![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2034058%2Fa998874b78733465dec91e28fe95e804%2FbestOne.png?generation=1595473390109465&alt=media\")
didn't select our silver solution!!
\n\n
didn't select our silver solution!!
\n\n
If everyone selected their best scores you would not be silver. I also had a very high scoring private LB kernel but would have never selected it, because it was a very early sub of mine just blending local model with public kernel.
", "rawMarkdown": "If everyone selected their best scores you would not be silver. I also had a very high scoring private LB kernel but would have never selected it, because it was a very early sub of mine just blending local model with public kernel.", "votes": 16, "replies": [ { "id": 941265, "postDate": "2020-07-23T06:39:15.763Z", "content": "that's true too @philippsinger :)
", "rawMarkdown": "that's true too @philippsinger :)\n" }, { "id": 941274, "postDate": "2020-07-23T06:43:12.830Z", "content": "@philippsinger one thing i noticed that in recent kaggle competitions (specially in some of them) we can't rely on cv nor on lb much,it really is very frustrating , where is machine learning here when you can't rely much on cv,nor on lb?
", "rawMarkdown": "@philippsinger one thing i noticed that in recent kaggle competitions (specially in some of them) we can't rely on cv nor on lb much,it really is very frustrating , where is machine learning here when you can't rely much on cv,nor on lb?", "votes": 2 }, { "id": 941284, "postDate": "2020-07-23T06:47:28.193Z", "content": "I think only this competition and M5 have been like this. Other competitions had quite good ways to evaluate.
", "rawMarkdown": "I think only this competition and M5 have been like this. Other competitions had quite good ways to evaluate." }, { "id": 942559, "postDate": "2020-07-23T20:19:10.167Z", "content": "It was quite new and interesting experience for me of learning how to deal with such problem
", "rawMarkdown": "It was quite new and interesting experience for me of learning how to deal with such problem", "votes": 3 } ] }, { "id": 943535, "postDate": "2020-07-24T12:24:26.437Z", "content": "The real question to ask is how could you have selected your best submission? Learning how to do it is the most precious skill one can learn out of kagggle: reliably assess model performance based on train data. If you get good at it then you are a very valuable ML practitioner.
", "rawMarkdown": "The real question to ask is how could you have selected your best submission? Learning how to do it is the most precious skill one can learn out of kagggle: reliably assess model performance based on train data. If you get good at it then you are a very valuable ML practitioner. ", "votes": 11, "replies": [ { "id": 943621, "postDate": "2020-07-24T13:29:17.597Z", "content": "Thank you @cpmpml \nTrying to learn it better from every competitions mistake and following great discussions from people like you!\nThank you :)
", "rawMarkdown": "Thank you @cpmpml \nTrying to learn it better from every competitions mistake and following great discussions from people like you!\nThank you :)", "votes": 1 }, { "id": 943773, "postDate": "2020-07-24T15:11:26.113Z", "content": "thank you @cpmpml ! great advice. will definitely keep this in mind and keep trying to improve. Cheers!
", "rawMarkdown": "thank you @cpmpml ! great advice. will definitely keep this in mind and keep trying to improve. Cheers! ", "votes": 1 } ] }, { "id": 942573, "postDate": "2020-07-23T20:39:16.637Z", "content": "It happens... There are always a plenty of opportunities that may be missed, but it was my own chose, and if everyone has selected the best sub, things, probably, wouldn't be that different:\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1212661%2F851af739ff078cd09e20edd308028d87%2F3.png?generation=1595536230724131&alt=media\")
\nInterestingly that first 3 best subs for me are using different from my main pipeline (and just have different seed and other minor modifications). Too bad that I decided to ignore them (and even not build an ensemble), and also I ignored an ensemble of models finetuned on a balanced set with other tricks added... but it happens, and in any way I have achieved my main objective for this competition.
wow! 0.94144 1st place solution from great kaggle grandmaster, you deserved that 1st place, sorry for you :(
", "rawMarkdown": "wow! 0.94144 1st place solution from great kaggle grandmaster, you deserved that 1st place, sorry for you :(", "votes": 2 }, { "id": 945693, "postDate": "2020-07-26T04:52:11.077Z", "content": "True winning efforts @iafoss , some competition are too dependent on selection based dynamics rather effort based ... \nHow to get attention of Kaggle team here .
