3141903, "postDate": "2025-03-06T01:17:47.830Z", "content": "All the best @optimistix and congratulations! I think simplicity was essence here and simple solutions worked well!
", "rawMarkdown": "All the best @optimistix and congratulations! I think simplicity was essence here and simple solutions worked well!", "votes": 1, "replies": [ { "id": 3141913, "postDate": "2025-03-06T01:23:26.473Z", "content": "Thanks! That, and judicious selection of the final two submissions (though there's always some luck involved).
", "rawMarkdown": "Thanks! That, and judicious selection of the final two submissions (though there's always some luck involved).", "votes": 1 } ] }, { "id": 3141888, "postDate": "2025-03-06T01:08:48.503Z", "content": "Congratulations @optimistix, quite the accomplishment! And a fun story there at the end on the last day. I've noticed H2O appears to do well on this site, perhaps the universe is telling us something . . .
", "rawMarkdown": "Congratulations @optimistix, quite the accomplishment! And a fun story there at the end on the last day. I've noticed H2O appears to do well on this site, perhaps the universe is telling us something . . .", "votes": 1, "replies": [ { "id": 3141909, "postDate": "2025-03-06T01:21:22.733Z", "content": "Thanks so much! Interestingly, Autogluon (AG) also did well in this competition, and in the AutoML Grand Prix last year, AG, FLAML and LightAutoML were usually the best performing AutoML packages - perhaps H2O wasn't explored enough?
", "rawMarkdown": "Thanks so much! Interestingly, Autogluon (AG) also did well in this competition, and in the AutoML Grand Prix last year, AG, FLAML and LightAutoML were usually the best performing AutoML packages - perhaps H2O wasn't explored enough?" } ] }, { "id": 3141880, "postDate": "2025-03-06T01:02:38.520Z", "content": "congratulations for First Silver @optimistix. Wishing you good luck for Future Competitions
", "rawMarkdown": "congratulations for First Silver @optimistix. Wishing you good luck for Future Competitions", "votes": 1, "replies": [ { "id": 3141911, "postDate": "2025-03-06T01:21:48.893Z", "content": "Thanks! Congratulations to you as well.
", "rawMarkdown": "Thanks! Congratulations to you as well.", "votes": 1 } ] }, { "id": 3142126, "postDate": "2025-03-06T05:40:57.523Z", "content": "Congratulations @optimistix! It's good to see you succeed outside TPS competitions as well.
", "rawMarkdown": "Congratulations @optimistix! It's good to see you succeed outside TPS competitions as well.", "votes": 2, "replies": [ { "id": 3142144, "postDate": "2025-03-06T06:05:39.520Z", "content": "Thanks so much - Kaggle helped by organizing some tabular ML competitions! Also, good to see you - you haven't been seen in the TPS competitions of late. Hope you're learning interesting stuff, and shall be back soon.
", "rawMarkdown": "Thanks so much - Kaggle helped by organizing some tabular ML competitions! Also, good to see you - you haven't been seen in the TPS competitions of late. Hope you're learning interesting stuff, and shall be back soon.", "votes": 2, "replies": [ { "id": 3142160, "postDate": "2025-03-06T06:30:53.793Z", "content": "I learned what I could from TPS and reached my goal of winning 1st place once, so I decided to move on and focus on competitions with real data. I might be back though when I'm in between competitions.
", "rawMarkdown": "I learned what I could from TPS and reached my goal of winning 1st place once, so I decided to move on and focus on competitions with real data. I might be back though when I'm in between competitions.", "votes": 1 }, { "id": 3142165, "postDate": "2025-03-06T06:39:19.060Z", "content": "Sensible. All the best for all your endeavors outside TPS!
", "rawMarkdown": "Sensible. All the best for all your endeavors outside TPS!", "votes": 1 } ] } ] }, { "id": 3217268, "postDate": "2025-06-04T18:42:20.517Z", "content": "Congratulations!!! can you share your highest scoring notebook so that we can also learn
", "rawMarkdown": "Congratulations!!! can you share your highest scoring notebook so that we can also learn\n" } ], "complete": true, "source": "live", "fetched_at": "2026-10-19T19:45:43Z", "competition_id": 70942, "competition_slug": "equity-post-HCT-survival-predictions", "team_id": 13082810, "solution_url": "/competitions/equity-post-HCT-survival-predictions/writeups/optimistix-first-silver-medal-over-the-moon" }