{ "id": 94303, "title": "Congrats to winners", "url": "/competitions/LANL-Earthquake-Prediction/discussion/94303", "author_name": "CPMP", "post_date": "2019-06-04T00:05:56.767000", "votes": 48, "comment_count": 48, "views": 0, "content": "

Congrats to winners, The Zoo.

\n\n

Our best model was submitted yesterday, and would have got us 2nd place at 2.28902, but I am still happy to have survived this monstrous shakeup at the top. And we selected our second best model, which is not so bad afterall...

\n\n

Looking forward to knowing more of the winning solution!

\n\n

And I want to thank my team mates, alone I would not have fared that well.

\n\n

Edit: @giba just told me one trick he used: adjust the mean of test predictions to estimate from the paper. I had estimated the mean to be 6.35 (it is actually 6.32). Moving our best selected model (private LB 2.30630) to that mean yields a private LB of 2.26598 which is pretty close to 1st place. I'm sure however that The Zoo solution is very different from ours, and I am eager to learn about it. We will disclose some of ours shortly, including the famous no magic feature ;)

", "messages": [ { "id": 542450, "postDate": "2019-06-04T00:05:56.767Z", "content": "

Congrats to winners, The Zoo.

\n\n

Our best model was submitted yesterday, and would have got us 2nd place at 2.28902, but I am still happy to have survived this monstrous shakeup at the top. And we selected our second best model, which is not so bad afterall...

\n\n

Looking forward to knowing more of the winning solution!

\n\n

And I want to thank my team mates, alone I would not have fared that well.

\n\n

Edit: @giba just told me one trick he used: adjust the mean of test predictions to estimate from the paper. I had estimated the mean to be 6.35 (it is actually 6.32). Moving our best selected model (private LB 2.30630) to that mean yields a private LB of 2.26598 which is pretty close to 1st place. I'm sure however that The Zoo solution is very different from ours, and I am eager to learn about it. We will disclose some of ours shortly, including the famous no magic feature ;)

", "rawMarkdown": "Congrats to winners, The Zoo.\n\nOur best model was submitted yesterday, and would have got us 2nd place at 2.28902, but I am still happy to have survived this monstrous shakeup at the top. And we selected our second best model, which is not so bad afterall...\n\nLooking forward to knowing more of the winning solution!\n\nAnd I want to thank my team mates, alone I would not have fared that well.\n\nEdit: @giba just told me one trick he used: adjust the mean of test predictions to estimate from the paper. I had estimated the mean to be 6.35 (it is actually 6.32). Moving our best selected model (private LB 2.30630) to that mean yields a private LB of 2.26598 which is pretty close to 1st place. I'm sure however that The Zoo solution is very different from ours, and I am eager to learn about it. We will disclose some of ours shortly, including the famous no magic feature ;)", "votes": 46 }, { "id": 542474, "postDate": "2019-06-04T00:19:22.493Z", "content": "

GG! Congratz to all. Wll post solution tomorrow if some of my mates doesnt do it by then.

", "rawMarkdown": "GG! Congratz to all. Wll post solution tomorrow if some of my mates doesnt do it by then.", "votes": 9 }, { "id": 542496, "postDate": "2019-06-04T00:28:10.550Z", "content": "

Big congrats to the winners \"The Zoo\". I teamed up with 2 of the guys before and I can say that they deserved it.

", "rawMarkdown": "Big congrats to the winners \"The Zoo\". I teamed up with 2 of the guys before and I can say that they deserved it.", "votes": 10 }, { "id": 542456, "postDate": "2019-06-04T00:09:12.473Z", "content": "

I survived but out of gold. However, I selected my best private submission. So it's no regret. Phew. What a crazy shake it is.

", "rawMarkdown": "I survived but out of gold. However, I selected my best private submission. So it's no regret. Phew. What a crazy shake it is.", "votes": 6, "replies": [ { "id": 542471, "postDate": "2019-06-04T00:16:51.407Z", "content": "

Congrats anyway @khahuras ! Big up to all survivors out there! :)

", "rawMarkdown": "Congrats anyway @khahuras ! Big up to all survivors out there! :)", "votes": 2 }, { "id": 542505, "postDate": "2019-06-04T00:36:00.547Z", "content": "

I chose a high mean model to blend, and it saved me ,thx a lot @khahuras

", "rawMarkdown": "I chose a high mean model to blend, and it saved me ,thx a lot @khahuras ", "votes": 1 } ] }, { "id": 543674, "postDate": "2019-06-04T17:31:07.403Z", "content": "

Congrats Winners :)

", "rawMarkdown": "Congrats Winners :)", "votes": 3 }, { "id": 542545, "postDate": "2019-06-04T01:16:55.317Z", "content": "

Congrats!

