{ "id": 81956, "title": "Deleted post", "url": "/competitions/elo-merchant-category-recommendation/discussion/81956", "author_name": "", "post_date": "2019-02-26T11:23:28.121273100Z", "votes": 51, "comment_count": 45, "views": 0, "content": "

To Kaggle team, this is really bad and unfair.

\n\n

Apparently, someone shared interesting info few hours ago, then got advice to have not posted it before competition end, then the post was deleted.

\n\n

Deleting something of value is even worse than publishing it. Indeed, it creates unfairness between those who could see it and those who could not.

\n\n

The post should be reposted IMHO. If someone got it via notifications in email, then please repost.

\n\n

Edited: link to the deleted post: https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/81915

\n\n

PS. I'm not worried about my rank given how low it is. But this may impact top of the LB, which is really bad.

", "messages": [ { "id": "478593", "postDate": "02/26/2019 11:23:28", "content": "

To Kaggle team, this is really bad and unfair.

\n\n

Apparently, someone shared interesting info few hours ago, then got advice to have not posted it before competition end, then the post was deleted.

\n\n

Deleting something of value is even worse than publishing it. Indeed, it creates unfairness between those who could see it and those who could not.

\n\n

The post should be reposted IMHO. If someone got it via notifications in email, then please repost.

\n\n

Edited: link to the deleted post: https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/81915

\n\n

PS. I'm not worried about my rank given how low it is. But this may impact top of the LB, which is really bad.

", "rawMarkdown": "To Kaggle team, this is really bad and unfair.\n\nApparently, someone shared interesting info few hours ago, then got advice to have not posted it before competition end, then the post was deleted.\n\nDeleting something of value is even worse than publishing it. Indeed, it creates unfairness between those who could see it and those who could not.\n\nThe post should be reposted IMHO. If someone got it via notifications in email, then please repost.\n\nEdited: link to the deleted post: https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/81915\n\nPS. I'm not worried about my rank given how low it is. But this may impact top of the LB, which is really bad.", "votes": null }, { "id": "478649", "postDate": "02/26/2019 12:52:45", "content": "

I advised that it would be nice to share it after the competition. I did not know if the publisher would delete the article.\nI was very embarrassed today. The Top2 Solution has come up before end of the competition and the ID of the competitor in Top10 has been deleted(170 submission).

", "rawMarkdown": "I advised that it would be nice to share it after the competition. I did not know if the publisher would delete the article.\nI was very embarrassed today. The Top2 Solution has come up before end of the competition and the ID of the competitor in Top10 has been deleted(170 submission).", "votes": null }, { "id": "478651", "postDate": "02/26/2019 12:53:31", "content": "

@cpmpml, are you refering to a post tiltled \"Something I want to say...\" ? If so, I saw it earlier and the whole discussion was all criptic and I wondered what the point of it was. The person said they will share tomorrow as of the time I read it. I do not know if more was discussed since I read it a few hours ago.

", "rawMarkdown": "cpmpml, are you refering to a post tiltled \"Something I want to say...\" ? If so, I saw it earlier and the whole discussion was all criptic and I wondered what the point of it was. The person said they will share tomorrow as of the time I read it. I do not know if more was discussed since I read it a few hours ago.", "votes": null }, { "id": "478653", "postDate": "02/26/2019 12:57:52", "content": "

Edit history under \"Options\" button for original post is all we need.

", "rawMarkdown": "Edit history under \"Options\" button for original post is all we need.", "votes": null }, { "id": "478654", "postDate": "02/26/2019 13:00:57", "content": "

It's overfitting

", "rawMarkdown": "It's overfitting", "votes": null }, { "id": "478662", "postDate": "02/26/2019 13:09:14", "content": "

Great idea, you should post it in the product feedback forum.

", "rawMarkdown": "Great idea, you should post it in the product feedback forum.", "votes": null }, { "id": "478663", "postDate": "02/26/2019 13:09:24", "content": "

What is overfitting?

", "rawMarkdown": "What is overfitting?", "votes": null }, { "id": "478664", "postDate": "02/26/2019 13:09:56", "content": "

I don't know what I am referring to, that's the whole point of the discussion here.

", "rawMarkdown": "I don't know what I am referring to, that's the whole point of the discussion here.", "votes": null }, { "id": "478665", "postDate": "02/26/2019 13:10:17", "content": "

You did not do anything wrong.

", "rawMarkdown": "You did not do anything wrong.", "votes": null }, { "id": "478667", "postDate": "02/26/2019 13:11:11", "content": "
\n

The Top2 Solution has come up before end of the competition

\n
\n\n

Wow, that's a serious breach.

", "rawMarkdown": "> The Top2 Solution has come up before end of the competition \n\nWow, that's a serious breach.", "votes": null }, { "id": "478670", "postDate": "02/26/2019 13:15:20", "content": "

@cpmpml, how did you know a post was deleted then? I mean how can one find out?

\n\n

Update:

\n\n

I forgot about the follow tab for the discussions. You probably got a notification as a result of following the discussion I guess.

", "rawMarkdown": "cpmpml, how did you know a post was deleted then? I mean how can one find out?\n\nUpdate:\n\nI forgot about the follow tab for the discussions. You probably got a notification as a result of following the discussion I guess.", "votes": null }, { "id": "478671", "postDate": "02/26/2019 13:17:31", "content": "
\n

... and the ID of the competitor in Top10 has been deleted(170 submission).

