Big thanks to @cpmpml for sharing a wealth of knowledge in this competition and many others:\n- Posting his OOF plots and describing his process for feature selection.\n- Encouraging us trust our CV\n- Reminding us the public LB is meaningless.\n- Explaining that less features is more.\nHe doesn't have to help us all by posting insights, but he does it anyways!
", "messages": [ { "id": "542551", "postDate": "06/04/2019 01:23:13", "content": "Big thanks to @cpmpml for sharing a wealth of knowledge in this competition and many others:\n- Posting his OOF plots and describing his process for feature selection.\n- Encouraging us trust our CV\n- Reminding us the public LB is meaningless.\n- Explaining that less features is more.\nHe doesn't have to help us all by posting insights, but he does it anyways!
", "rawMarkdown": "Big thanks to @cpmpml for sharing a wealth of knowledge in this competition and many others:\n- Posting his OOF plots and describing his process for feature selection.\n- Encouraging us trust our CV\n- Reminding us the public LB is meaningless.\n- Explaining that less features is more.\nHe doesn't have to help us all by posting insights, but he does it anyways!", "votes": null }, { "id": "542557", "postDate": "06/04/2019 01:25:48", "content": "Thanks, I swore I would not share anything in this competition, yet I couldn't avoid it ;)
", "rawMarkdown": "Thanks, I swore I would not share anything in this competition, yet I couldn't avoid it ;)", "votes": null }, { "id": "542565", "postDate": "06/04/2019 01:37:19", "content": "Haha.. thanks man
", "rawMarkdown": "Haha.. thanks man", "votes": null }, { "id": "542573", "postDate": "06/04/2019 01:39:33", "content": "It is crazy to see that change between public and private LB and CPMP is one of those with least change on the top LB...really looking forward to their solutions.
", "rawMarkdown": "It is crazy to see that change between public and private LB and CPMP is one of those with least change on the top LB...really looking forward to their solutions.", "votes": null }, { "id": "542588", "postDate": "06/04/2019 01:49:55", "content": "Thanks @cpmpml. Your forum posts are always welcome, specially if you disclose some leaks ;)
", "rawMarkdown": "Thanks @cpmpml. Your forum posts are always welcome, specially if you disclose some leaks ;)", "votes": null }, { "id": "542600", "postDate": "06/04/2019 01:57:55", "content": "Also thanks to CPMP.
\n\nHere is my suggestion: Hang up a shingle as a Kaggle competition organizer, mainly to make sure that the data organization makes sense.
", "rawMarkdown": "Also thanks to CPMP.\n\nHere is my suggestion: Hang up a shingle as a Kaggle competition organizer, mainly to make sure that the data organization makes sense.", "votes": null }, { "id": "542681", "postDate": "06/04/2019 03:27:29", "content": "Yeah absolutely I am also thinking about public score and private score how both are independent to each other.... BTW @cpmpml please post kernel we will eagerly waiting for this.
", "rawMarkdown": "Yeah absolutely I am also thinking about public score and private score how both are independent to each other.... BTW @cpmpml please post kernel we will eagerly waiting for this.", "votes": null }, { "id": "542696", "postDate": "06/04/2019 03:36:52", "content": "I also want to thank you @cpmpml !\nI really appreciate all the time you spent sharing your knowledge. Honestly I would've probably quit this competition, so little sense the difference between LB and CV made to me in the beginning. But I took your advice to trust my CV ;)
", "rawMarkdown": "I also want to thank you @cpmpml !\nI really appreciate all the time you spent sharing your knowledge. Honestly I would've probably quit this competition, so little sense the difference between LB and CV made to me in the beginning. But I took your advice to trust my CV ;)", "votes": null }, { "id": "542878", "postDate": "06/04/2019 06:50:11", "content": "Lots of awesome insights indeed. Thanks for sharing. Now, I am waiting for the ABC team's solution. ;)
", "rawMarkdown": "Lots of awesome insights indeed. Thanks for sharing. Now, I am waiting for the ABC team's solution. ;)", "votes": null }, { "id": "542881", "postDate": "06/04/2019 06:52:31", "content": "Yes thank you @cpmpml. Because of your concerns about the LB and the amount of training data i tried to make my model as general as possible. Result: 124th\nThanks!
