We are already aware of the leak in this competition and Kaggle response on the same about revising the competition at some foreseen future. I personally see lots of participants losing interest in the competition.
\n\nDedicating this thread on the impact of the leak for this competition. Honestly, the Santander Competition and its leak took out a large flavour of ML from the competition. I personally didn't enjoy much solving a pseudo data property competition. Would like to know what do you think? Will this competition also end up exploiting the already leaked property to gain huge score even after the revision by Kaggle.
\n\nIs there really a way to fix this competition or this will be let's get on with leak property competition again? Let me know if one with pure ML inisight can stay or switch to other competition for more learning...
We are already aware of the leak in this competition and Kaggle response on the same about revising the competition at some foreseen future. I personally see lots of participants losing interest in the competition.
\n\nDedicating this thread on the impact of the leak for this competition. Honestly, the Santander Competition and its leak took out a large flavour of ML from the competition. I personally didn't enjoy much solving a pseudo data property competition. Would like to know what do you think? Will this competition also end up exploiting the already leaked property to gain huge score even after the revision by Kaggle.
\n\nIs there really a way to fix this competition or this will be let's get on with leak property competition again? Let me know if one with pure ML inisight can stay or switch to other competition for more learning...
It has turned into a web scraping competition in my opinion. Although it is not possible to retrieve every visitor data from the demo account, people can still get total user revenues categorized by some specific dimensions eg. geo, device etc. As far as I know, the ga api is not for the demo account. But if someone is skillful to design a scraper or patient enough to download weekly, or even daily data, it is very easy to build a nearly perfect \"predictive\" model. For me, it is meaningless. It is a waste of time trying to mine answer from the demo account only to have a higher ranking.
\n\nI enjoyed so much reading the discussion and kernels and I learned a lot from it. If the competition continues with allowing the leaks to be exploited, it is probable that the discussion will focus on how they can further exploit the leak to catch up to the best answer. I won't learn as much as I did before the leaks
\n\nI cannot think of any way to minimize the impact of leaked target data on this competition. But I think the Kaggle admins are much smarter and they may come up with a way to deal with that.
", "rawMarkdown": "It has turned into a web scraping competition in my opinion. Although it is not possible to retrieve every visitor data from the demo account, people can still get total user revenues categorized by some specific dimensions eg. geo, device etc. As far as I know, the ga api is not for the demo account. But if someone is skillful to design a scraper or patient enough to download weekly, or even daily data, it is very easy to build a nearly perfect \"predictive\" model. For me, it is meaningless. It is a waste of time trying to mine answer from the demo account only to have a higher ranking.\n\nI enjoyed so much reading the discussion and kernels and I learned a lot from it. If the competition continues with allowing the leaks to be exploited, it is probable that the discussion will focus on how they can further exploit the leak to catch up to the best answer. I won't learn as much as I did before the leaks\n\nI cannot think of any way to minimize the impact of leaked target data on this competition. But I think the Kaggle admins are much smarter and they may come up with a way to deal with that.", "votes": null }, { "id": "403767", "postDate": "10/14/2018 15:02:24", "content": "Has Kaggle really planned to relaunch the competition ? ( Because given the submissions dates on LB , it seems people are still fighting to exploit the leak/property or whatever you call it :p )
\n\nAs for me , glad I didn't jump and spend time on this competition ^^ ..Santander was a huge waste of time for me.
", "rawMarkdown": "Has Kaggle really planned to relaunch the competition ? ( Because given the submissions dates on LB , it seems people are still fighting to exploit the leak/property or whatever you call it :p )\n\nAs for me , glad I didn't jump and spend time on this competition ^^ ..Santander was a huge waste of time for me.", "votes": null }, { "id": "403793", "postDate": "10/14/2018 15:49:12", "content": "I am still waiting for the official response.\nBTW, it's frustrating to see rank drop ;(
", "rawMarkdown": "I am still waiting for the official response.\nBTW, it's frustrating to see rank drop ;(", "votes": null }, { "id": "403796", "postDate": "10/14/2018 15:56:41", "content": "I can relate to this. Aside from ranking, you can continue to challenge yourself by improving ur model. Probably what you have worked on may still be applicable after the rules are revised. Keep it up!