\n
\n\n

@yeomin, I don't think I understand what you mean. Can you elaborate?

", "rawMarkdown": ">... and the ID of the competitor in Top10 has been deleted(170 submission).\n\n@yeomin, I don't think I understand what you mean. Can you elaborate?", "votes": null }, { "id": "478673", "postDate": "02/26/2019 13:19:02", "content": "

It was @cmsm1

", "rawMarkdown": "It was @cmsm1", "votes": null }, { "id": "478675", "postDate": "02/26/2019 13:23:12", "content": "

I saw it mentioned in KggleNoobs slack.

", "rawMarkdown": "I saw it mentioned in KggleNoobs slack.", "votes": null }, { "id": "478676", "postDate": "02/26/2019 13:25:51", "content": "

Wow! @yeomin, you meant an actual solution. I have not seen that post. The post I referred to in my post below had a link to a feature selection code. That is why I came back to look for it but found that it is no longer there.

", "rawMarkdown": "Wow! @yeomin, you meant an actual solution. I have not seen that post. The post I referred to in my post below had a link to a feature selection code. That is why I came back to look for it but found that it is no longer there.", "votes": null }, { "id": "478677", "postDate": "02/26/2019 13:28:40", "content": "

I see. It looks like it is a different post from the one I was talking about. That post had a link to a feature selection code. That is why I came back to look for it but found that it is no longer there as well.

", "rawMarkdown": "I see. It looks like it is a different post from the one I was talking about. That post had a link to a feature selection code. That is why I came back to look for it but found that it is no longer there as well.", "votes": null }, { "id": "478681", "postDate": "02/26/2019 13:34:24", "content": "

And today there are so many deleted teams.:)

", "rawMarkdown": "And today there are so many deleted teams.:)", "votes": null }, { "id": "478683", "postDate": "02/26/2019 13:37:17", "content": "

I didn't do anything today but I advanced three places. very interesting

", "rawMarkdown": "I didn't do anything today but I advanced three places. very interesting", "votes": null }, { "id": "478684", "postDate": "02/26/2019 13:39:23", "content": "

It was not exactly a solution, but in summary\n1. Simplified model usage information (xgb, nn, lgbm ridge, etc.), number of features used\n2. outlier model rank pairwise ensemble\n3. It seems to use the target value divided by discrete. The target value can be converted to log2 (x + 1e-10). It seemed here that 10 of 1e-10 was divided into discrete ranges from 1 to 10. For example, -31 to -33

", "rawMarkdown": "It was not exactly a solution, but in summary\n1. Simplified model usage information (xgb, nn, lgbm ridge, etc.), number of features used\n2. outlier model rank pairwise ensemble\n3. It seems to use the target value divided by discrete. The target value can be converted to log2 (x + 1e-10). It seemed here that 10 of 1e-10 was divided into discrete ranges from 1 to 10. For example, -31 to -33", "votes": null }, { "id": "478685", "postDate": "02/26/2019 13:39:49", "content": "

People may think it will prevent them from being caught cheating.

\n\n

But it won't :)

", "rawMarkdown": "People may think it will prevent them from being caught cheating.\n\nBut it won't :)", "votes": null }, { "id": "478687", "postDate": "02/26/2019 13:41:18", "content": "

Thanks!

\n\n

Made a thread request on the Product Forum

", "rawMarkdown": "Thanks!\n\nMade a [thread](https://www.kaggle.com/product-feedback/81968) request on the Product Forum", "votes": null }, { "id": "478689", "postDate": "02/26/2019 13:43:58", "content": "

In Chinese idiom, it's Plug one's ears while stealing a bell.

", "rawMarkdown": "In Chinese idiom, it's Plug one's ears while stealing a bell.", "votes": null }, { "id": "478691", "postDate": "02/26/2019 13:44:19", "content": "

Upvoted here and there!

", "rawMarkdown": "Upvoted here and there!", "votes": null }, { "id": "478692", "postDate": "02/26/2019 13:46:29", "content": "

It's my pleasure that you are in kaggle. You are right, i see this, but don't try this and make a boost, actually hard for me to do the correct things.

", "rawMarkdown": "It's my pleasure that you are in kaggle. You are right, i see this, but don't try this and make a boost, actually hard for me to do the correct things.", "votes": null }, { "id": "478703", "postDate": "02/26/2019 14:01:03", "content": "

？？？

", "rawMarkdown": "？？？", "votes": null }, { "id": "478707", "postDate": "02/26/2019 14:08:08", "content": "

thank you!

", "rawMarkdown": "thank you!", "votes": null }, { "id": "478790", "postDate": "02/26/2019 16:41:56", "content": "

nobody missing anying, top2 team message was simple,' share ideas,but share it tomorrow‘

", "rawMarkdown": "nobody missing anying, top2 team message was simple,' share ideas,but share it tomorrow‘", "votes": null }, { "id": "478800", "postDate": "02/26/2019 16:52:34", "content": "

Hey all,

\n\n

We're aware of this. The Kaggle team does not repost deleted comments. Comments are deleted at the discretion of the user, and we do not reinstate them against that users wishes. Should the user want to repost it they can do so.