", "rawMarkdown": "Yes thank you @cpmpml. Because of your concerns about the LB and the amount of training data i tried to make my model as general as possible. Result: 124th\nThanks!", "votes": null }, { "id": "542915", "postDate": "06/04/2019 07:31:46", "content": "\n\n\nBTW @cpmpml please post kernel we will eagerly waiting for this.
\n
I cannot, rules are too demanding (liability section) for me to post kernels.
", "rawMarkdown": "> BTW @cpmpml please post kernel we will eagerly waiting for this. \n\nI cannot, rules are too demanding (liability section) for me to post kernels.", "votes": null }, { "id": "542972", "postDate": "06/04/2019 08:44:32", "content": "Thank you Uncle, it's always insightful when you drop some knowledge!
", "rawMarkdown": "Thank you Uncle, it's always insightful when you drop some knowledge!", "votes": null }, { "id": "543033", "postDate": "06/04/2019 09:33:25", "content": "@cpmpml Can you elaborate a bit on what you mean with that? Which liabilty would be violated?
\n\nAnd on posting kernels: I actually prefere to talk about ideas behind kernels. It sometimes is nice to have an example, but just handing out a winning solution is nonsense. We will never takle the same problem again (and that's good and why ML is fun!).
", "rawMarkdown": "cpmpml Can you elaborate a bit on what you mean with that? Which liabilty would be violated?\n\nAnd on posting kernels: I actually prefere to talk about ideas behind kernels. It sometimes is nice to have an example, but just handing out a winning solution is nonsense. We will never takle the same problem again (and that's good and why ML is fun!).", "votes": null }, { "id": "543089", "postDate": "06/04/2019 10:20:57", "content": "I took a legal advice on Kaggle rules. I can only tell you to do the same if you want to know more. I discussed this in a post in another competition:\nhttps://www.kaggle.com/c/two-sigma-financial-news/discussion/66828#393743
\n\nMaybe there is no issue when posting a kernel and not claiming a prize, but I'll stay on the safe side.
", "rawMarkdown": "I took a legal advice on Kaggle rules. I can only tell you to do the same if you want to know more. I discussed this in a post in another competition:\nhttps://www.kaggle.com/c/two-sigma-financial-news/discussion/66828#393743\n\nMaybe there is no issue when posting a kernel and not claiming a prize, but I'll stay on the safe side.", "votes": null }, { "id": "543266", "postDate": "06/04/2019 12:28:17", "content": "Thank you CPMP, I tried a lot of feature engineering after I saw your post.
", "rawMarkdown": "Thank you CPMP, I tried a lot of feature engineering after I saw your post.", "votes": null }, { "id": "543298", "postDate": "06/04/2019 13:00:49", "content": "Thank you!
\n\nWell, this is indeed worrisome! And how the hell can Dropout be patented...?! I'm a bit puzzled at the moment.\nHowever, to my knowledge scientific theory and mathematical models cannot be patented. At least in Germany. Still, those models (like Dropout) could be a pending patent (that will be rejected). So, in any way, a sentence like \"[...] does not contain confidential information or trade secrets and is not the subject of a registered patent or pending patent application\" is worrisome.
", "rawMarkdown": "Thank you!\n\nWell, this is indeed worrisome! And how the hell can Dropout be patented...?! I'm a bit puzzled at the moment.\nHowever, to my knowledge scientific theory and mathematical models cannot be patented. At least in Germany. Still, those models (like Dropout) could be a pending patent (that will be rejected). So, in any way, a sentence like \"[...] does not contain confidential information or trade secrets and is not the subject of a registered patent or pending patent application\" is worrisome.", "votes": null }, { "id": "543307", "postDate": "06/04/2019 13:06:30", "content": "I work for a US company, and they take this very seriously.
", "rawMarkdown": "I work for a US company, and they take this very seriously.", "votes": null }, { "id": "543742", "postDate": "06/04/2019 18:42:51", "content": "Thank you! The advice to minimise number of features was very helpfull!