\n\n

And thanks to CPMP, basically 90% of the hints I got came from you, except Andrew and Vettejeep's Kernels.

", "rawMarkdown": "Congrats! \n\nAnd thanks to CPMP, basically 90% of the hints I got came from you, except Andrew and Vettejeep's Kernels. ", "votes": 3 }, { "id": 542468, "postDate": "2019-06-04T00:16:00.610Z", "content": "

didn't survived the shake up, had 3 submissions that got medals but unfortunately where not selected. funny enough 2 were my 2nd and 3rd submission in the competition.. congrats y'all

", "rawMarkdown": "didn't survived the shake up, had 3 submissions that got medals but unfortunately where not selected. funny enough 2 were my 2nd and 3rd submission in the competition.. congrats y'all", "votes": 3, "replies": [ { "id": 542518, "postDate": "2019-06-04T00:49:11.463Z", "content": "

The public LB was so misleading with very small sample size. So sorry to hear that. None of my submissions were far from winning anything. The test samples were not random. I think the competition was poorly designed. I did training, CV and test, and had models CV mae between 0.5-07, mae of test set were also in the range. These models have done poorly both in public and private LBs. It's really hard to guess which were the right models.

", "rawMarkdown": "The public LB was so misleading with very small sample size. So sorry to hear that. None of my submissions were far from winning anything. The test samples were not random. I think the competition was poorly designed. I did training, CV and test, and had models CV mae between 0.5-07, mae of test set were also in the range. These models have done poorly both in public and private LBs. It's really hard to guess which were the right models. " }, { "id": 542815, "postDate": "2019-06-04T05:39:28.650Z", "content": "

There should be a 3rd leaderboard showing the results assuming everyone selected their best submission ;)

", "rawMarkdown": "There should be a 3rd leaderboard showing the results assuming everyone selected their best submission ;) ", "votes": 3 } ] }, { "id": 542453, "postDate": "2019-06-04T00:07:53.600Z", "content": "

Congrats what a survival :)

", "rawMarkdown": "Congrats what a survival :)", "votes": 3 }, { "id": 542540, "postDate": "2019-06-04T01:07:38.647Z", "content": "

I'm looking forward to seeing what people's magic/non-magic features were. I had a sudden insight in the last few days that we can redefine the target to engineer a totally different variable - the ratio of the length of the overall earthquake period. It's trivial for models to determine the start and end of a quake, so if you can predict the approximate length of the earthquake period as a whole you have an excellent feature to interact with the rest of your variables. I didn't have time to explore this thoroughly but maybe others did. One example of this feature is shown below:

\n\n

\"\"

\n\n

It was very good at identifying the ends of earthquake periods. I only used it in my other, less successful submission. It overfit but not nearly as much as I feared. The model was happy to predict TTF values over 14, and ultimately would have gotten a bronze even though my other models that predicted values this high always did terribly. I'm sure with enough time I could have made this a lot better and scored higher...

", "rawMarkdown": "I'm looking forward to seeing what people's magic/non-magic features were. I had a sudden insight in the last few days that we can redefine the target to engineer a totally different variable - the ratio of the length of the overall earthquake period. It's trivial for models to determine the start and end of a quake, so if you can predict the approximate length of the earthquake period as a whole you have an excellent feature to interact with the rest of your variables. I didn't have time to explore this thoroughly but maybe others did. One example of this feature is shown below:\n\n![](https://i.imgur.com/ZqBsPHF.png)\n\nIt was very good at identifying the ends of earthquake periods. I only used it in my other, less successful submission. It overfit but not nearly as much as I feared. The model was happy to predict TTF values over 14, and ultimately would have gotten a bronze even though my other models that predicted values this high always did terribly. I'm sure with enough time I could have made this a lot better and scored higher...", "votes": 4 }, { "id": 542482, "postDate": "2019-06-04T00:22:14.023Z", "content": "

Congratulations to @CPMP and your team. You managed drop the least of any team near the top of the leaderboard, which is an amazing accomplishment in a competition with this big a shakeup.

\n\n

I also appreciate your open sharing of wonderful kernels. You provided a lot of insight to the community and helped make the competition interesting and informative.

", "rawMarkdown": "Congratulations to @CPMP and your team. You managed drop the least of any team near the top of the leaderboard, which is an amazing accomplishment in a competition with this big a shakeup. \n\nI also appreciate your open sharing of wonderful kernels. You provided a lot of insight to the community and helped make the competition interesting and informative. ", "votes": 4 }, { "id": 543245, "postDate": "2019-06-04T12:09:34.823Z", "content": "

Congrats to all the winners!! Special congrats to ABC, the only constant between many variables!

