You are working on your ideas putting in effort on your own, stars get aligned and rank closes in towards gold and then... A high scoring kernel is published all hard work goes in vain as a lot of people just fork and submit. All done in not more than three clicks.
\n\nThere should be some sort of honor code and ethics regarding this, I wish if kernels could contain complete EDA, visualizations and summaries with code but only clear, concise step by step hints given for modelling.
\n\nHave to start all over again.
", "messages": [ { "id": 547378, "postDate": "2019-06-07T16:28:59.787Z", "content": "This competition started 9 days ago! Only 9 days. Within 2 months these high scores will seem to be quite low.\nAlso I want to repeat a point, which I often say: if I'm in a medal range and there is a public kernel with a higher score, then I don't deserve this medal.\nBut I can use this public kernel to improve my own solution.
", "rawMarkdown": "This competition started 9 days ago! Only 9 days. Within 2 months these high scores will seem to be quite low.\nAlso I want to repeat a point, which I often say: if I'm in a medal range and there is a public kernel with a higher score, then I don't deserve this medal.\nBut I can use this public kernel to improve my own solution.", "votes": 19, "replies": [ { "id": 547501, "postDate": "2019-06-07T20:05:47.653Z", "content": "Yes and no. All your hard work has put you in the medal position but when a hundred people swoop in and use the kernel to just generate the same submissions and push you out even though they did little work, that is frustrating. I left the competitions for a while after that happened to me. No, I don't take them a serious because I figure it can happen just as easily again.
", "rawMarkdown": "Yes and no. All your hard work has put you in the medal position but when a hundred people swoop in and use the kernel to just generate the same submissions and push you out even though they did little work, that is frustrating. I left the competitions for a while after that happened to me. No, I don't take them a serious because I figure it can happen just as easily again.", "votes": 1 }, { "id": 547650, "postDate": "2019-06-08T04:14:52.393Z", "content": "@artgor your kernels are always thoughtful, well explained and give credit where you have used others ideas and cannot recall one that is beyond bronze or lower scores by the end anyways. And early in competitions, much is learned from kernels, discussions, links. But have seen your work included elsewhere without a simple paste of the link to your kernel to credit the source. That seems far worse to me than what a public kernel scores.
", "rawMarkdown": "@artgor your kernels are always thoughtful, well explained and give credit where you have used others ideas and cannot recall one that is beyond bronze or lower scores by the end anyways. And early in competitions, much is learned from kernels, discussions, links. But have seen your work included elsewhere without a simple paste of the link to your kernel to credit the source. That seems far worse to me than what a public kernel scores. ", "votes": 1 }, { "id": 547665, "postDate": "2019-06-08T05:12:56.013Z", "content": "@cbryant well, if this new kernel is different from your solution, you can use it to improve your own score, right?
\n\n@something4kag I got used to the fact that some people simply copy kernels, little can be done about it sadly.
", "rawMarkdown": "@cbryant well, if this new kernel is different from your solution, you can use it to improve your own score, right?\n\n@something4kag I got used to the fact that some people simply copy kernels, little can be done about it sadly.", "votes": 1 }, { "id": 548029, "postDate": "2019-06-08T16:45:54.107Z", "content": "thanks Andrew for sane voice
", "rawMarkdown": "thanks Andrew for sane voice", "votes": 1 }, { "id": 549133, "postDate": "2019-06-10T10:28:14.703Z", "content": "First of all a huge thanks to you for all the amazing work you have done so far in this competition and the future work.
\n\n\n\n\nI often say: if I'm in a medal range and there is a public kernel with a higher score, then I don't deserve this medal.
\n
This shows your high credibility which is present in very few these days. I follow your work very closely and found it very inspiring. In fact I have a list of all the kernels you have ever written on kaggle and I learn from them. Infact I am modifying one of your kernels. So there is no doubt that a kernel is needed as part of learning on Kaggle
\n\nNow what my point is there are users who never look into kernels and all they do is three clicks to get a position on LB. This practice is OK for beginners but it is expected that eventually at some stage they would start to write their own kernels and original stuff but they never do that it results into an increase in number of those who fall in not more than three clicks category now the people who were writing their own kernels don't get what they deserved.
This has been a topic of debate for a long time on kaggle not just us having this conversation. Probably in future I would love to see something in which people who don't do more than three clicks to reach LB don't get medals.