\n\n

In addition, as many of you have pointed out - we have our own systems in place to detect and remove cheaters, which occurs after the competition has concluded, which is why you'll see the total number of teams decrease.

", "rawMarkdown": "Hey all,\n\nWe're aware of this. The Kaggle team does not repost deleted comments. Comments are deleted at the discretion of the user, and we do not reinstate them against that users wishes. Should the user want to repost it they can do so.\n\nIn addition, as many of you have pointed out - we have our own systems in place to detect and remove cheaters, which occurs *after* the competition has concluded, which is why you'll see the total number of teams decrease.", "votes": null }, { "id": "478808", "postDate": "02/26/2019 17:02:34", "content": "

At a time when competition is over, revealing the tips or codes that are effective at the final rank is very unfair.\nAnd I think there should be some guidelines for it.\nAt least kernels or posts that contain these content can be viewed by the Kaggle team and removed promptly.

", "rawMarkdown": "At a time when competition is over, revealing the tips or codes that are effective at the final rank is very unfair.\nAnd I think there should be some guidelines for it.\nAt least kernels or posts that contain these content can be viewed by the Kaggle team and removed promptly.", "votes": null }, { "id": "478814", "postDate": "02/26/2019 17:06:19", "content": "

Allowing post deletion as you do is opening the door for lots problems. You don't think it is a problem that a guy in top LB shares his solution overview, then erase it? The erasure is the problem here.

", "rawMarkdown": "Allowing post deletion as you do is opening the door for lots problems. You don't think it is a problem that a guy in top LB shares his solution overview, then erase it? The erasure is the problem here.", "votes": null }, { "id": "478900", "postDate": "02/26/2019 19:13:47", "content": "

Someone shared with me the content:

\n\n
\n

Perhaps you have discovered that outliers have a very strong impact on the accuracy of the model. The first thing I am worried about is whether the distribution of outliers of the switched data is consistent with the current one. So I have to declare it in advance, and I am very worried that my post-processing method will cause me to fit on the existing list.

\n \n

Yep, I am going to talk about my scoring skills at the end of the game(now), and I don't guarantee the stability of this method. My score improvement comes from three parts. The first part is the grouping combination of multiple sets of features. The framework is as follows: (original feature + original stitching + feature filtering after stitching + fixed 300 features and tuning) * (lgb single model /nn single model /xgb-lgb-nn-catboost-ridge's Repeat (base model of BayesRidge)); the second part is to obtain two models for different models after obtaining 300 models, but I executed It is not \"binary\", but \"rank:pairwise\", making full use of the characteristics of \"Recommend\". After constructing the candidate set, you can fully interact with the -33 of the sample with the highest rank score. The third part, in fact, everyone has already Observed, -33 is actually log2(1e-10). Similarly, we can also find log(1e-n), n∈(1,10), and in this discrete interval, we can put the regression back. The value is corrected to a discrete value to reduce the prediction error. For example, we can correct the value of -31 to -33.

\n \n

In fact, the profit that rank:pairwise gives me is very high, but I have not found a very stable way to prevent overfitting in terms of threshold selection. If I am still in the gold medal area after switching data, I will share all my skills and handling in this question.

\n
", "rawMarkdown": "Someone shared with me the content:\n\n> Perhaps you have discovered that outliers have a very strong impact on the accuracy of the model. The first thing I am worried about is whether the distribution of outliers of the switched data is consistent with the current one. So I have to declare it in advance, and I am very worried that my post-processing method will cause me to fit on the existing list.\n\n> Yep, I am going to talk about my scoring skills at the end of the game(now), and I don't guarantee the stability of this method. My score improvement comes from three parts. The first part is the grouping combination of multiple sets of features. The framework is as follows: (original feature + original stitching + feature filtering after stitching + fixed 300 features and tuning) * (lgb single model /nn single model /xgb-lgb-nn-catboost-ridge's Repeat (base model of BayesRidge)); the second part is to obtain two models for different models after obtaining 300 models, but I executed It is not \"binary\", but \"rank:pairwise\", making full use of the characteristics of \"Recommend\". After constructing the candidate set, you can fully interact with the -33 of the sample with the highest rank score. The third part, in fact, everyone has already Observed, -33 is actually log2(1e-10). Similarly, we can also find log(1e-n), n∈(1,10), and in this discrete interval, we can put the regression back. The value is corrected to a discrete value to reduce the prediction error. For example, we can correct the value of -31 to -33.\n\n> In fact, the profit that rank:pairwise gives me is very high, but I have not found a very stable way to prevent overfitting in terms of threshold selection. If I am still in the gold medal area after switching data, I will share all my skills and handling in this question.", "votes": null }, { "id": "478926", "postDate": "02/26/2019 20:07:04", "content": "

next days i will chek this trand for today

", "rawMarkdown": "next days i will chek this [trand] for today \n\n[trand]: https://trends.google.com/trends/explore?date=today%205-y&q=rank%20pairwise", "votes": null }, { "id": "479063", "postDate": "02/27/2019 00:34:10", "content": "

I'm curious to know if this helped anyone.