", "rawMarkdown": "Thank you! The advice to minimise number of features was very helpfull!", "votes": null } ], "comments": [ { "id": 542557, "author_name": "cpmpml", "author_url": "", "post_date": "06/04/2019 01:25:48", "content": "Thanks, I swore I would not share anything in this competition, yet I couldn't avoid it ;)
", "votes": null, "replies": [ { "id": 542565, "author_name": "carlospk", "author_url": "", "post_date": "06/04/2019 01:37:19", "content": "Haha.. thanks man
", "votes": null, "replies": [] }, { "id": 542696, "author_name": "antaresnyc", "author_url": "", "post_date": "06/04/2019 03:36:52", "content": "I also want to thank you @cpmpml !\nI really appreciate all the time you spent sharing your knowledge. Honestly I would've probably quit this competition, so little sense the difference between LB and CV made to me in the beginning. But I took your advice to trust my CV ;)
", "votes": null, "replies": [] }, { "id": 543742, "author_name": "bfishh", "author_url": "", "post_date": "06/04/2019 18:42:51", "content": "Thank you! The advice to minimise number of features was very helpfull!
", "votes": null, "replies": [] } ] }, { "id": 542573, "author_name": "zhixuanliu", "author_url": "", "post_date": "06/04/2019 01:39:33", "content": "It is crazy to see that change between public and private LB and CPMP is one of those with least change on the top LB...really looking forward to their solutions.
", "votes": null, "replies": [] }, { "id": 542588, "author_name": "titericz", "author_url": "", "post_date": "06/04/2019 01:49:55", "content": "Thanks @cpmpml. Your forum posts are always welcome, specially if you disclose some leaks ;)
", "votes": null, "replies": [] }, { "id": 542600, "author_name": "petewills", "author_url": "", "post_date": "06/04/2019 01:57:55", "content": "Also thanks to CPMP.
\n\nHere is my suggestion: Hang up a shingle as a Kaggle competition organizer, mainly to make sure that the data organization makes sense.
", "votes": null, "replies": [] }, { "id": 542681, "author_name": "himusoni", "author_url": "", "post_date": "06/04/2019 03:27:29", "content": "Yeah absolutely I am also thinking about public score and private score how both are independent to each other.... BTW @cpmpml please post kernel we will eagerly waiting for this.
", "votes": null, "replies": [ { "id": 542915, "author_name": "cpmpml", "author_url": "", "post_date": "06/04/2019 07:31:46", "content": "\n\n\nBTW @cpmpml please post kernel we will eagerly waiting for this.
\n
I cannot, rules are too demanding (liability section) for me to post kernels.
", "votes": null, "replies": [] }, { "id": 543033, "author_name": "bernir", "author_url": "", "post_date": "06/04/2019 09:33:25", "content": "@cpmpml Can you elaborate a bit on what you mean with that? Which liabilty would be violated?
\n\nAnd on posting kernels: I actually prefere to talk about ideas behind kernels. It sometimes is nice to have an example, but just handing out a winning solution is nonsense. We will never takle the same problem again (and that's good and why ML is fun!).
", "votes": null, "replies": [] }, { "id": 543089, "author_name": "cpmpml", "author_url": "", "post_date": "06/04/2019 10:20:57", "content": "I took a legal advice on Kaggle rules. I can only tell you to do the same if you want to know more. I discussed this in a post in another competition:\nhttps://www.kaggle.com/c/two-sigma-financial-news/discussion/66828#393743
\n\nMaybe there is no issue when posting a kernel and not claiming a prize, but I'll stay on the safe side.
", "votes": null, "replies": [] }, { "id": 543298, "author_name": "bernir", "author_url": "", "post_date": "06/04/2019 13:00:49", "content": "Thank you!
\n\nWell, this is indeed worrisome! And how the hell can Dropout be patented...?! I'm a bit puzzled at the moment.\nHowever, to my knowledge scientific theory and mathematical models cannot be patented. At least in Germany. Still, those models (like Dropout) could be a pending patent (that will be rejected). So, in any way, a sentence like \"[...] does not contain confidential information or trade secrets and is not the subject of a registered patent or pending patent application\" is worrisome.
", "votes": null, "replies": [] }, { "id": 543307, "author_name": "cpmpml", "author_url": "", "post_date": "06/04/2019 13:06:30", "content": "I work for a US company, and they take this very seriously.