", "rawMarkdown": "First of all a huge thanks to you for all the amazing work you have done so far in this competition and the future work. \n> I often say: if I'm in a medal range and there is a public kernel with a higher score, then I don't deserve this medal.\n\nThis shows your high credibility which is present in very few these days. I follow your work very closely and found it very inspiring. In fact I have a list of all the kernels you have ever written on kaggle and I learn from them. Infact I am modifying one of your kernels. **So there is no doubt that a kernel is needed as part of learning on Kaggle**\n\nNow what my point is there are users who never look into kernels and all they do is three clicks to get a position on LB. This practice is OK for beginners but it is expected that eventually at some stage they would start to write their own kernels and original stuff but they never do that it results into an increase in number of those who fall in `not more than three clicks category` now the people who were writing their own kernels don't get what they deserved.\n\nThis has been a topic of debate for a long time on kaggle not just us having this conversation. Probably in future I would love to see something in which people who don't do more than three clicks to reach LB don't get medals." }, { "id": 549141, "postDate": "2019-06-10T10:48:24.317Z", "content": "Thank you for your kind words!
\n\nI agree that if people can get medals without any efforts it is bad. It diminishes the value of medals and discourages honest people to take part in further competition.
\n\nBut I want to make several points:\n- lately getting silver medals becomes more and more difficult, often by the end of a competition there are no public kernels in silver range;\n- publishing high scoring kernels is prohibited within last week of a competition, so most top teams will exceed scores of public kernels;\n- I'm sure that in this particular competition it would be very difficult to get even silver medal by training models only on kernels. The datafiles in this competition are quite big, if you generate several hundreds of features, you could hit memory limit. Also it is obvious that approaches using graph networks and graph neural nets will be very useful and I'm not sure whether it is possible to implement these approaches in kernels with good enough quality. So I think that most of people at the top of the leaderboard will use their own hardware and not kaggle kernels.
\n\nStill, despite these points I want to repeat that getting medals without efforts is bad.
", "rawMarkdown": "Thank you for your kind words!\n\nI agree that if people can get medals without any efforts it is bad. It diminishes the value of medals and discourages honest people to take part in further competition.\n\nBut I want to make several points:\n- lately getting silver medals becomes more and more difficult, often by the end of a competition there are no public kernels in silver range;\n- publishing high scoring kernels is prohibited within last week of a competition, so most top teams will exceed scores of public kernels;\n- I'm sure that in this particular competition it would be very difficult to get even silver medal by training models only on kernels. The datafiles in this competition are quite big, if you generate several hundreds of features, you could hit memory limit. Also it is obvious that approaches using graph networks and graph neural nets will be very useful and I'm not sure whether it is possible to implement these approaches in kernels with good enough quality. So I think that most of people at the top of the leaderboard will use their own hardware and not kaggle kernels.\n\nStill, despite these points I want to repeat that getting medals without efforts is bad.", "votes": 5 }, { "id": 583684, "postDate": "2019-07-24T20:27:05.250Z", "content": "I don’t agree. Imagine if my team publish a -2.1 kernel, then 100 teams below you will jump up. Then you will understand that you’re wrong.
", "rawMarkdown": "I don’t agree. Imagine if my team publish a -2.1 kernel, then 100 teams below you will jump up. Then you will understand that you’re wrong. " }, { "id": 583820, "postDate": "2019-07-25T03:20:08.737Z", "content": "No, I'd be glad that there is a new public kernel with unique approach/features, which we could analyze and add to our solution.
", "rawMarkdown": "No, I'd be glad that there is a new public kernel with unique approach/features, which we could analyze and add to our solution." } ] }, { "id": 548167, "postDate": "2019-06-08T22:55:27.023Z", "content": "Have you asked yourself why you are spending your precious time on these Kaggle competitions instead of any other alternatives? If you are after the prize money, good luck with that if you can't even beat a public kernel. If you like to hoard internet points, there are many alternative sites that you don't have to work so hard.
\n\nIf like most of us, you are trying to learn new things, you should probably be thankful that someone spent their time sharing their knowledge so that you can learn from them, especially if they are better than you.
", "rawMarkdown": "Have you asked yourself why you are spending your precious time on these Kaggle competitions instead of any other alternatives? If you are after the prize money, good luck with that if you can't even beat a public kernel. If you like to hoard internet points, there are many alternative sites that you don't have to work so hard.\n\nIf like most of us, you are trying to learn new things, you should probably be thankful that someone spent their time sharing their knowledge so that you can learn from them, especially if they are better than you.", "votes": 17 }, { "id": 547274, "postDate": "2019-06-07T13:42:14.857Z", "content": "There are 3 months left in the competition, don't worry. This will not be the highest scoring kernel. People will innovate and beat it, especially you.
", "rawMarkdown": "There are 3 months left in the competition, don't worry. This will not be the highest scoring kernel. People will innovate and beat it, especially you.", "votes": 10, "replies": [ { "id": 547278, "postDate": "2019-06-07T13:58:02.020Z", "content": "I agree. its too early for not to share.Otherwise we would not able to see the true potential of the data.