", "rawMarkdown": "I'm curious to know if this helped anyone.", "votes": null }, { "id": "479067", "postDate": "02/27/2019 00:39:07", "content": "

No, the private will be better if no post processing.model stacking is also good.

", "rawMarkdown": "No, the private will be better if no post processing.model stacking is also good.", "votes": null }, { "id": "479080", "postDate": "02/27/2019 00:49:31", "content": "

yeah, indeed.

", "rawMarkdown": "yeah, indeed.", "votes": null }, { "id": "479113", "postDate": "02/27/2019 01:15:05", "content": "

Thanks for sharing the deleted post to you and the person who shared it you.. However can the deleted post be reposted by Kaggle now !!

", "rawMarkdown": "Thanks for sharing the deleted post to you and the person who shared it you.. However can the deleted post be reposted by Kaggle now !!", "votes": null }, { "id": "479123", "postDate": "02/27/2019 01:21:16", "content": "

I don't entirely understand a way around this. I know it is unfair to have disproportionate information sharing, but if someone accidentally posted a password or credit card number or something else potentially damaging they should have the right to delete. I agree it sucks but I don't think we can have Kaggle be set in stone as soon as a post is made.

", "rawMarkdown": "I don't entirely understand a way around this. I know it is unfair to have disproportionate information sharing, but if someone accidentally posted a password or credit card number or something else potentially damaging they should have the right to delete. I agree it sucks but I don't think we can have Kaggle be set in stone as soon as a post is made.", "votes": null }, { "id": "479160", "postDate": "02/27/2019 01:55:02", "content": "

The one delete the post is a nice guy and with strong technological strength. Sometimes we should be perceptual since it is really a bad experience to fall out of bronze medal from the 2nd place. If he wants, he can repost, if not, it is still understandable. See it or not see it is luck and luck is also the key to many competitions.

", "rawMarkdown": "The one delete the post is a nice guy and with strong technological strength. Sometimes we should be perceptual since it is really a bad experience to fall out of bronze medal from the 2nd place. If he wants, he can repost, if not, it is still understandable. See it or not see it is luck and luck is also the key to many competitions.", "votes": null }, { "id": "479183", "postDate": "02/27/2019 02:19:33", "content": "

蛇老太惨了

", "rawMarkdown": "蛇老太惨了", "votes": null }, { "id": "479221", "postDate": "02/27/2019 02:59:54", "content": "

Interesting. @gmhost, OOF stacking did not work for me this time. I got better result with simple averaging in stages.

", "rawMarkdown": "Interesting. @gmhost, OOF stacking did not work for me this time. I got better result with simple averaging in stages.", "votes": null }, { "id": "479229", "postDate": "02/27/2019 03:02:33", "content": "

simple linear stack, don't use gbm model for stack.:)

", "rawMarkdown": "simple linear stack, don't use gbm model for stack.:)", "votes": null }, { "id": "479238", "postDate": "02/27/2019 03:06:01", "content": "

Thanks @gmhost, I will try it later and do a late submission.

\n\n

Here is my post on feature selection techniques https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052

", "rawMarkdown": "Thanks @gmhost, I will try it later and do a late submission.\n\nHere is my post on feature selection techniques https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052", "votes": null }, { "id": "479241", "postDate": "02/27/2019 03:07:19", "content": "

Thanks for sharing!

", "rawMarkdown": "Thanks for sharing!", "votes": null }, { "id": "479673", "postDate": "02/27/2019 10:34:22", "content": "

@longyin2 I didn't know who posted it and certainly wasn't pointing fingers at him/she. I just wanted the post back given it had been seen by some competitors. The real issue is to allow post deletion.

\n\n

> See it or not see it is luck and luck is also the key to many competitions.

\n\n

Well, private sharing outside teaming is against competition rules. Once deleted, the post becomes private sharing between the author and those who read it. So, yes, this was breaking competition rules.

", "rawMarkdown": "longyin2 I didn't know who posted it and certainly wasn't pointing fingers at him/she. I just wanted the post back given it had been seen by some competitors. The real issue is to allow post deletion.\n\n> See it or not see it is luck and luck is also the key to many competitions.\n\nWell, private sharing outside teaming is against competition rules. Once deleted, the post becomes private sharing between the author and those who read it. So, yes, this was breaking competition rules.", "votes": null }, { "id": "479679", "postDate": "02/27/2019 10:36:02", "content": "

@addisonhoward Let me try another argument:

\n\n

Private sharing outside teaming is against competition rules. Once deleted, the post becomes private sharing between the author and those who read it, hence it is against the rules.

\n\n

I am not asking for any retaliation against the poster that deleted his post. I am asking that you take action to prevent this from happening again.

", "rawMarkdown": "addisonhoward Let me try another argument:\n\nPrivate sharing outside teaming is against competition rules. Once deleted, the post becomes private sharing between the author and those who read it, hence it is against the rules. \n\nI am not asking for any retaliation against the poster that deleted his post. I am asking that you take action to prevent this from happening again.", "votes": null }, { "id": "1120923", "postDate": "12/21/2020 07:52:06", "content": "

well done bro

", "rawMarkdown": "well done bro", "votes": null } ], "comments": [ { "id": 1120923, "author_name": "monish459", "author_url": "", "post_date": "12/21/2020 07:52:06", "content": "

well done bro

", "votes": null, "replies": [] }, { "id": 478649, "author_name": "yeonmin", "author_url": "", "post_date": "02/26/2019 12:52:45", "content": "

I advised that it would be nice to share it after the competition. I did not know if the publisher would delete the article.\nI was very embarrassed today. The Top2 Solution has come up before end of the competition and the ID of the competitor in Top10 has been deleted(170 submission).