", "rawMarkdown": "Congrats to all the winners!! Special congrats to ABC, the only constant between many variables!", "votes": 1 }, { "id": 543233, "postDate": "2019-06-04T11:59:34.673Z", "content": "

Congratulations to all survivals

", "rawMarkdown": "Congratulations to all survivals", "votes": 1 }, { "id": 543218, "postDate": "2019-06-04T11:42:12.883Z", "content": "

Congrats to all winners and all participants!!\nIt was very tough competition since public LB was not reliable.

\n\n

I'm impressed that @cpmpml keep top on both public & private (without test mean scaling), I am looking forward to for your next post of no magic feature!

", "rawMarkdown": "Congrats to all winners and all participants!!\nIt was very tough competition since public LB was not reliable.\n\nI'm impressed that @cpmpml keep top on both public & private (without test mean scaling), I am looking forward to for your next post of no magic feature!", "votes": 1, "replies": [ { "id": 543236, "postDate": "2019-06-04T12:02:51.383Z", "content": "
\n

I'm impressed that @cpmpml keep top on both public & private

\n
\n\n

Don't forget my team mates, we did this as a team.

\n\n

I wish the final ranking was based on a blend of public and private LB ranking! We would be 1st by a comfortable margin ;)

", "rawMarkdown": "> I'm impressed that @cpmpml keep top on both public & private\n\nDon't forget my team mates, we did this as a team.\n\nI wish the final ranking was based on a blend of public and private LB ranking! We would be 1st by a comfortable margin ;)", "votes": 3 }, { "id": 543310, "postDate": "2019-06-04T13:11:34.970Z", "content": "

Right, conglatulations @stecasasso and @areveillon too! :)

\n\n

Yeah keep high score on both public/private, even with such a unstable competition is amazing.

", "rawMarkdown": "Right, conglatulations @stecasasso and @areveillon too! :)\n\nYeah keep high score on both public/private, even with such a unstable competition is amazing." } ] }, { "id": 542980, "postDate": "2019-06-04T08:50:02.017Z", "content": "

Congratulations to all winners!

", "rawMarkdown": "Congratulations to all winners!", "votes": 1 }, { "id": 542770, "postDate": "2019-06-04T04:45:57.263Z", "content": "

The other high rank may be lucky by the high variance of dataset, But you and your team's job was so good that make sure you can survival in the big \"earthquake\". I believe it's a difficult work to choose a good submission in so many overfitting submissions.\nThank you for you reply in my other discussion, and congratulations!

", "rawMarkdown": "The other high rank may be lucky by the high variance of dataset, But you and your team's job was so good that make sure you can survival in the big \"earthquake\". I believe it's a difficult work to choose a good submission in so many overfitting submissions.\nThank you for you reply in my other discussion, and congratulations!", "votes": 1 }, { "id": 542741, "postDate": "2019-06-04T04:19:50.697Z", "content": "

Congrats @cpmpml \nWell played ;)

", "rawMarkdown": "Congrats @cpmpml \nWell played ;)", "votes": 1 }, { "id": 542503, "postDate": "2019-06-04T00:31:46.047Z", "content": "

Congrats @cpmpml ! :) I also survived, but couldn't recognize my best solution - it would have been 180 positions above my position. But anyway I am glad.\nAnd congrats to the winners!

", "rawMarkdown": "Congrats @cpmpml ! :) I also survived, but couldn't recognize my best solution - it would have been 180 positions above my position. But anyway I am glad.\nAnd congrats to the winners!", "votes": 1 }, { "id": 542478, "postDate": "2019-06-04T00:20:38.100Z", "content": "

Cant wait to see the winners share solutions!!

", "rawMarkdown": "Cant wait to see the winners share solutions!!", "votes": 1 }, { "id": 542472, "postDate": "2019-06-04T00:17:44.397Z", "content": "

Congrats to all winners, this time i selected my oldest submission but it didn't work :D

", "rawMarkdown": "Congrats to all winners, this time i selected my oldest submission but it didn't work :D", "votes": 1 }, { "id": 542452, "postDate": "2019-06-04T00:07:15.103Z", "content": "

Congrats to your team as well!

", "rawMarkdown": "Congrats to your team as well!", "votes": 1 }, { "id": 543110, "postDate": "2019-06-04T10:33:45.903Z", "content": "

Congrats @cpmpml . I learned a lot from you.

", "rawMarkdown": "Congrats @cpmpml . I learned a lot from you. ", "votes": 2 }, { "id": 542673, "postDate": "2019-06-04T03:21:22.380Z", "content": "

Many many congratulations!!!!

", "rawMarkdown": "Many many congratulations!!!! ", "votes": 2 }, { "id": 542539, "postDate": "2019-06-04T01:03:38.033Z", "content": "

Congrats The Zoo and all the other top teams!! Excellent work and I'm excited to reading more about how they did it.