", "rawMarkdown": "I agree. its too early for not to share.Otherwise we would not able to see the true potential of the data.", "votes": 2 } ] }, { "id": 548026, "postDate": "2019-06-08T16:43:09.547Z", "content": "I downvoted you. I don't understand people who upvoted you.
\n\nWhat are you talking about?
\n\nCompetition just started, I am just getting familar with the data. And you already did \"all hard work\"? When?
\n\nKaggle is for learning, I think some people don't understand that at all.
", "rawMarkdown": "I downvoted you. I don't understand people who upvoted you.\n\nWhat are you talking about?\n\nCompetition just started, I am just getting familar with the data. And you already did \"all hard work\"? When?\n\nKaggle is for learning, I think some people don't understand that at all.\n\n\n\n\n", "votes": 7, "replies": [ { "id": 549119, "postDate": "2019-06-10T10:07:23.200Z", "content": "I upvoted you. Why? because of the following
\n\n\nWhat are you talking about?
Issue comes when there are some who did not even looked at the kernel, clicked not more than thrice and are on better position than those who have invested time and energy to get a good position.
\n\nI am actually learning a lot from Andrew Lukyanenko in this competition. And his contribution in this competition is unparallel
\n\n\n\n\nI don't understand people who upvoted you.
\n
They are the ones who have at some stage felt the pain of working on something and seeing that going in vain and most importantly look at big picture of machine learning as a whole not limited to any competition hosting site.
\n\nRemember : All models are same. No model is better than other but kaggle makes them appear better than other by ranking and LB. Sometimes it is hard to help people realize what they are losing by following an approach.
\n\nI agree with most of your points. Wish you luck.
", "rawMarkdown": "I upvoted you. Why? because of the following\n\n1. You showed good communication skills you asked \n> What are you talking about?\n\nI as others want to improve my tier. I worked hard to find a way to improve my score on LB but someone who was more skillful than me released a kernel. Now so far this is perfectly fine as I learnt a lot from the kernel released I spent time to go through the kernel and trying to add the knowledge I have on top of public kernel to improve my score now others who do this are perfectly fine. I have no issues so far. \n\nIssue comes when there are some who did not even looked at the kernel, **clicked not more than thrice** and are on better position than those who have invested time and energy to get a good position. \n\n**I am actually learning a lot from Andrew Lukyanenko in this competition. And his contribution in this competition is unparallel**\n\n> I don't understand people who upvoted you.\n\nThey are the ones who have at some stage felt the pain of working on something and seeing that going in vain and most importantly look at big picture of machine learning as a whole not limited to any competition hosting site. \n\nRemember : All models are same. No model is better than other but kaggle makes them appear better than other by ranking and LB. Sometimes it is hard to help people realize what they are losing by following an approach.\n\nI agree with most of your points. Wish you luck.", "votes": 2 } ] }, { "id": 547834, "postDate": "2019-06-08T11:14:23.563Z", "content": "The greatest value of these competitions is learning. For that purpose public kernels and discussions can be very useful. Medals should not be viewed as the goal but as the consequence of learning :)
", "rawMarkdown": "The greatest value of these competitions is learning. For that purpose public kernels and discussions can be very useful. Medals should not be viewed as the goal but as the consequence of learning :) ", "votes": 5 }, { "id": 547393, "postDate": "2019-06-07T16:52:19.873Z", "content": "If you did something original, then high scoring kernels are also useful for you. You can add features or methods from another kernel to your solution and gain your score. At least you can simply blend submissions.
", "rawMarkdown": "If you did something original, then high scoring kernels are also useful for you. You can add features or methods from another kernel to your solution and gain your score. At least you can simply blend submissions.", "votes": 5 }, { "id": 548659, "postDate": "2019-06-09T17:12:21.037Z", "content": "As mentioned in the comments a lot of us are here to learn. For a Kaggle beginner like me, having this kind of kernels helps me understand how other people approach these modelling \"puzzles\". Plenty of time left for you to improve your kernel and who knows you might find a solution in another public kernel that helps win the competition!
", "rawMarkdown": "As mentioned in the comments a lot of us are here to learn. For a Kaggle beginner like me, having this kind of kernels helps me understand how other people approach these modelling \"puzzles\". Plenty of time left for you to improve your kernel and who knows you might find a solution in another public kernel that helps win the competition!", "votes": 4 }, { "id": 547254, "postDate": "2019-06-07T12:55:55.603Z", "content": "You are working on your ideas putting in effort on your own, stars get aligned and rank closes in towards gold and then... A high scoring kernel is published all hard work goes in vain as a lot of people just fork and submit. All done in not more than three clicks.
\n\nThere should be some sort of honor code and ethics regarding this, I wish if kernels could contain complete EDA, visualizations and summaries with code but only clear, concise step by step hints given for modelling.