", "votes": null, "replies": [ { "id": 478665, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 13:10:17", "content": "

You did not do anything wrong.

", "votes": null, "replies": [] }, { "id": 478667, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 13:11:11", "content": "
\n

The Top2 Solution has come up before end of the competition

\n
\n\n

Wow, that's a serious breach.

", "votes": null, "replies": [] }, { "id": 478671, "author_name": "sheriytm", "author_url": "", "post_date": "02/26/2019 13:17:31", "content": "
\n

... and the ID of the competitor in Top10 has been deleted(170 submission).

\n
\n\n

@yeomin, I don't think I understand what you mean. Can you elaborate?

", "votes": null, "replies": [] }, { "id": 478673, "author_name": "bolkonsky", "author_url": "", "post_date": "02/26/2019 13:19:02", "content": "

It was @cmsm1

", "votes": null, "replies": [] }, { "id": 478676, "author_name": "sheriytm", "author_url": "", "post_date": "02/26/2019 13:25:51", "content": "

Wow! @yeomin, you meant an actual solution. I have not seen that post. The post I referred to in my post below had a link to a feature selection code. That is why I came back to look for it but found that it is no longer there.

", "votes": null, "replies": [] }, { "id": 478683, "author_name": "fangwenjun", "author_url": "", "post_date": "02/26/2019 13:37:17", "content": "

I didn't do anything today but I advanced three places. very interesting

", "votes": null, "replies": [] }, { "id": 478684, "author_name": "yeonmin", "author_url": "", "post_date": "02/26/2019 13:39:23", "content": "

It was not exactly a solution, but in summary\n1. Simplified model usage information (xgb, nn, lgbm ridge, etc.), number of features used\n2. outlier model rank pairwise ensemble\n3. It seems to use the target value divided by discrete. The target value can be converted to log2 (x + 1e-10). It seemed here that 10 of 1e-10 was divided into discrete ranges from 1 to 10. For example, -31 to -33

", "votes": null, "replies": [] }, { "id": 478707, "author_name": "raddar", "author_url": "", "post_date": "02/26/2019 14:08:08", "content": "

thank you!

", "votes": null, "replies": [] }, { "id": 478790, "author_name": "youwbpyr", "author_url": "", "post_date": "02/26/2019 16:41:56", "content": "

nobody missing anying, top2 team message was simple,' share ideas,but share it tomorrow‘

", "votes": null, "replies": [] } ] }, { "id": 478651, "author_name": "sheriytm", "author_url": "", "post_date": "02/26/2019 12:53:31", "content": "

@cpmpml, are you refering to a post tiltled \"Something I want to say...\" ? If so, I saw it earlier and the whole discussion was all criptic and I wondered what the point of it was. The person said they will share tomorrow as of the time I read it. I do not know if more was discussed since I read it a few hours ago.

", "votes": null, "replies": [ { "id": 478664, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 13:09:56", "content": "

I don't know what I am referring to, that's the whole point of the discussion here.

", "votes": null, "replies": [] }, { "id": 478670, "author_name": "sheriytm", "author_url": "", "post_date": "02/26/2019 13:15:20", "content": "

@cpmpml, how did you know a post was deleted then? I mean how can one find out?

\n\n

Update:

\n\n

I forgot about the follow tab for the discussions. You probably got a notification as a result of following the discussion I guess.

", "votes": null, "replies": [] }, { "id": 478675, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 13:23:12", "content": "

I saw it mentioned in KggleNoobs slack.

", "votes": null, "replies": [] }, { "id": 478677, "author_name": "sheriytm", "author_url": "", "post_date": "02/26/2019 13:28:40", "content": "

I see. It looks like it is a different post from the one I was talking about. That post had a link to a feature selection code. That is why I came back to look for it but found that it is no longer there as well.

", "votes": null, "replies": [] } ] }, { "id": 478653, "author_name": "shaz13", "author_url": "", "post_date": "02/26/2019 12:57:52", "content": "

Edit history under \"Options\" button for original post is all we need.

", "votes": null, "replies": [ { "id": 478662, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 13:09:14", "content": "

Great idea, you should post it in the product feedback forum.

", "votes": null, "replies": [] }, { "id": 478687, "author_name": "shaz13", "author_url": "", "post_date": "02/26/2019 13:41:18", "content": "

Thanks!

\n\n

Made a thread request on the Product Forum

", "votes": null, "replies": [] }, { "id": 478691, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 13:44:19", "content": "

Upvoted here and there!

", "votes": null, "replies": [] } ] }, { "id": 478654, "author_name": "moussaid", "author_url": "", "post_date": "02/26/2019 13:00:57", "content": "

It's overfitting

", "votes": null, "replies": [ { "id": 478663, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 13:09:24", "content": "

What is overfitting?