", "rawMarkdown": "Congrats The Zoo and all the other top teams!! Excellent work and I'm excited to reading more about how they did it.", "votes": 2 }, { "id": 542477, "postDate": "2019-06-04T00:20:35.693Z", "content": "

Congrats on your golden 8th place finish @cpmpml

", "rawMarkdown": "Congrats on your golden 8th place finish @cpmpml ", "votes": 2 }, { "id": 542455, "postDate": "2019-06-04T00:09:08.120Z", "content": "

Congratulations to the winners and to all the top of the leaderboard!

", "rawMarkdown": "Congratulations to the winners and to all the top of the leaderboard!", "votes": 2 }, { "id": 542454, "postDate": "2019-06-04T00:08:23.167Z", "content": "

well done!

", "rawMarkdown": "well done!", "votes": 2 }, { "id": 542451, "postDate": "2019-06-04T00:06:58Z", "content": "

congratulations @cpmpml. So glad to see you didn't shake that much.

", "rawMarkdown": "congratulations @cpmpml. So glad to see you didn't shake that much.", "votes": 2, "replies": [ { "id": 542457, "postDate": "2019-06-04T00:09:58.903Z", "content": "

It seems I survive shakeups in general ;)

", "rawMarkdown": "It seems I survive shakeups in general ;)", "votes": 4 }, { "id": 542534, "postDate": "2019-06-04T01:00:29.193Z", "content": "

what is the secret ? :P

", "rawMarkdown": "what is the secret ? :P " }, { "id": 542536, "postDate": "2019-06-04T01:02:25.583Z", "content": "

Don't gamble, trust your CV.

", "rawMarkdown": "Don't gamble, trust your CV. ", "votes": 6 }, { "id": 542767, "postDate": "2019-06-04T04:45:05.847Z", "content": "

@cpmpml Do you chose the best CV score to submit? I really wonder how to avoid shake up :P

", "rawMarkdown": " @cpmpml Do you chose the best CV score to submit? I really wonder how to avoid shake up :P" }, { "id": 543092, "postDate": "2019-06-04T10:24:16.033Z", "content": "

Yes, we selected best CV score, but we used CV with two different evaluation functions, see my write up.

", "rawMarkdown": "Yes, we selected best CV score, but we used CV with two different evaluation functions, see my write up.", "votes": 1 } ] }, { "id": 547821, "postDate": "2019-06-08T10:41:11.733Z", "content": "

Congrats winners !

", "rawMarkdown": "Congrats winners !" }, { "id": 543532, "postDate": "2019-06-04T15:23:32.570Z", "content": "

Congrats for all ! and thanks for all who contributed in this competition. We learned a lot. Also I hope their solution would be applicable to save all human beings <3

", "rawMarkdown": "Congrats for all ! and thanks for all who contributed in this competition. We learned a lot. Also I hope their solution would be applicable to save all human beings <3" }, { "id": 543327, "postDate": "2019-06-04T13:25:13.333Z", "content": "

Many Many congratulations to the Winners..!! ..From what I have heard, this was a very hard competition and me still being a toddler in Machine Learning .. I look forward to learning from all the greats..!!!

", "rawMarkdown": "Many Many congratulations to the Winners..!! ..From what I have heard, this was a very hard competition and me still being a toddler in Machine Learning .. I look forward to learning from all the greats..!!!" }, { "id": 543308, "postDate": "2019-06-04T13:07:21.430Z", "content": "

Congrats on the golden 8th place finish to @cpmpml, I truly appreciate you sharing your thoughts, kernel after the competition and helped make the discussion post interesting:)!

", "rawMarkdown": "Congrats on the golden 8th place finish to @cpmpml, I truly appreciate you sharing your thoughts, kernel after the competition and helped make the discussion post interesting:)!" }, { "id": 542691, "postDate": "2019-06-04T03:34:50.240Z", "content": "

cannot wait to see how good did NN based solutions get

", "rawMarkdown": "cannot wait to see how good did NN based solutions get " }, { "id": 542643, "postDate": "2019-06-04T02:34:29.660Z", "content": "

One question plz, I only submitted the prediction .csv file but not my model, it only contains test id and some numbers... how does Kaggle know how is my model perform for the private test dataset. Thanks

", "rawMarkdown": "One question plz, I only submitted the prediction .csv file but not my model, it only contains test id and some numbers... how does Kaggle know how is my model perform for the private test dataset. Thanks", "replies": [ { "id": 542752, "postDate": "2019-06-04T04:33:19.323Z", "content": "

the prediction .csv contains predictions for both the public and private leaderboards. so, when you submit, kaggle knows your score for both

", "rawMarkdown": "the prediction .csv contains predictions for both the public and private leaderboards. so, when you submit, kaggle knows your score for both", "votes": 1 }, { "id": 542798, "postDate": "2019-06-04T05:09:49.423Z", "content": "

Thanks for your answer. But prediction file only has 2624 items, where does the private dataset come from? Or that approximately 87% of the test data is somehow hidden in 2624 public test data?