", "votes": null, "replies": [] } ] }, { "id": 542878, "author_name": "yassinealouini", "author_url": "", "post_date": "06/04/2019 06:50:11", "content": "Lots of awesome insights indeed. Thanks for sharing. Now, I am waiting for the ABC team's solution. ;)
", "votes": null, "replies": [] }, { "id": 542881, "author_name": "sjakob", "author_url": "", "post_date": "06/04/2019 06:52:31", "content": "Yes thank you @cpmpml. Because of your concerns about the LB and the amount of training data i tried to make my model as general as possible. Result: 124th\nThanks!
", "votes": null, "replies": [] }, { "id": 542972, "author_name": "stanislavblinov", "author_url": "", "post_date": "06/04/2019 08:44:32", "content": "Thank you Uncle, it's always insightful when you drop some knowledge!
", "votes": null, "replies": [] }, { "id": 543266, "author_name": "corochann", "author_url": "", "post_date": "06/04/2019 12:28:17", "content": "Thank you CPMP, I tried a lot of feature engineering after I saw your post.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "542551": "Big thanks to @cpmpml for sharing a wealth of knowledge in this competition and many others:\n- Posting his OOF plots and describing his process for feature selection.\n- Encouraging us trust our CV\n- Reminding us the public LB is meaningless.\n- Explaining that less features is more.\nHe doesn't have to help us all by posting insights, but he does it anyways!", "542557": "Thanks, I swore I would not share anything in this competition, yet I couldn't avoid it ;)", "542565": "Haha.. thanks man", "542573": "It is crazy to see that change between public and private LB and CPMP is one of those with least change on the top LB...really looking forward to their solutions.", "542588": "Thanks @cpmpml. Your forum posts are always welcome, specially if you disclose some leaks ;)", "542600": "Also thanks to CPMP.\n\nHere is my suggestion: Hang up a shingle as a Kaggle competition organizer, mainly to make sure that the data organization makes sense.", "542681": "Yeah absolutely I am also thinking about public score and private score how both are independent to each other.... BTW @cpmpml please post kernel we will eagerly waiting for this.", "542696": "I also want to thank you @cpmpml !\nI really appreciate all the time you spent sharing your knowledge. Honestly I would've probably quit this competition, so little sense the difference between LB and CV made to me in the beginning. But I took your advice to trust my CV ;)", "542878": "Lots of awesome insights indeed. Thanks for sharing. Now, I am waiting for the ABC team's solution. ;)", "542881": "Yes thank you @cpmpml. Because of your concerns about the LB and the amount of training data i tried to make my model as general as possible. Result: 124th\nThanks!", "542915": "> BTW @cpmpml please post kernel we will eagerly waiting for this. \n\nI cannot, rules are too demanding (liability section) for me to post kernels.", "542972": "Thank you Uncle, it's always insightful when you drop some knowledge!", "543033": "cpmpml Can you elaborate a bit on what you mean with that? Which liabilty would be violated?\n\nAnd on posting kernels: I actually prefere to talk about ideas behind kernels. It sometimes is nice to have an example, but just handing out a winning solution is nonsense. We will never takle the same problem again (and that's good and why ML is fun!).", "543089": "I took a legal advice on Kaggle rules. I can only tell you to do the same if you want to know more. I discussed this in a post in another competition:\nhttps://www.kaggle.com/c/two-sigma-financial-news/discussion/66828#393743\n\nMaybe there is no issue when posting a kernel and not claiming a prize, but I'll stay on the safe side.", "543266": "Thank you CPMP, I tried a lot of feature engineering after I saw your post.", "543298": "Thank you!\n\nWell, this is indeed worrisome! And how the hell can Dropout be patented...?! I'm a bit puzzled at the moment.\nHowever, to my knowledge scientific theory and mathematical models cannot be patented. At least in Germany. Still, those models (like Dropout) could be a pending patent (that will be rejected). So, in any way, a sentence like \"[...] does not contain confidential information or trade secrets and is not the subject of a registered patent or pending patent application\" is worrisome.", "543307": "I work for a US company, and they take this very seriously.", "543742": "Thank you! The advice to minimise number of features was very helpfull!" }, "source": "meta" }