", "rawMarkdown": "I can relate to this. Aside from ranking, you can continue to challenge yourself by improving ur model. Probably what you have worked on may still be applicable after the rules are revised. Keep it up!", "votes": null }, { "id": "403803", "postDate": "10/14/2018 16:10:35", "content": "Me too.\nEvery time I check LB, I get sad...
", "rawMarkdown": "Me too.\nEvery time I check LB, I get sad...", "votes": null }, { "id": "403822", "postDate": "10/14/2018 17:08:33", "content": "I think we should wait a few more days for the official response. But I agree with your point, if the competition goes on with the leak I'll just switch to another one. However, I think kaggle will come up with a better solution
", "rawMarkdown": "I think we should wait a few more days for the official response. But I agree with your point, if the competition goes on with the leak I'll just switch to another one. However, I think kaggle will come up with a better solution", "votes": null }, { "id": "403846", "postDate": "10/14/2018 18:05:27", "content": "Agree @serigne. I thought by end of Santander I would take a lot of amazing feature engineering ideas given the dimensionality and anonymized features. But, in the end, we ended up running for loops to search the target.
", "rawMarkdown": "Agree @serigne. I thought by end of Santander I would take a lot of amazing feature engineering ideas given the dimensionality and anonymized features. But, in the end, we ended up running for loops to search the target.", "votes": null }, { "id": "403900", "postDate": "10/14/2018 21:02:46", "content": "I would consider switching, but there are no other tabular data prize money competition available, other than the astronomy one.
", "rawMarkdown": "I would consider switching, but there are no other tabular data prize money competition available, other than the astronomy one.", "votes": null }, { "id": "404215", "postDate": "10/15/2018 12:54:35", "content": "We should wait for the official response.
", "rawMarkdown": "We should wait for the official response.", "votes": null }, { "id": "404299", "postDate": "10/15/2018 14:38:45", "content": "Don't know why some people are still wasting their neurons over a data which has severe leakage, at the end it would be for no good. Better to switch to some other competitions until some official reset has been done.
", "rawMarkdown": "Don't know why some people are still wasting their neurons over a data which has severe leakage, at the end it would be for no good. Better to switch to some other competitions until some official reset has been done.", "votes": null } ], "comments": [ { "id": 403715, "author_name": "rglimitless", "author_url": "", "post_date": "10/14/2018 12:26:25", "content": "It has turned into a web scraping competition in my opinion. Although it is not possible to retrieve every visitor data from the demo account, people can still get total user revenues categorized by some specific dimensions eg. geo, device etc. As far as I know, the ga api is not for the demo account. But if someone is skillful to design a scraper or patient enough to download weekly, or even daily data, it is very easy to build a nearly perfect \"predictive\" model. For me, it is meaningless. It is a waste of time trying to mine answer from the demo account only to have a higher ranking.
\n\nI enjoyed so much reading the discussion and kernels and I learned a lot from it. If the competition continues with allowing the leaks to be exploited, it is probable that the discussion will focus on how they can further exploit the leak to catch up to the best answer. I won't learn as much as I did before the leaks
\n\nI cannot think of any way to minimize the impact of leaked target data on this competition. But I think the Kaggle admins are much smarter and they may come up with a way to deal with that.
", "votes": null, "replies": [] }, { "id": 403767, "author_name": "serigne", "author_url": "", "post_date": "10/14/2018 15:02:24", "content": "Has Kaggle really planned to relaunch the competition ? ( Because given the submissions dates on LB , it seems people are still fighting to exploit the leak/property or whatever you call it :p )
\n\nAs for me , glad I didn't jump and spend time on this competition ^^ ..Santander was a huge waste of time for me.
", "votes": null, "replies": [ { "id": 403846, "author_name": "shaz13", "author_url": "", "post_date": "10/14/2018 18:05:27", "content": "Agree @serigne. I thought by end of Santander I would take a lot of amazing feature engineering ideas given the dimensionality and anonymized features. But, in the end, we ended up running for loops to search the target.
", "votes": null, "replies": [] } ] }, { "id": 403793, "author_name": "zhengyichen", "author_url": "", "post_date": "10/14/2018 15:49:12", "content": "I am still waiting for the official response.\nBTW, it's frustrating to see rank drop ;(
", "votes": null, "replies": [ { "id": 403796, "author_name": "rglimitless", "author_url": "", "post_date": "10/14/2018 15:56:41", "content": "I can relate to this. Aside from ranking, you can continue to challenge yourself by improving ur model. Probably what you have worked on may still be applicable after the rules are revised. Keep it up!