", "rawMarkdown": "Thanks for your answer. But prediction file only has 2624 items, where does the private dataset come from? Or that approximately 87% of the test data is somehow hidden in 2624 public test data?" }, { "id": 543095, "postDate": "2019-06-04T10:25:59.687Z", "content": "

yes, approximately 13% of the 2624 items is the pubilc dataset, and approximately 87% of the 2624 items is the private dataset

", "rawMarkdown": "yes, approximately 13% of the 2624 items is the pubilc dataset, and approximately 87% of the 2624 items is the private dataset" }, { "id": 543957, "postDate": "2019-06-05T00:59:54.890Z", "content": "

Thank you, I get it!

", "rawMarkdown": "Thank you, I get it!" } ] }, { "id": 543223, "postDate": "2019-06-04T11:48:40.307Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 542474, "author_name": "Psi", "author_url": "", "post_date": "2019-06-04T00:19:22.493000", "content": "

GG! Congratz to all. Wll post solution tomorrow if some of my mates doesnt do it by then.

", "votes": 9, "replies": [] }, { "id": 542496, "author_name": "Giba", "author_url": "", "post_date": "2019-06-04T00:28:10.550000", "content": "

Big congrats to the winners \"The Zoo\". I teamed up with 2 of the guys before and I can say that they deserved it.

", "votes": 10, "replies": [] }, { "id": 542456, "author_name": "Kha Vo", "author_url": "", "post_date": "2019-06-04T00:09:12.473000", "content": "

I survived but out of gold. However, I selected my best private submission. So it's no regret. Phew. What a crazy shake it is.

", "votes": 6, "replies": [ { "id": 542471, "author_name": "bluetrain", "author_url": "", "post_date": "2019-06-04T00:16:51.407000", "content": "

Congrats anyway @khahuras ! Big up to all survivors out there! :)

", "votes": 2, "replies": [] }, { "id": 542505, "author_name": "ZeroWen", "author_url": "", "post_date": "2019-06-04T00:36:00.547000", "content": "

I chose a high mean model to blend, and it saved me ,thx a lot @khahuras

", "votes": 1, "replies": [] } ] }, { "id": 543674, "author_name": "Gaurav Bhardwaj", "author_url": "", "post_date": "2019-06-04T17:31:07.403000", "content": "

Congrats Winners :)

", "votes": 3, "replies": [] }, { "id": 542545, "author_name": "vroom", "author_url": "", "post_date": "2019-06-04T01:16:55.317000", "content": "

Congrats!

\n\n

And thanks to CPMP, basically 90% of the hints I got came from you, except Andrew and Vettejeep's Kernels.

", "votes": 3, "replies": [] }, { "id": 542468, "author_name": "Holar", "author_url": "", "post_date": "2019-06-04T00:16:00.610000", "content": "

didn't survived the shake up, had 3 submissions that got medals but unfortunately where not selected. funny enough 2 were my 2nd and 3rd submission in the competition.. congrats y'all

", "votes": 3, "replies": [ { "id": 542518, "author_name": "joejeo1", "author_url": "", "post_date": "2019-06-04T00:49:11.463000", "content": "

The public LB was so misleading with very small sample size. So sorry to hear that. None of my submissions were far from winning anything. The test samples were not random. I think the competition was poorly designed. I did training, CV and test, and had models CV mae between 0.5-07, mae of test set were also in the range. These models have done poorly both in public and private LBs. It's really hard to guess which were the right models.

", "votes": 0, "replies": [] }, { "id": 542815, "author_name": "Jussi Collin", "author_url": "", "post_date": "2019-06-04T05:39:28.650000", "content": "

There should be a 3rd leaderboard showing the results assuming everyone selected their best submission ;)

", "votes": 3, "replies": [] } ] }, { "id": 542453, "author_name": "Manoj", "author_url": "", "post_date": "2019-06-04T00:07:53.600000", "content": "

Congrats what a survival :)

", "votes": 3, "replies": [] }, { "id": 542540, "author_name": "RNA", "author_url": "", "post_date": "2019-06-04T01:07:38.647000", "content": "

I'm looking forward to seeing what people's magic/non-magic features were. I had a sudden insight in the last few days that we can redefine the target to engineer a totally different variable - the ratio of the length of the overall earthquake period. It's trivial for models to determine the start and end of a quake, so if you can predict the approximate length of the earthquake period as a whole you have an excellent feature to interact with the rest of your variables. I didn't have time to explore this thoroughly but maybe others did. One example of this feature is shown below:

\n\n

\"\"

\n\n

It was very good at identifying the ends of earthquake periods. I only used it in my other, less successful submission. It overfit but not nearly as much as I feared. The model was happy to predict TTF values over 14, and ultimately would have gotten a bronze even though my other models that predicted values this high always did terribly. I'm sure with enough time I could have made this a lot better and scored higher...