", "votes": null, "replies": [] } ] }, { "id": 478681, "author_name": "gmhost", "author_url": "", "post_date": "02/26/2019 13:34:24", "content": "

And today there are so many deleted teams.:)

", "votes": null, "replies": [ { "id": 478685, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 13:39:49", "content": "

People may think it will prevent them from being caught cheating.

\n\n

But it won't :)

", "votes": null, "replies": [] }, { "id": 478689, "author_name": "gmhost", "author_url": "", "post_date": "02/26/2019 13:43:58", "content": "

In Chinese idiom, it's Plug one's ears while stealing a bell.

", "votes": null, "replies": [] }, { "id": 479080, "author_name": "zjucor", "author_url": "", "post_date": "02/27/2019 00:49:31", "content": "

yeah, indeed.

", "votes": null, "replies": [] } ] }, { "id": 478692, "author_name": "codlife", "author_url": "", "post_date": "02/26/2019 13:46:29", "content": "

It's my pleasure that you are in kaggle. You are right, i see this, but don't try this and make a boost, actually hard for me to do the correct things.

", "votes": null, "replies": [] }, { "id": 478703, "author_name": "niujiahe13", "author_url": "", "post_date": "02/26/2019 14:01:03", "content": "

？？？

", "votes": null, "replies": [] }, { "id": 478800, "author_name": "addisonhoward", "author_url": "", "post_date": "02/26/2019 16:52:34", "content": "

Hey all,

\n\n

We're aware of this. The Kaggle team does not repost deleted comments. Comments are deleted at the discretion of the user, and we do not reinstate them against that users wishes. Should the user want to repost it they can do so.

\n\n

In addition, as many of you have pointed out - we have our own systems in place to detect and remove cheaters, which occurs after the competition has concluded, which is why you'll see the total number of teams decrease.

", "votes": null, "replies": [ { "id": 478814, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 17:06:19", "content": "

Allowing post deletion as you do is opening the door for lots problems. You don't think it is a problem that a guy in top LB shares his solution overview, then erase it? The erasure is the problem here.

", "votes": null, "replies": [] }, { "id": 479679, "author_name": "cpmpml", "author_url": "", "post_date": "02/27/2019 10:36:02", "content": "

@addisonhoward Let me try another argument:

\n\n

Private sharing outside teaming is against competition rules. Once deleted, the post becomes private sharing between the author and those who read it, hence it is against the rules.

\n\n

I am not asking for any retaliation against the poster that deleted his post. I am asking that you take action to prevent this from happening again.

", "votes": null, "replies": [] } ] }, { "id": 478808, "author_name": "mjbahmani", "author_url": "", "post_date": "02/26/2019 17:02:34", "content": "

At a time when competition is over, revealing the tips or codes that are effective at the final rank is very unfair.\nAnd I think there should be some guidelines for it.\nAt least kernels or posts that contain these content can be viewed by the Kaggle team and removed promptly.

", "votes": null, "replies": [] }, { "id": 478900, "author_name": "cpmpml", "author_url": "", "post_date": "02/26/2019 19:13:47", "content": "

Someone shared with me the content:

\n\n
\n

Perhaps you have discovered that outliers have a very strong impact on the accuracy of the model. The first thing I am worried about is whether the distribution of outliers of the switched data is consistent with the current one. So I have to declare it in advance, and I am very worried that my post-processing method will cause me to fit on the existing list.

\n \n

Yep, I am going to talk about my scoring skills at the end of the game(now), and I don't guarantee the stability of this method. My score improvement comes from three parts. The first part is the grouping combination of multiple sets of features. The framework is as follows: (original feature + original stitching + feature filtering after stitching + fixed 300 features and tuning) * (lgb single model /nn single model /xgb-lgb-nn-catboost-ridge's Repeat (base model of BayesRidge)); the second part is to obtain two models for different models after obtaining 300 models, but I executed It is not \"binary\", but \"rank:pairwise\", making full use of the characteristics of \"Recommend\". After constructing the candidate set, you can fully interact with the -33 of the sample with the highest rank score. The third part, in fact, everyone has already Observed, -33 is actually log2(1e-10). Similarly, we can also find log(1e-n), n∈(1,10), and in this discrete interval, we can put the regression back. The value is corrected to a discrete value to reduce the prediction error. For example, we can correct the value of -31 to -33.

\n \n

In fact, the profit that rank:pairwise gives me is very high, but I have not found a very stable way to prevent overfitting in terms of threshold selection. If I am still in the gold medal area after switching data, I will share all my skills and handling in this question.

\n
", "votes": null, "replies": [] }, { "id": 478926, "author_name": "steubk", "author_url": "", "post_date": "02/26/2019 20:07:04", "content": "

next days i will chek this trand for today

", "votes": null, "replies": [] }, { "id": 479063, "author_name": "cpmpml", "author_url": "", "post_date": "02/27/2019 00:34:10", "content": "

I'm curious to know if this helped anyone.

", "votes": null, "replies": [ { "id": 479067, "author_name": "gmhost", "author_url": "", "post_date": "02/27/2019 00:39:07", "content": "

No, the private will be better if no post processing.model stacking is also good.