\n\nHave to start all over again.
", "rawMarkdown": "You are working on your ideas putting in effort on your own, stars get aligned and rank closes in towards gold and then... A high scoring kernel is published all hard work goes in vain as a lot of people just fork and submit. All done in not more than three clicks.\n\nThere should be some sort of honor code and ethics regarding this, I wish if kernels could contain complete EDA, visualizations and summaries with code but only clear, concise step by step hints given for modelling.\n\nHave to start all over again.", "votes": -15 }, { "id": 547656, "postDate": "2019-06-08T04:37:52.633Z", "content": "Exactly what I was thinking about. I want to make my kernel public, but then this( forking and submitting ) stops me. But it seems it's ok to share at this early stage of the competition and I still can't decide ( someone help me ). I want to have some medals so that I can also be an expert on kaggle :)
", "rawMarkdown": "Exactly what I was thinking about. I want to make my kernel public, but then this( forking and submitting ) stops me. But it seems it's ok to share at this early stage of the competition and I still can't decide ( someone help me ). I want to have some medals so that I can also be an expert on kaggle :) ", "votes": 2 }, { "id": 547264, "postDate": "2019-06-07T13:13:48.303Z", "content": "I guess it's in human nature to \"learn\" from someone else's work to make your product better. Isn't this how someone came up with the concept of Transfer Learning? I get it that copying & submitting a high scoring public kernel on the leaderboard is practically just cheating but you could use that to your advantage! All that I'm trying to convey here is that why not try gaining some insights from the high scoring kernels, adding your own spice and ingredients to try and come up with a better model?
\n\nDon't get me wrong here, I'm just trying to be optimistic and prefer to look out-of-the-box to beat the ones who take the easy way out.
", "rawMarkdown": "I guess it's in human nature to \"*learn*\" from someone else's work to make your product better. Isn't this how someone came up with the concept of *Transfer Learning*? I get it that copying & submitting a high scoring public kernel on the leaderboard is practically just cheating but you could use that to your advantage! All that I'm trying to convey here is that why not try gaining some insights from the high scoring kernels, adding your own spice and ingredients to try and come up with a better model?\n\nDon't get me wrong here, I'm just trying to be optimistic and prefer to look out-of-the-box to beat the ones who take the easy way out.", "votes": 2 }, { "id": 549224, "postDate": "2019-06-10T12:38:28.040Z", "content": "lol
", "rawMarkdown": "lol", "votes": 1 }, { "id": 547694, "postDate": "2019-06-08T06:10:03.993Z", "content": "yes,exactly..i believe public kernels should contain only baseline models.Tuning should be done in private mode,so that this issue won't happen.But this becomes a limiting factor for kernel authors as kernel having best score gets more views and upvotes.
", "rawMarkdown": "yes,exactly..i believe public kernels should contain only baseline models.Tuning should be done in private mode,so that this issue won't happen.But this becomes a limiting factor for kernel authors as kernel having best score gets more views and upvotes.", "votes": -1, "replies": [ { "id": 548028, "postDate": "2019-06-08T16:45:09.490Z", "content": "\n\n\ni believe public kernels should contain only baseline models.
\n
what happened to Kaggle? I dont't believe what I am reading
", "rawMarkdown": "> i believe public kernels should contain only baseline models.\n\nwhat happened to Kaggle? I dont't believe what I am reading", "votes": 3 }, { "id": 548427, "postDate": "2019-06-09T11:19:47.600Z", "content": "I think that Kaggle need a makes special block for education with kernels, with explain all problem.
", "rawMarkdown": "I think that Kaggle need a makes special block for education with kernels, with explain all problem. " } ] }, { "id": 583463, "postDate": "2019-07-24T14:09:52.030Z", "content": "Wow, -1.5 in public kernels now
", "rawMarkdown": "Wow, -1.5 in public kernels now", "replies": [ { "id": 583467, "postDate": "2019-07-24T14:14:13.133Z", "content": "Yeah, they got from -1.3 to -1.5 with blending. There is no new public kernel which would actually create a model like for two weeks or more.
", "rawMarkdown": "Yeah, they got from -1.3 to -1.5 with blending. There is no new public kernel which would actually create a model like for two weeks or more." }, { "id": 583638, "postDate": "2019-07-24T18:23:50.037Z", "content": "It's surprising for me. Probably there are some new features for last 2 weeks?! Because such a boost > 0.1 is high for blending, I think.