", "votes": 4, "replies": [] }, { "id": 542482, "author_name": "Bill Holst", "author_url": "", "post_date": "2019-06-04T00:22:14.023000", "content": "

Congratulations to @CPMP and your team. You managed drop the least of any team near the top of the leaderboard, which is an amazing accomplishment in a competition with this big a shakeup.

\n\n

I also appreciate your open sharing of wonderful kernels. You provided a lot of insight to the community and helped make the competition interesting and informative.

", "votes": 4, "replies": [] }, { "id": 543245, "author_name": "gmobaz", "author_url": "", "post_date": "2019-06-04T12:09:34.823000", "content": "

Congrats to all the winners!! Special congrats to ABC, the only constant between many variables!

", "votes": 1, "replies": [] }, { "id": 543233, "author_name": "belgraviton", "author_url": "", "post_date": "2019-06-04T11:59:34.673000", "content": "

Congratulations to all survivals

", "votes": 1, "replies": [] }, { "id": 543218, "author_name": "corochann", "author_url": "", "post_date": "2019-06-04T11:42:12.883000", "content": "

Congrats to all winners and all participants!!\nIt was very tough competition since public LB was not reliable.

\n\n

I'm impressed that @cpmpml keep top on both public & private (without test mean scaling), I am looking forward to for your next post of no magic feature!

", "votes": 1, "replies": [ { "id": 543236, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-04T12:02:51.383000", "content": "
\n

I'm impressed that @cpmpml keep top on both public & private

\n
\n\n

Don't forget my team mates, we did this as a team.

\n\n

I wish the final ranking was based on a blend of public and private LB ranking! We would be 1st by a comfortable margin ;)

", "votes": 3, "replies": [] }, { "id": 543310, "author_name": "corochann", "author_url": "", "post_date": "2019-06-04T13:11:34.970000", "content": "

Right, conglatulations @stecasasso and @areveillon too! :)

\n\n

Yeah keep high score on both public/private, even with such a unstable competition is amazing.

", "votes": 0, "replies": [] } ] }, { "id": 542980, "author_name": "Stanislav Blinov", "author_url": "", "post_date": "2019-06-04T08:50:02.017000", "content": "

Congratulations to all winners!

", "votes": 1, "replies": [] }, { "id": 542770, "author_name": "Kyle", "author_url": "", "post_date": "2019-06-04T04:45:57.263000", "content": "

The other high rank may be lucky by the high variance of dataset, But you and your team's job was so good that make sure you can survival in the big \"earthquake\". I believe it's a difficult work to choose a good submission in so many overfitting submissions.\nThank you for you reply in my other discussion, and congratulations!

", "votes": 1, "replies": [] }, { "id": 542741, "author_name": "Carlos Prades K.", "author_url": "", "post_date": "2019-06-04T04:19:50.697000", "content": "

Congrats @cpmpml \nWell played ;)

", "votes": 1, "replies": [] }, { "id": 542503, "author_name": "AM", "author_url": "", "post_date": "2019-06-04T00:31:46.047000", "content": "

Congrats @cpmpml ! :) I also survived, but couldn't recognize my best solution - it would have been 180 positions above my position. But anyway I am glad.\nAnd congrats to the winners!

", "votes": 1, "replies": [] }, { "id": 542478, "author_name": "Heisenger", "author_url": "", "post_date": "2019-06-04T00:20:38.100000", "content": "

Cant wait to see the winners share solutions!!

", "votes": 1, "replies": [] }, { "id": 542472, "author_name": "Murat Korkmaz", "author_url": "", "post_date": "2019-06-04T00:17:44.397000", "content": "

Congrats to all winners, this time i selected my oldest submission but it didn't work :D

", "votes": 1, "replies": [] }, { "id": 542452, "author_name": "lucaskg", "author_url": "", "post_date": "2019-06-04T00:07:15.103000", "content": "

Congrats to your team as well!

", "votes": 1, "replies": [] }, { "id": 543110, "author_name": "Karan Jakhar", "author_url": "", "post_date": "2019-06-04T10:33:45.903000", "content": "

Congrats @cpmpml . I learned a lot from you.