", "votes": null, "replies": [] }, { "id": 479221, "author_name": "sheriytm", "author_url": "", "post_date": "02/27/2019 02:59:54", "content": "

Interesting. @gmhost, OOF stacking did not work for me this time. I got better result with simple averaging in stages.

", "votes": null, "replies": [] }, { "id": 479229, "author_name": "gmhost", "author_url": "", "post_date": "02/27/2019 03:02:33", "content": "

simple linear stack, don't use gbm model for stack.:)

", "votes": null, "replies": [] }, { "id": 479238, "author_name": "sheriytm", "author_url": "", "post_date": "02/27/2019 03:06:01", "content": "

Thanks @gmhost, I will try it later and do a late submission.

\n\n

Here is my post on feature selection techniques https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052

", "votes": null, "replies": [] }, { "id": 479241, "author_name": "gmhost", "author_url": "", "post_date": "02/27/2019 03:07:19", "content": "

Thanks for sharing!

", "votes": null, "replies": [] } ] }, { "id": 479113, "author_name": "viswanathravindran", "author_url": "", "post_date": "02/27/2019 01:15:05", "content": "

Thanks for sharing the deleted post to you and the person who shared it you.. However can the deleted post be reposted by Kaggle now !!

", "votes": null, "replies": [] }, { "id": 479123, "author_name": "ryches", "author_url": "", "post_date": "02/27/2019 01:21:16", "content": "

I don't entirely understand a way around this. I know it is unfair to have disproportionate information sharing, but if someone accidentally posted a password or credit card number or something else potentially damaging they should have the right to delete. I agree it sucks but I don't think we can have Kaggle be set in stone as soon as a post is made.

", "votes": null, "replies": [] }, { "id": 479160, "author_name": "longyin2", "author_url": "", "post_date": "02/27/2019 01:55:02", "content": "

The one delete the post is a nice guy and with strong technological strength. Sometimes we should be perceptual since it is really a bad experience to fall out of bronze medal from the 2nd place. If he wants, he can repost, if not, it is still understandable. See it or not see it is luck and luck is also the key to many competitions.

", "votes": null, "replies": [ { "id": 479183, "author_name": "youwbpyr", "author_url": "", "post_date": "02/27/2019 02:19:33", "content": "

蛇老太惨了

", "votes": null, "replies": [] }, { "id": 479673, "author_name": "cpmpml", "author_url": "", "post_date": "02/27/2019 10:34:22", "content": "

@longyin2 I didn't know who posted it and certainly wasn't pointing fingers at him/she. I just wanted the post back given it had been seen by some competitors. The real issue is to allow post deletion.

\n\n

> See it or not see it is luck and luck is also the key to many competitions.