", "rawMarkdown": "It's surprising for me. Probably there are some new features for last 2 weeks?! Because such a boost > 0.1 is high for blending, I think." } ] }, { "id": 547294, "postDate": "2019-06-07T14:26:16.280Z", "rawMarkdown": "", "votes": 13, "isDeleted": true, "replies": [ { "id": 547299, "postDate": "2019-06-07T14:33:47.643Z", "content": "No one could put it as elegantly as you have. 👍
", "rawMarkdown": "No one could put it as elegantly as you have. 👍" }, { "id": 547352, "postDate": "2019-06-07T16:00:07.917Z", "content": "It is not clear to me that the sponsor actually benefits from the excessive sharing of high-scoring kernels. It hurts the sponsor by damping the motivation of Kagglers and thus reducing the variety in approaches.
", "rawMarkdown": "It is not clear to me that the sponsor actually benefits from the excessive sharing of high-scoring kernels. It hurts the sponsor by damping the motivation of Kagglers and thus reducing the variety in approaches.", "votes": 1 }, { "id": 547354, "postDate": "2019-06-07T16:01:48.897Z", "content": "I did see the warning message in the earthquake competition. It must be implemented already.
", "rawMarkdown": "I did see the warning message in the earthquake competition. It must be implemented already." }, { "id": 549087, "postDate": "2019-06-10T09:31:38.260Z", "content": "\n\n\nIt is not clear to me that the sponsor actually benefits from the excessive sharing of high-scoring kernels
\n
Good point. When it comes to hosting a competition. The goal is to crowdsource the problem (developing a model) which could be used by the host/sponsor to generate value by deploying it for their business. If there is not even one evidence/user story of a sponsor putting the winning solution into production in the entire history of Kaggle. You should be able to understand the hidden picture no one talks about or the one who talks about get downvotes 😄 .
\n\nIf anyone knows of any resource on internet where it is mentioned that sponsor got anything out of a competition which was realistic. Please let me know, I would be highly thankful for that, I have searched but never found anything.
", "rawMarkdown": "> It is not clear to me that the sponsor actually benefits from the excessive sharing of high-scoring kernels\n\nGood point. When it comes to hosting a competition. The goal is to crowdsource the problem (developing a model) which could be used by the host/sponsor to generate value by deploying it for their business. If there is not even one evidence/user story of a sponsor putting the winning solution into production in the entire history of Kaggle. You should be able to understand the hidden picture no one talks about or the one who talks about get downvotes 😄 . \n\nIf anyone knows of any resource on internet where it is mentioned that sponsor got anything out of a competition which was realistic. Please let me know, I would be highly thankful for that, I have searched but never found anything.", "votes": 1 } ] } ], "comments": [ { "id": 547378, "author_name": "Andrey Lukyanenko", "author_url": "", "post_date": "2019-06-07T16:28:59.787000", "content": "This competition started 9 days ago! Only 9 days. Within 2 months these high scores will seem to be quite low.\nAlso I want to repeat a point, which I often say: if I'm in a medal range and there is a public kernel with a higher score, then I don't deserve this medal.\nBut I can use this public kernel to improve my own solution.
", "votes": 19, "replies": [ { "id": 547501, "author_name": "TheGruffalo", "author_url": "", "post_date": "2019-06-07T20:05:47.653000", "content": "Yes and no. All your hard work has put you in the medal position but when a hundred people swoop in and use the kernel to just generate the same submissions and push you out even though they did little work, that is frustrating. I left the competitions for a while after that happened to me. No, I don't take them a serious because I figure it can happen just as easily again.
", "votes": 1, "replies": [] }, { "id": 547650, "author_name": "something4kag", "author_url": "", "post_date": "2019-06-08T04:14:52.393000", "content": "@artgor your kernels are always thoughtful, well explained and give credit where you have used others ideas and cannot recall one that is beyond bronze or lower scores by the end anyways. And early in competitions, much is learned from kernels, discussions, links. But have seen your work included elsewhere without a simple paste of the link to your kernel to credit the source. That seems far worse to me than what a public kernel scores.
", "votes": 1, "replies": [] }, { "id": 547665, "author_name": "Andrey Lukyanenko", "author_url": "", "post_date": "2019-06-08T05:12:56.013000", "content": "@cbryant well, if this new kernel is different from your solution, you can use it to improve your own score, right?
\n\n@something4kag I got used to the fact that some people simply copy kernels, little can be done about it sadly.
", "votes": 1, "replies": [] }, { "id": 548029, "author_name": "Jacek Poplawski", "author_url": "", "post_date": "2019-06-08T16:45:54.107000", "content": "thanks Andrew for sane voice
", "votes": 1, "replies": [] }, { "id": 549133, "author_name": "cyberia", "author_url": "", "post_date": "2019-06-10T10:28:14.703000", "content": "First of all a huge thanks to you for all the amazing work you have done so far in this competition and the future work.