", "votes": null, "replies": [] }, { "id": 403803, "author_name": "go5kuramubon", "author_url": "", "post_date": "10/14/2018 16:10:35", "content": "Me too.\nEvery time I check LB, I get sad...
", "votes": null, "replies": [] } ] }, { "id": 403822, "author_name": "jsaguiar", "author_url": "", "post_date": "10/14/2018 17:08:33", "content": "I think we should wait a few more days for the official response. But I agree with your point, if the competition goes on with the leak I'll just switch to another one. However, I think kaggle will come up with a better solution
", "votes": null, "replies": [] }, { "id": 403900, "author_name": "sdoria", "author_url": "", "post_date": "10/14/2018 21:02:46", "content": "I would consider switching, but there are no other tabular data prize money competition available, other than the astronomy one.
", "votes": null, "replies": [] }, { "id": 404215, "author_name": "abhishekmamidi", "author_url": "", "post_date": "10/15/2018 12:54:35", "content": "We should wait for the official response.
", "votes": null, "replies": [] }, { "id": 404299, "author_name": "manu845", "author_url": "", "post_date": "10/15/2018 14:38:45", "content": "Don't know why some people are still wasting their neurons over a data which has severe leakage, at the end it would be for no good. Better to switch to some other competitions until some official reset has been done.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "403703": "We are already aware of the leak in this competition and Kaggle response on the same about revising the competition at some foreseen future. I personally see lots of participants losing interest in the competition. \n\nDedicating this thread on the impact of the leak for this competition. Honestly, the Santander Competition and its leak took out a large flavour of ML from the competition. I personally didn't enjoy much solving a pseudo data property competition. Would like to know what do you think? Will this competition also end up exploiting the already leaked property to gain huge score even after the revision by Kaggle. \n\nIs there really a way to fix this competition or this will be `let's get on with leak property` competition again? Let me know if one with pure ML inisight can stay or switch to other competition for more learning...", "403715": "It has turned into a web scraping competition in my opinion. Although it is not possible to retrieve every visitor data from the demo account, people can still get total user revenues categorized by some specific dimensions eg. geo, device etc. As far as I know, the ga api is not for the demo account. But if someone is skillful to design a scraper or patient enough to download weekly, or even daily data, it is very easy to build a nearly perfect \"predictive\" model. For me, it is meaningless. It is a waste of time trying to mine answer from the demo account only to have a higher ranking.\n\nI enjoyed so much reading the discussion and kernels and I learned a lot from it. If the competition continues with allowing the leaks to be exploited, it is probable that the discussion will focus on how they can further exploit the leak to catch up to the best answer. I won't learn as much as I did before the leaks\n\nI cannot think of any way to minimize the impact of leaked target data on this competition. But I think the Kaggle admins are much smarter and they may come up with a way to deal with that.", "403767": "Has Kaggle really planned to relaunch the competition ? ( Because given the submissions dates on LB , it seems people are still fighting to exploit the leak/property or whatever you call it :p )\n\nAs for me , glad I didn't jump and spend time on this competition ^^ ..Santander was a huge waste of time for me.", "403793": "I am still waiting for the official response.\nBTW, it's frustrating to see rank drop ;(", "403796": "I can relate to this. Aside from ranking, you can continue to challenge yourself by improving ur model. Probably what you have worked on may still be applicable after the rules are revised. Keep it up!", "403803": "Me too.\nEvery time I check LB, I get sad...", "403822": "I think we should wait a few more days for the official response. But I agree with your point, if the competition goes on with the leak I'll just switch to another one. However, I think kaggle will come up with a better solution", "403846": "Agree @serigne. I thought by end of Santander I would take a lot of amazing feature engineering ideas given the dimensionality and anonymized features. But, in the end, we ended up running for loops to search the target.", "403900": "I would consider switching, but there are no other tabular data prize money competition available, other than the astronomy one.", "404215": "We should wait for the official response.", "404299": "Don't know why some people are still wasting their neurons over a data which has severe leakage, at the end it would be for no good. Better to switch to some other competitions until some official reset has been done." }, "source": "meta" }