", "votes": 2, "replies": [] }, { "id": 542673, "author_name": "Himanshu Soni", "author_url": "", "post_date": "2019-06-04T03:21:22.380000", "content": "

Many many congratulations!!!!

", "votes": 2, "replies": [] }, { "id": 542539, "author_name": "Rob Mulla", "author_url": "", "post_date": "2019-06-04T01:03:38.033000", "content": "

Congrats The Zoo and all the other top teams!! Excellent work and I'm excited to reading more about how they did it.

", "votes": 2, "replies": [] }, { "id": 542477, "author_name": "Abhinand", "author_url": "", "post_date": "2019-06-04T00:20:35.693000", "content": "

Congrats on your golden 8th place finish @cpmpml

", "votes": 2, "replies": [] }, { "id": 542455, "author_name": "bluetrain", "author_url": "", "post_date": "2019-06-04T00:09:08.120000", "content": "

Congratulations to the winners and to all the top of the leaderboard!

", "votes": 2, "replies": [] }, { "id": 542454, "author_name": "spongebob", "author_url": "", "post_date": "2019-06-04T00:08:23.167000", "content": "

well done!

", "votes": 2, "replies": [] }, { "id": 542451, "author_name": "Massoud Hosseinali", "author_url": "", "post_date": "2019-06-04T00:06:58", "content": "

congratulations @cpmpml. So glad to see you didn't shake that much.

", "votes": 2, "replies": [ { "id": 542457, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-04T00:09:58.903000", "content": "

It seems I survive shakeups in general ;)

", "votes": 4, "replies": [] }, { "id": 542534, "author_name": "Dimosthenis Karaflos", "author_url": "", "post_date": "2019-06-04T01:00:29.193000", "content": "

what is the secret ? :P

", "votes": 0, "replies": [] }, { "id": 542536, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-04T01:02:25.583000", "content": "

Don't gamble, trust your CV.

", "votes": 6, "replies": [] }, { "id": 542767, "author_name": "ZeroWen", "author_url": "", "post_date": "2019-06-04T04:45:05.847000", "content": "

@cpmpml Do you chose the best CV score to submit? I really wonder how to avoid shake up :P

", "votes": 0, "replies": [] }, { "id": 543092, "author_name": "CPMP", "author_url": "", "post_date": "2019-06-04T10:24:16.033000", "content": "

Yes, we selected best CV score, but we used CV with two different evaluation functions, see my write up.

", "votes": 1, "replies": [] } ] }, { "id": 547821, "author_name": "Asif Bin Latif", "author_url": "", "post_date": "2019-06-08T10:41:11.733000", "content": "

Congrats winners !

", "votes": 0, "replies": [] }, { "id": 543532, "author_name": "YasserSinjab", "author_url": "", "post_date": "2019-06-04T15:23:32.570000", "content": "

Congrats for all ! and thanks for all who contributed in this competition. We learned a lot. Also I hope their solution would be applicable to save all human beings <3

", "votes": 0, "replies": [] }, { "id": 543327, "author_name": "Saikat Biswas", "author_url": "", "post_date": "2019-06-04T13:25:13.333000", "content": "

Many Many congratulations to the Winners..!! ..From what I have heard, this was a very hard competition and me still being a toddler in Machine Learning .. I look forward to learning from all the greats..!!!