", "rawMarkdown": "True winning efforts @iafoss , some competition are too dependent on selection based dynamics rather effort based ... \nHow to get attention of Kaggle team here .\n", "votes": -5 } ] }, { "id": 950039, "postDate": "2020-07-29T06:25:51.450Z", "content": "https://kaggledb.com/c/prostate-cancer-grade-assessment/best\nWe can see the private ranking if our best subs were selected.. Interesting to see the gold zone does not change much.\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3508221%2F965653db7e3c44693fb6d01265359ad9%2F2020-07-29%20152531.png?generation=1596003946941046&alt=media\")
thanks @kyoshioka47 ! Nice to see your Private(final) versus Private (best) did not change much. \nI have to learn to be able to do that. \nMy Private (final) is 478 and My Private (best) is 54. I definitely started over-training my models at some point... sigh.. but it is a good lesson to learn. thanks for sharing.
", "rawMarkdown": "thanks @kyoshioka47 ! Nice to see your Private(final) versus Private (best) did not change much. \nI have to learn to be able to do that. \nMy Private (final) is 478 and My Private (best) is 54. I definitely started over-training my models at some point... sigh.. but it is a good lesson to learn. thanks for sharing. " } ] }, { "id": 941568, "postDate": "2020-07-23T09:56:29.417Z", "content": "I had this last minute genius idea to weight my ensemble based on LB scores(v3). Turns out the simple average (v2) would've been just fine..π§ \n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2039088%2F382976dd937c5a8257a7d25bc600083a%2Flb.png?generation=1595497580053043&alt=media\")
\nThough I'm grateful for all I've learned.
congrats @kiminya
", "rawMarkdown": "congrats @kiminya " } ] }, { "id": 943964, "postDate": "2020-07-24T18:06:00.670Z", "content": "I got to learn a lot from this even though I submitted only one my private leaderboard performance got me to rank 377 which is very surprising for meπ
", "rawMarkdown": "I got to learn a lot from this even though I submitted only one my private leaderboard performance got me to rank 377 which is very surprising for meπ ", "votes": 1 }, { "id": 942295, "postDate": "2020-07-23T17:25:50.157Z", "content": "this competition it happened to a lot of people. here is mine.
", "rawMarkdown": "this competition it happened to a lot of people. here is mine. ", "votes": 1, "replies": [ { "id": 942297, "postDate": "2020-07-23T17:26:03.123Z", "content": "![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F159759%2Ffafc02c584c80802115a7323bd62f2c8%2Ffinal_submissions.png?generation=1595525161895842&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F455315%2F195c55167f132bb58c9ea257cd08c5df%2Fsad.PNG?generation=1595474447175820&alt=media\")
\nThe submission I made by adding TTA to the panda-inference-w-36-tiles-256 could have gotten me a medal, i was not able to train with my limitted set up and my kernels timed out while training. But even then just the addition of TTA could have given good results in private leader board. Goodluck for the next competitions.
sorry for you! i believe you will select the medal submission next time,best of luck
", "rawMarkdown": "sorry for you! i believe you will select the medal submission next time,best of luck", "votes": 1 } ] }, { "id": 940651, "postDate": "2020-07-23T04:20:22.953Z", "content": "![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3205386%2F8f925b6eed6c0d34862dee8b030bd422%2FUntitled.png?generation=1595478003528276&alt=media\")
@karrak3256 0.927 omg!!!
", "rawMarkdown": "@karrak3256 0.927 omg!!! ", "votes": 1 }, { "id": 940666, "postDate": "2020-07-23T04:35:55.020Z", "content": "this is what we got , whom to complaint now. Only way was to get dream of this π
", "rawMarkdown": "this is what we got , whom to complaint now. Only way was to get dream of this π ", "votes": 2 }, { "id": 940667, "postDate": "2020-07-23T04:37:32.287Z", "content": "@rajnishe bad luck and i know it is heartbreaking :( and i wish you luck for next competitions
", "rawMarkdown": "@rajnishe bad luck and i know it is heartbreaking :( and i wish you luck for next competitions" } ] }, { "id": 952170, "postDate": "2020-07-30T17:13:03.140Z", "content": "Hmm
", "rawMarkdown": "Hmm" }, { "id": 941238, "postDate": "2020-07-23T06:30:58.903Z", "rawMarkdown": "", "votes": 2, "isDeleted": true, "replies": [ { "id": 941271, "postDate": "2020-07-23T06:41:04.133Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 940635, "postDate": "2020-07-23T04:06:19.223Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 941259, "author_name": "Psi", "author_url": "", "post_date": "2020-07-23T06:37:49.410000", "content": "If everyone selected their best scores you would not be silver. I also had a very high scoring private LB kernel but would have never selected it, because it was a very early sub of mine just blending local model with public kernel.