\n\n

Well, private sharing outside teaming is against competition rules. Once deleted, the post becomes private sharing between the author and those who read it. So, yes, this was breaking competition rules.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "478593": "To Kaggle team, this is really bad and unfair.\n\nApparently, someone shared interesting info few hours ago, then got advice to have not posted it before competition end, then the post was deleted.\n\nDeleting something of value is even worse than publishing it. Indeed, it creates unfairness between those who could see it and those who could not.\n\nThe post should be reposted IMHO. If someone got it via notifications in email, then please repost.\n\nEdited: link to the deleted post: https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/81915\n\nPS. I'm not worried about my rank given how low it is. But this may impact top of the LB, which is really bad.", "478649": "I advised that it would be nice to share it after the competition. I did not know if the publisher would delete the article.\nI was very embarrassed today. The Top2 Solution has come up before end of the competition and the ID of the competitor in Top10 has been deleted(170 submission).", "478651": "cpmpml, are you refering to a post tiltled \"Something I want to say...\" ? If so, I saw it earlier and the whole discussion was all criptic and I wondered what the point of it was. The person said they will share tomorrow as of the time I read it. I do not know if more was discussed since I read it a few hours ago.", "478653": "Edit history under \"Options\" button for original post is all we need.", "478654": "It's overfitting", "478662": "Great idea, you should post it in the product feedback forum.", "478663": "What is overfitting?", "478664": "I don't know what I am referring to, that's the whole point of the discussion here.", "478665": "You did not do anything wrong.", "478667": "> The Top2 Solution has come up before end of the competition \n\nWow, that's a serious breach.", "478670": "cpmpml, how did you know a post was deleted then? I mean how can one find out?\n\nUpdate:\n\nI forgot about the follow tab for the discussions. You probably got a notification as a result of following the discussion I guess.", "478671": ">... and the ID of the competitor in Top10 has been deleted(170 submission).\n\n@yeomin, I don't think I understand what you mean. Can you elaborate?", "478673": "It was @cmsm1", "478675": "I saw it mentioned in KggleNoobs slack.", "478676": "Wow! @yeomin, you meant an actual solution. I have not seen that post. The post I referred to in my post below had a link to a feature selection code. That is why I came back to look for it but found that it is no longer there.", "478677": "I see. It looks like it is a different post from the one I was talking about. That post had a link to a feature selection code. That is why I came back to look for it but found that it is no longer there as well.", "478681": "And today there are so many deleted teams.:)", "478683": "I didn't do anything today but I advanced three places. very interesting", "478684": "It was not exactly a solution, but in summary\n1. Simplified model usage information (xgb, nn, lgbm ridge, etc.), number of features used\n2. outlier model rank pairwise ensemble\n3. It seems to use the target value divided by discrete. The target value can be converted to log2 (x + 1e-10). It seemed here that 10 of 1e-10 was divided into discrete ranges from 1 to 10. For example, -31 to -33", "478685": "People may think it will prevent them from being caught cheating.\n\nBut it won't :)", "478687": "Thanks!\n\nMade a [thread](https://www.kaggle.com/product-feedback/81968) request on the Product Forum", "478689": "In Chinese idiom, it's Plug one's ears while stealing a bell.", "478691": "Upvoted here and there!", "478692": "It's my pleasure that you are in kaggle. You are right, i see this, but don't try this and make a boost, actually hard for me to do the correct things.", "478703": "？？？", "478707": "thank you!", "478790": "nobody missing anying, top2 team message was simple,' share ideas,but share it tomorrow‘", "478800": "Hey all,\n\nWe're aware of this. The Kaggle team does not repost deleted comments. Comments are deleted at the discretion of the user, and we do not reinstate them against that users wishes. Should the user want to repost it they can do so.\n\nIn addition, as many of you have pointed out - we have our own systems in place to detect and remove cheaters, which occurs *after* the competition has concluded, which is why you'll see the total number of teams decrease.", "478808": "At a time when competition is over, revealing the tips or codes that are effective at the final rank is very unfair.\nAnd I think there should be some guidelines for it.\nAt least kernels or posts that contain these content can be viewed by the Kaggle team and removed promptly.", "478814": "Allowing post deletion as you do is opening the door for lots problems. You don't think it is a problem that a guy in top LB shares his solution overview, then erase it? The erasure is the problem here.", "478900": "Someone shared with me the content:\n\n> Perhaps you have discovered that outliers have a very strong impact on the accuracy of the model. The first thing I am worried about is whether the distribution of outliers of the switched data is consistent with the current one. So I have to declare it in advance, and I am very worried that my post-processing method will cause me to fit on the existing list.\n\n> Yep, I am going to talk about my scoring skills at the end of the game(now), and I don't guarantee the stability of this method. My score improvement comes from three parts. The first part is the grouping combination of multiple sets of features. The framework is as follows: (original feature + original stitching + feature filtering after stitching + fixed 300 features and tuning) * (lgb single model /nn single model /xgb-lgb-nn-catboost-ridge's Repeat (base model of BayesRidge)); the second part is to obtain two models for different models after obtaining 300 models, but I executed It is not \"binary\", but \"rank:pairwise\", making full use of the characteristics of \"Recommend\". After constructing the candidate set, you can fully interact with the -33 of the sample with the highest rank score. The third part, in fact, everyone has already Observed, -33 is actually log2(1e-10). Similarly, we can also find log(1e-n), n∈(1,10), and in this discrete interval, we can put the regression back. The value is corrected to a discrete value to reduce the prediction error. For example, we can correct the value of -31 to -33.\n\n> In fact, the profit that rank:pairwise gives me is very high, but I have not found a very stable way to prevent overfitting in terms of threshold selection. If I am still in the gold medal area after switching data, I will share all my skills and handling in this question.", "478926": "next days i will chek this [trand] for today \n\n[trand]: https://trends.google.com/trends/explore?date=today%205-y&q=rank%20pairwise", "479063": "I'm curious to know if this helped anyone.", "479067": "No, the private will be better if no post processing.model stacking is also good.", "479080": "yeah, indeed.", "479113": "Thanks for sharing the deleted post to you and the person who shared it you.. However can the deleted post be reposted by Kaggle now !!", "479123": "I don't entirely understand a way around this. I know it is unfair to have disproportionate information sharing, but if someone accidentally posted a password or credit card number or something else potentially damaging they should have the right to delete. I agree it sucks but I don't think we can have Kaggle be set in stone as soon as a post is made.", "479160": "The one delete the post is a nice guy and with strong technological strength. Sometimes we should be perceptual since it is really a bad experience to fall out of bronze medal from the 2nd place. If he wants, he can repost, if not, it is still understandable. See it or not see it is luck and luck is also the key to many competitions.", "479183": "蛇老太惨了", "479221": "Interesting. @gmhost, OOF stacking did not work for me this time. I got better result with simple averaging in stages.", "479229": "simple linear stack, don't use gbm model for stack.:)", "479238": "Thanks @gmhost, I will try it later and do a late submission.\n\nHere is my post on feature selection techniques https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/82052", "479241": "Thanks for sharing!", "479673": "longyin2 I didn't know who posted it and certainly wasn't pointing fingers at him/she. I just wanted the post back given it had been seen by some competitors. The real issue is to allow post deletion.\n\n> See it or not see it is luck and luck is also the key to many competitions.\n\nWell, private sharing outside teaming is against competition rules. Once deleted, the post becomes private sharing between the author and those who read it. So, yes, this was breaking competition rules.", "479679": "addisonhoward Let me try another argument:\n\nPrivate sharing outside teaming is against competition rules. Once deleted, the post becomes private sharing between the author and those who read it, hence it is against the rules. \n\nI am not asking for any retaliation against the poster that deleted his post. I am asking that you take action to prevent this from happening again.", "1120923": "well done bro" }, "source": "meta" }