\n\n\n\n\nI often say: if I'm in a medal range and there is a public kernel with a higher score, then I don't deserve this medal.
\n
This shows your high credibility which is present in very few these days. I follow your work very closely and found it very inspiring. In fact I have a list of all the kernels you have ever written on kaggle and I learn from them. Infact I am modifying one of your kernels. So there is no doubt that a kernel is needed as part of learning on Kaggle
\n\nNow what my point is there are users who never look into kernels and all they do is three clicks to get a position on LB. This practice is OK for beginners but it is expected that eventually at some stage they would start to write their own kernels and original stuff but they never do that it results into an increase in number of those who fall in not more than three clicks category now the people who were writing their own kernels don't get what they deserved.
This has been a topic of debate for a long time on kaggle not just us having this conversation. Probably in future I would love to see something in which people who don't do more than three clicks to reach LB don't get medals.
", "votes": 0, "replies": [] }, { "id": 549141, "author_name": "Andrey Lukyanenko", "author_url": "", "post_date": "2019-06-10T10:48:24.317000", "content": "Thank you for your kind words!
\n\nI agree that if people can get medals without any efforts it is bad. It diminishes the value of medals and discourages honest people to take part in further competition.
\n\nBut I want to make several points:\n- lately getting silver medals becomes more and more difficult, often by the end of a competition there are no public kernels in silver range;\n- publishing high scoring kernels is prohibited within last week of a competition, so most top teams will exceed scores of public kernels;\n- I'm sure that in this particular competition it would be very difficult to get even silver medal by training models only on kernels. The datafiles in this competition are quite big, if you generate several hundreds of features, you could hit memory limit. Also it is obvious that approaches using graph networks and graph neural nets will be very useful and I'm not sure whether it is possible to implement these approaches in kernels with good enough quality. So I think that most of people at the top of the leaderboard will use their own hardware and not kaggle kernels.
\n\nStill, despite these points I want to repeat that getting medals without efforts is bad.
", "votes": 5, "replies": [] }, { "id": 583684, "author_name": "Kha Vo", "author_url": "", "post_date": "2019-07-24T20:27:05.250000", "content": "I don’t agree. Imagine if my team publish a -2.1 kernel, then 100 teams below you will jump up. Then you will understand that you’re wrong.
", "votes": 0, "replies": [] }, { "id": 583820, "author_name": "Andrey Lukyanenko", "author_url": "", "post_date": "2019-07-25T03:20:08.737000", "content": "No, I'd be glad that there is a new public kernel with unique approach/features, which we could analyze and add to our solution.
", "votes": 0, "replies": [] } ] }, { "id": 548167, "author_name": "Ming Zhao", "author_url": "", "post_date": "2019-06-08T22:55:27.023000", "content": "Have you asked yourself why you are spending your precious time on these Kaggle competitions instead of any other alternatives? If you are after the prize money, good luck with that if you can't even beat a public kernel. If you like to hoard internet points, there are many alternative sites that you don't have to work so hard.
\n\nIf like most of us, you are trying to learn new things, you should probably be thankful that someone spent their time sharing their knowledge so that you can learn from them, especially if they are better than you.
", "votes": 17, "replies": [] }, { "id": 547274, "author_name": "CoreyJamesLevinson", "author_url": "", "post_date": "2019-06-07T13:42:14.857000", "content": "There are 3 months left in the competition, don't worry. This will not be the highest scoring kernel. People will innovate and beat it, especially you.
", "votes": 10, "replies": [ { "id": 547278, "author_name": "Manoj", "author_url": "", "post_date": "2019-06-07T13:58:02.020000", "content": "I agree. its too early for not to share.Otherwise we would not able to see the true potential of the data.
", "votes": 2, "replies": [] } ] }, { "id": 548026, "author_name": "Jacek Poplawski", "author_url": "", "post_date": "2019-06-08T16:43:09.547000", "content": "I downvoted you. I don't understand people who upvoted you.
\n\nWhat are you talking about?
\n\nCompetition just started, I am just getting familar with the data. And you already did \"all hard work\"? When?
\n\nKaggle is for learning, I think some people don't understand that at all.
", "votes": 7, "replies": [ { "id": 549119, "author_name": "cyberia", "author_url": "", "post_date": "2019-06-10T10:07:23.200000", "content": "I upvoted you. Why? because of the following
\n\n\nWhat are you talking about?
Issue comes when there are some who did not even looked at the kernel, clicked not more than thrice and are on better position than those who have invested time and energy to get a good position.
\n\nI am actually learning a lot from Andrew Lukyanenko in this competition. And his contribution in this competition is unparallel
\n\n\n\n\nI don't understand people who upvoted you.
\n
They are the ones who have at some stage felt the pain of working on something and seeing that going in vain and most importantly look at big picture of machine learning as a whole not limited to any competition hosting site.