", "votes": 16, "replies": [ { "id": 941265, "author_name": "Mobassir", "author_url": "", "post_date": "2020-07-23T06:39:15.763000", "content": "that's true too @philippsinger :)
", "votes": 0, "replies": [] }, { "id": 941274, "author_name": "Mobassir", "author_url": "", "post_date": "2020-07-23T06:43:12.830000", "content": "@philippsinger one thing i noticed that in recent kaggle competitions (specially in some of them) we can't rely on cv nor on lb much,it really is very frustrating , where is machine learning here when you can't rely much on cv,nor on lb?
", "votes": 2, "replies": [] }, { "id": 941284, "author_name": "Psi", "author_url": "", "post_date": "2020-07-23T06:47:28.193000", "content": "I think only this competition and M5 have been like this. Other competitions had quite good ways to evaluate.
", "votes": 0, "replies": [] }, { "id": 942559, "author_name": "Iafoss", "author_url": "", "post_date": "2020-07-23T20:19:10.167000", "content": "It was quite new and interesting experience for me of learning how to deal with such problem
", "votes": 3, "replies": [] } ] }, { "id": 943535, "author_name": "CPMP", "author_url": "", "post_date": "2020-07-24T12:24:26.437000", "content": "The real question to ask is how could you have selected your best submission? Learning how to do it is the most precious skill one can learn out of kagggle: reliably assess model performance based on train data. If you get good at it then you are a very valuable ML practitioner.
", "votes": 11, "replies": [ { "id": 943621, "author_name": "Mobassir", "author_url": "", "post_date": "2020-07-24T13:29:17.597000", "content": "Thank you @cpmpml \nTrying to learn it better from every competitions mistake and following great discussions from people like you!\nThank you :)
", "votes": 1, "replies": [] }, { "id": 943773, "author_name": "Mehul Sampat", "author_url": "", "post_date": "2020-07-24T15:11:26.113000", "content": "thank you @cpmpml ! great advice. will definitely keep this in mind and keep trying to improve. Cheers!
", "votes": 1, "replies": [] } ] }, { "id": 942573, "author_name": "Iafoss", "author_url": "", "post_date": "2020-07-23T20:39:16.637000", "content": "It happens... There are always a plenty of opportunities that may be missed, but it was my own chose, and if everyone has selected the best sub, things, probably, wouldn't be that different:\n\nInterestingly that first 3 best subs for me are using different from my main pipeline (and just have different seed and other minor modifications). Too bad that I decided to ignore them (and even not build an ensemble), and also I ignored an ensemble of models finetuned on a balanced set with other tricks added... but it happens, and in any way I have achieved my main objective for this competition.
wow! 0.94144 1st place solution from great kaggle grandmaster, you deserved that 1st place, sorry for you :(
", "votes": 2, "replies": [] }, { "id": 945693, "author_name": "Jaideep", "author_url": "", "post_date": "2020-07-26T04:52:11.077000", "content": "True winning efforts @iafoss , some competition are too dependent on selection based dynamics rather effort based ... \nHow to get attention of Kaggle team here .
", "votes": -5, "replies": [] } ] }, { "id": 950039, "author_name": "arutema47", "author_url": "", "post_date": "2020-07-29T06:25:51.450000", "content": "https://kaggledb.com/c/prostate-cancer-grade-assessment/best\nWe can see the private ranking if our best subs were selected.. Interesting to see the gold zone does not change much.\n
thanks @kyoshioka47 ! Nice to see your Private(final) versus Private (best) did not change much. \nI have to learn to be able to do that. \nMy Private (final) is 478 and My Private (best) is 54. I definitely started over-training my models at some point... sigh.. but it is a good lesson to learn. thanks for sharing.