", "votes": 0, "replies": [] }, { "id": 543308, "author_name": "", "author_url": "", "post_date": "2019-06-04T13:07:21.430000", "content": "", "votes": 0, "replies": [] }, { "id": 542691, "author_name": "", "author_url": "", "post_date": "2019-06-04T03:34:50.240000", "content": "", "votes": 0, "replies": [] }, { "id": 542643, "author_name": "", "author_url": "", "post_date": "2019-06-04T02:34:29.660000", "content": "", "votes": 0, "replies": [ { "id": 542752, "author_name": "", "author_url": "", "post_date": "2019-06-04T04:33:19.323000", "content": "", "votes": 1, "replies": [] }, { "id": 542798, "author_name": "", "author_url": "", "post_date": "2019-06-04T05:09:49.423000", "content": "", "votes": 0, "replies": [] }, { "id": 543095, "author_name": "", "author_url": "", "post_date": "2019-06-04T10:25:59.687000", "content": "", "votes": 0, "replies": [] }, { "id": 543957, "author_name": "", "author_url": "", "post_date": "2019-06-05T00:59:54.890000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 543223, "author_name": "", "author_url": "", "post_date": "2019-06-04T11:48:40.307000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "542450": "Congrats to winners, The Zoo.\n\nOur best model was submitted yesterday, and would have got us 2nd place at 2.28902, but I am still happy to have survived this monstrous shakeup at the top. And we selected our second best model, which is not so bad afterall...\n\nLooking forward to knowing more of the winning solution!\n\nAnd I want to thank my team mates, alone I would not have fared that well.\n\nEdit: @giba just told me one trick he used: adjust the mean of test predictions to estimate from the paper. I had estimated the mean to be 6.35 (it is actually 6.32). Moving our best selected model (private LB 2.30630) to that mean yields a private LB of 2.26598 which is pretty close to 1st place. I'm sure however that The Zoo solution is very different from ours, and I am eager to learn about it. We will disclose some of ours shortly, including the famous no magic feature ;)", "542474": "GG! Congratz to all. Wll post solution tomorrow if some of my mates doesnt do it by then.", "542496": "Big congrats to the winners \"The Zoo\". I teamed up with 2 of the guys before and I can say that they deserved it.", "542456": "I survived but out of gold. However, I selected my best private submission. So it's no regret. Phew. What a crazy shake it is.", "543674": "Congrats Winners :)", "542545": "Congrats! \n\nAnd thanks to CPMP, basically 90% of the hints I got came from you, except Andrew and Vettejeep's Kernels. ", "542468": "didn't survived the shake up, had 3 submissions that got medals but unfortunately where not selected. funny enough 2 were my 2nd and 3rd submission in the competition.. congrats y'all", "542453": "Congrats what a survival :)", "542540": "I'm looking forward to seeing what people's magic/non-magic features were. I had a sudden insight in the last few days that we can redefine the target to engineer a totally different variable - the ratio of the length of the overall earthquake period. It's trivial for models to determine the start and end of a quake, so if you can predict the approximate length of the earthquake period as a whole you have an excellent feature to interact with the rest of your variables. I didn't have time to explore this thoroughly but maybe others did. One example of this feature is shown below:\n\n![](https://i.imgur.com/ZqBsPHF.png)\n\nIt was very good at identifying the ends of earthquake periods. I only used it in my other, less successful submission. It overfit but not nearly as much as I feared. The model was happy to predict TTF values over 14, and ultimately would have gotten a bronze even though my other models that predicted values this high always did terribly. I'm sure with enough time I could have made this a lot better and scored higher...", "542482": "Congratulations to @CPMP and your team. You managed drop the least of any team near the top of the leaderboard, which is an amazing accomplishment in a competition with this big a shakeup. \n\nI also appreciate your open sharing of wonderful kernels. You provided a lot of insight to the community and helped make the competition interesting and informative. ", "543245": "Congrats to all the winners!! Special congrats to ABC, the only constant between many variables!", "543233": "Congratulations to all survivals", "543218": "Congrats to all winners and all participants!!\nIt was very tough competition since public LB was not reliable.\n\nI'm impressed that @cpmpml keep top on both public & private (without test mean scaling), I am looking forward to for your next post of no magic feature!", "542980": "Congratulations to all winners!", "542770": "The other high rank may be lucky by the high variance of dataset, But you and your team's job was so good that make sure you can survival in the big \"earthquake\". I believe it's a difficult work to choose a good submission in so many overfitting submissions.\nThank you for you reply in my other discussion, and congratulations!", "542741": "Congrats @cpmpml \nWell played ;)", "542503": "Congrats @cpmpml ! :) I also survived, but couldn't recognize my best solution - it would have been 180 positions above my position. But anyway I am glad.\nAnd congrats to the winners!", "542478": "Cant wait to see the winners share solutions!!", "542472": "Congrats to all winners, this time i selected my oldest submission but it didn't work :D", "542452": "Congrats to your team as well!", "543110": "Congrats @cpmpml . I learned a lot from you. ", "542673": "Many many congratulations!!!! ", "542539": "Congrats The Zoo and all the other top teams!! Excellent work and I'm excited to reading more about how they did it.", "542477": "Congrats on your golden 8th place finish @cpmpml ", "542455": "Congratulations to the winners and to all the top of the leaderboard!", "542454": "well done!", "542451": "congratulations @cpmpml. So glad to see you didn't shake that much.", "547821": "Congrats winners !", "543532": "Congrats for all ! and thanks for all who contributed in this competition. We learned a lot. Also I hope their solution would be applicable to save all human beings <3", "543327": "Many Many congratulations to the Winners..!! ..From what I have heard, this was a very hard competition and me still being a toddler in Machine Learning .. I look forward to learning from all the greats..!!!", "543308": "Congrats on the golden 8th place finish to @cpmpml, I truly appreciate you sharing your thoughts, kernel after the competition and helped make the discussion post interesting:)!", "542691": "cannot wait to see how good did NN based solutions get ", "542643": "One question plz, I only submitted the prediction .csv file but not my model, it only contains test id and some numbers... how does Kaggle know how is my model perform for the private test dataset. Thanks", "543223": "" } }