\n\nRemember : All models are same. No model is better than other but kaggle makes them appear better than other by ranking and LB. Sometimes it is hard to help people realize what they are losing by following an approach.
\n\nI agree with most of your points. Wish you luck.
", "votes": 2, "replies": [] } ] }, { "id": 547834, "author_name": "Miguel Pinto", "author_url": "", "post_date": "2019-06-08T11:14:23.563000", "content": "The greatest value of these competitions is learning. For that purpose public kernels and discussions can be very useful. Medals should not be viewed as the goal but as the consequence of learning :)
", "votes": 5, "replies": [] }, { "id": 547393, "author_name": "IggiSv9t", "author_url": "", "post_date": "2019-06-07T16:52:19.873000", "content": "If you did something original, then high scoring kernels are also useful for you. You can add features or methods from another kernel to your solution and gain your score. At least you can simply blend submissions.
", "votes": 5, "replies": [] }, { "id": 548659, "author_name": "Mohammed A. Hassan", "author_url": "", "post_date": "2019-06-09T17:12:21.037000", "content": "As mentioned in the comments a lot of us are here to learn. For a Kaggle beginner like me, having this kind of kernels helps me understand how other people approach these modelling \"puzzles\". Plenty of time left for you to improve your kernel and who knows you might find a solution in another public kernel that helps win the competition!
", "votes": 4, "replies": [] }, { "id": 547656, "author_name": "Dhananjay Raut", "author_url": "", "post_date": "2019-06-08T04:37:52.633000", "content": "Exactly what I was thinking about. I want to make my kernel public, but then this( forking and submitting ) stops me. But it seems it's ok to share at this early stage of the competition and I still can't decide ( someone help me ). I want to have some medals so that I can also be an expert on kaggle :)
", "votes": 2, "replies": [] }, { "id": 547264, "author_name": "Somraj Saha", "author_url": "", "post_date": "2019-06-07T13:13:48.303000", "content": "I guess it's in human nature to \"learn\" from someone else's work to make your product better. Isn't this how someone came up with the concept of Transfer Learning? I get it that copying & submitting a high scoring public kernel on the leaderboard is practically just cheating but you could use that to your advantage! All that I'm trying to convey here is that why not try gaining some insights from the high scoring kernels, adding your own spice and ingredients to try and come up with a better model?
\n\nDon't get me wrong here, I'm just trying to be optimistic and prefer to look out-of-the-box to beat the ones who take the easy way out.
", "votes": 2, "replies": [] }, { "id": 549224, "author_name": "mephistopheies", "author_url": "", "post_date": "2019-06-10T12:38:28.040000", "content": "lol
", "votes": 1, "replies": [] }, { "id": 547694, "author_name": "Shahules", "author_url": "", "post_date": "2019-06-08T06:10:03.993000", "content": "yes,exactly..i believe public kernels should contain only baseline models.Tuning should be done in private mode,so that this issue won't happen.But this becomes a limiting factor for kernel authors as kernel having best score gets more views and upvotes.
", "votes": -1, "replies": [ { "id": 548028, "author_name": "Jacek Poplawski", "author_url": "", "post_date": "2019-06-08T16:45:09.490000", "content": "\n\n\ni believe public kernels should contain only baseline models.
\n
what happened to Kaggle? I dont't believe what I am reading
", "votes": 3, "replies": [] }, { "id": 548427, "author_name": "Ivan Dobryaev", "author_url": "", "post_date": "2019-06-09T11:19:47.600000", "content": "I think that Kaggle need a makes special block for education with kernels, with explain all problem.
", "votes": 0, "replies": [] } ] }, { "id": 583463, "author_name": "Sergey Zlobin", "author_url": "", "post_date": "2019-07-24T14:09:52.030000", "content": "Wow, -1.5 in public kernels now
", "votes": 0, "replies": [ { "id": 583467, "author_name": "somnambWl", "author_url": "", "post_date": "2019-07-24T14:14:13.133000", "content": "Yeah, they got from -1.3 to -1.5 with blending. There is no new public kernel which would actually create a model like for two weeks or more.
", "votes": 0, "replies": [] }, { "id": 583638, "author_name": "Sergey Zlobin", "author_url": "", "post_date": "2019-07-24T18:23:50.037000", "content": "It's surprising for me. Probably there are some new features for last 2 weeks?! Because such a boost > 0.1 is high for blending, I think.