", "votes": 0, "replies": [] } ] }, { "id": 941568, "author_name": "Kiminya", "author_url": "", "post_date": "2020-07-23T09:56:29.417000", "content": "I had this last minute genius idea to weight my ensemble based on LB scores(v3). Turns out the simple average (v2) would've been just fine..π§ \n\nThough I'm grateful for all I've learned.
congrats @kiminya
", "votes": 0, "replies": [] } ] }, { "id": 943964, "author_name": "Digvijay Yadav", "author_url": "", "post_date": "2020-07-24T18:06:00.670000", "content": "I got to learn a lot from this even though I submitted only one my private leaderboard performance got me to rank 377 which is very surprising for meπ
", "votes": 1, "replies": [] }, { "id": 942295, "author_name": "Mehul Sampat", "author_url": "", "post_date": "2020-07-23T17:25:50.157000", "content": "this competition it happened to a lot of people. here is mine.
", "votes": 1, "replies": [ { "id": 942297, "author_name": "Mehul Sampat", "author_url": "", "post_date": "2020-07-23T17:26:03.123000", "content": "
\nThe submission I made by adding TTA to the panda-inference-w-36-tiles-256 could have gotten me a medal, i was not able to train with my limitted set up and my kernels timed out while training. But even then just the addition of TTA could have given good results in private leader board. Goodluck for the next competitions.
sorry for you! i believe you will select the medal submission next time,best of luck
", "votes": 1, "replies": [] } ] }, { "id": 940651, "author_name": "Karan", "author_url": "", "post_date": "2020-07-23T04:20:22.953000", "content": "
@karrak3256 0.927 omg!!!
", "votes": 1, "replies": [] }, { "id": 940666, "author_name": "Rajnish Chauhan", "author_url": "", "post_date": "2020-07-23T04:35:55.020000", "content": "this is what we got , whom to complaint now. Only way was to get dream of this π
", "votes": 2, "replies": [] }, { "id": 940667, "author_name": "Mobassir", "author_url": "", "post_date": "2020-07-23T04:37:32.287000", "content": "@rajnishe bad luck and i know it is heartbreaking :( and i wish you luck for next competitions
", "votes": 0, "replies": [] } ] }, { "id": 952170, "author_name": "Raju Kumar Mishra", "author_url": "", "post_date": "2020-07-30T17:13:03.140000", "content": "Hmm
", "votes": 0, "replies": [] }, { "id": 941238, "author_name": "", "author_url": "", "post_date": "2020-07-23T06:30:58.903000", "content": "", "votes": 2, "replies": [ { "id": 941271, "author_name": "", "author_url": "", "post_date": "2020-07-23T06:41:04.133000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 940635, "author_name": "", "author_url": "", "post_date": "2020-07-23T04:06:19.223000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "940568": "didn't select our silver solution!! \n\n\n", "941259": "If everyone selected their best scores you would not be silver. I also had a very high scoring private LB kernel but would have never selected it, because it was a very early sub of mine just blending local model with public kernel.", "943535": "The real question to ask is how could you have selected your best submission? Learning how to do it is the most precious skill one can learn out of kagggle: reliably assess model performance based on train data. If you get good at it then you are a very valuable ML practitioner. ", "942573": "It happens... There are always a plenty of opportunities that may be missed, but it was my own chose, and if everyone has selected the best sub, things, probably, wouldn't be that different:\n\nInterestingly that first 3 best subs for me are using different from my main pipeline (and just have different seed and other minor modifications). Too bad that I decided to ignore them (and even not build an ensemble), and also I ignored an ensemble of models finetuned on a balanced set with other tricks added... but it happens, and in any way I have achieved my main objective for this competition.", "950039": "https://kaggledb.com/c/prostate-cancer-grade-assessment/best\nWe can see the private ranking if our best subs were selected.. Interesting to see the gold zone does not change much.\n\n", "941568": "I had this last minute genius idea to weight my ensemble based on LB scores(v3). Turns out the simple average (v2) would've been just fine..π§ \n\nThough I'm grateful for all I've learned.", "943964": "I got to learn a lot from this even though I submitted only one my private leaderboard performance got me to rank 377 which is very surprising for meπ ", "942295": "this competition it happened to a lot of people. here is mine. ", "940584": "\nThe submission I made by adding TTA to the [panda-inference-w-36-tiles-256](https://www.kaggle.com/abh001/panda-inference-w-36-tiles-256?scriptVersionId=37418079) could have gotten me a medal, i was not able to train with my limitted set up and my kernels timed out while training. But even then just the addition of TTA could have given good results in private leader board. Goodluck for the next competitions.", "940651": "\n", "952170": "Hmm", "941238": "", "940635": "" } }