", "votes": 0, "replies": [] } ] }, { "id": 547294, "author_name": "", "author_url": "", "post_date": "2019-06-07T14:26:16.280000", "content": "", "votes": 13, "replies": [ { "id": 547299, "author_name": "cyberia", "author_url": "", "post_date": "2019-06-07T14:33:47.643000", "content": "No one could put it as elegantly as you have. 👍
", "votes": 0, "replies": [] }, { "id": 547352, "author_name": "Mycroft Holmes", "author_url": "", "post_date": "2019-06-07T16:00:07.917000", "content": "It is not clear to me that the sponsor actually benefits from the excessive sharing of high-scoring kernels. It hurts the sponsor by damping the motivation of Kagglers and thus reducing the variety in approaches.
", "votes": 1, "replies": [] }, { "id": 547354, "author_name": "Mycroft Holmes", "author_url": "", "post_date": "2019-06-07T16:01:48.897000", "content": "I did see the warning message in the earthquake competition. It must be implemented already.
", "votes": 0, "replies": [] }, { "id": 549087, "author_name": "cyberia", "author_url": "", "post_date": "2019-06-10T09:31:38.260000", "content": "\n\n\nIt is not clear to me that the sponsor actually benefits from the excessive sharing of high-scoring kernels
\n
Good point. When it comes to hosting a competition. The goal is to crowdsource the problem (developing a model) which could be used by the host/sponsor to generate value by deploying it for their business. If there is not even one evidence/user story of a sponsor putting the winning solution into production in the entire history of Kaggle. You should be able to understand the hidden picture no one talks about or the one who talks about get downvotes 😄 .
\n\nIf anyone knows of any resource on internet where it is mentioned that sponsor got anything out of a competition which was realistic. Please let me know, I would be highly thankful for that, I have searched but never found anything.
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "547378": "This competition started 9 days ago! Only 9 days. Within 2 months these high scores will seem to be quite low.\nAlso I want to repeat a point, which I often say: if I'm in a medal range and there is a public kernel with a higher score, then I don't deserve this medal.\nBut I can use this public kernel to improve my own solution.", "548167": "Have you asked yourself why you are spending your precious time on these Kaggle competitions instead of any other alternatives? If you are after the prize money, good luck with that if you can't even beat a public kernel. If you like to hoard internet points, there are many alternative sites that you don't have to work so hard.\n\nIf like most of us, you are trying to learn new things, you should probably be thankful that someone spent their time sharing their knowledge so that you can learn from them, especially if they are better than you.", "547274": "There are 3 months left in the competition, don't worry. This will not be the highest scoring kernel. People will innovate and beat it, especially you.", "548026": "I downvoted you. I don't understand people who upvoted you.\n\nWhat are you talking about?\n\nCompetition just started, I am just getting familar with the data. And you already did \"all hard work\"? When?\n\nKaggle is for learning, I think some people don't understand that at all.\n\n\n\n\n", "547834": "The greatest value of these competitions is learning. For that purpose public kernels and discussions can be very useful. Medals should not be viewed as the goal but as the consequence of learning :) ", "547393": "If you did something original, then high scoring kernels are also useful for you. You can add features or methods from another kernel to your solution and gain your score. At least you can simply blend submissions.", "548659": "As mentioned in the comments a lot of us are here to learn. For a Kaggle beginner like me, having this kind of kernels helps me understand how other people approach these modelling \"puzzles\". Plenty of time left for you to improve your kernel and who knows you might find a solution in another public kernel that helps win the competition!", "547254": "You are working on your ideas putting in effort on your own, stars get aligned and rank closes in towards gold and then... A high scoring kernel is published all hard work goes in vain as a lot of people just fork and submit. All done in not more than three clicks.\n\nThere should be some sort of honor code and ethics regarding this, I wish if kernels could contain complete EDA, visualizations and summaries with code but only clear, concise step by step hints given for modelling.\n\nHave to start all over again.", "547656": "Exactly what I was thinking about. I want to make my kernel public, but then this( forking and submitting ) stops me. But it seems it's ok to share at this early stage of the competition and I still can't decide ( someone help me ). I want to have some medals so that I can also be an expert on kaggle :) ", "547264": "I guess it's in human nature to \"*learn*\" from someone else's work to make your product better. Isn't this how someone came up with the concept of *Transfer Learning*? I get it that copying & submitting a high scoring public kernel on the leaderboard is practically just cheating but you could use that to your advantage! All that I'm trying to convey here is that why not try gaining some insights from the high scoring kernels, adding your own spice and ingredients to try and come up with a better model?\n\nDon't get me wrong here, I'm just trying to be optimistic and prefer to look out-of-the-box to beat the ones who take the easy way out.", "549224": "lol", "547694": "yes,exactly..i believe public kernels should contain only baseline models.Tuning should be done in private mode,so that this issue won't happen.But this becomes a limiting factor for kernel authors as kernel having best score gets more views and upvotes.", "583463": "Wow, -1.5 in public kernels now", "547294": "" } }