This was Kaggle's first analytics competition and my first attempt at competing in a competition. The first time you do anything, you learn a lot. I made my own personal lessons learned list, and I created this topic to give Kaggle feedback on the competition.
\n\nWhat worked? What didn't? What would want to see different next time?
", "messages": [ { "id": "455002", "postDate": "01/12/2019 18:15:33", "content": "This was Kaggle's first analytics competition and my first attempt at competing in a competition. The first time you do anything, you learn a lot. I made my own personal lessons learned list, and I created this topic to give Kaggle feedback on the competition.
\n\nWhat worked? What didn't? What would want to see different next time?
", "rawMarkdown": "This was Kaggle's first analytics competition and my first attempt at competing in a competition. The first time you do anything, you learn a lot. I made my own personal lessons learned list, and I created this topic to give Kaggle feedback on the competition.\n\nWhat worked? What didn't? What would want to see different next time?", "votes": null }, { "id": "455008", "postDate": "01/12/2019 18:34:22", "content": "How we were to split between the kernel and presentation was unclear to me until Chris clarified it on the last weekend. That would have been great to have at the beginning.
", "rawMarkdown": "How we were to split between the kernel and presentation was unclear to me until Chris clarified it on the last weekend. That would have been great to have at the beginning.", "votes": null }, { "id": "455030", "postDate": "01/12/2019 19:32:45", "content": "I loved pretty much everything about this opportunity to apply my data science skills to my favorite sport. But I do have these issues:\n- I found the Kaggle editor unreliable, so I ended up using a local notebook\n- A large majority of highly upvoted Kernels propose the same boring rule - incentivize a fair catch. So in my opinion, the popularity of Kernels seems to be based on style over substance.
", "rawMarkdown": "I loved pretty much everything about this opportunity to apply my data science skills to my favorite sport. But I do have these issues:\n- I found the Kaggle editor unreliable, so I ended up using a local notebook\n- A large majority of highly upvoted Kernels propose the same boring rule - incentivize a fair catch. So in my opinion, the popularity of Kernels seems to be based on style over substance.", "votes": null }, { "id": "455221", "postDate": "01/13/2019 10:00:28", "content": "Well, this comp format implied like 150 kernels being published at the same time, what makes it difficult for all of us to digest them, or even read them all... (in fact I find it a problem of the format, we dont share before and we dont discuss much afterwards because 1) all opinions are already strongly expressed in final kernels and 2) we are sooo tired.)
\n\nThat said, I think the path to a conclussion of analysis is as important as the conclussion itself. Especially true if analytics are, like in this case, supporting measures to be shared with a wide audience. Data Science is also about communication. So style matters.
\n\nImo substance is also there, in a good part of the Kernels, popular and not so popular. Recommendations should be \"effective\", or \"possibly useful\", what is already a lot considering data available.
\n\nI guess the comment on \"boring\" measures refers to making the game less dynamic, but that has to do with actionability of measure and will be decided by judges of competition. I focused in other aspects but from a Data Science point of view nothing boring about that measures as long as they are founded and can help to reduce concussions.
", "rawMarkdown": "Well, this comp format implied like 150 kernels being published at the same time, what makes it difficult for all of us to digest them, or even read them all... (in fact I find it a problem of the format, we dont share before and we dont discuss much afterwards because 1) all opinions are already strongly expressed in final kernels and 2) we are sooo tired.)\n\nThat said, I think the path to a conclussion of analysis is as important as the conclussion itself. Especially true if analytics are, like in this case, supporting measures to be shared with a wide audience. Data Science is also about communication. So style matters.\n\nImo substance is also there, in a good part of the Kernels, popular and not so popular. Recommendations should be \"effective\", or \"possibly useful\", what is already a lot considering data available.\n\n I guess the comment on \"boring\" measures refers to making the game less dynamic, but that has to do with actionability of measure and will be decided by judges of competition. I focused in other aspects but from a Data Science point of view nothing boring about that measures as long as they are founded and can help to reduce concussions.", "votes": null }, { "id": "455281", "postDate": "01/13/2019 13:27:29", "content": "I agree with your first point about the lack of sharing kernels. Having been one of a small few to publish a kernel with suggested rule changes / general EDA quite early (around two weeks in) I can understand why people didn't because I and a few others had their kernels copied entirely and reposted as part of other kernels. This was the only reason why I didn't provide regular updates to my own kernel until the very end.
", "rawMarkdown": "I agree with your first point about the lack of sharing kernels. Having been one of a small few to publish a kernel with suggested rule changes / general EDA quite early (around two weeks in) I can understand why people didn't because I and a few others had their kernels copied entirely and reposted as part of other kernels. This was the only reason why I didn't provide regular updates to my own kernel until the very end.", "votes": null }, { "id": "455328", "postDate": "01/13/2019 16:01:49", "content": "I took \"boring\" rule to mean it was a pretty obvious rule and it didn't take a lot of creativity to come up with it.
\n\nI agree that style matters. I haven't seen a popular kernel with poor substance. They all had good substance, so the ones that stood out had better visuals and a better presentation.
\n\nAs a kaggle noob, I'm curious why people publish kernels early in other competitions? Is it that they are so far past that kernel, it's no longer competitive, so they share it? Or are they trying to get rewarded for sharing kernels?
", "rawMarkdown": "I took \"boring\" rule to mean it was a pretty obvious rule and it didn't take a lot of creativity to come up with it. \n\nI agree that style matters. I haven't seen a popular kernel with poor substance. They all had good substance, so the ones that stood out had better visuals and a better presentation. \n\nAs a kaggle noob, I'm curious why people publish kernels early in other competitions? Is it that they are so far past that kernel, it's no longer competitive, so they share it? Or are they trying to get rewarded for sharing kernels?", "votes": null }, { "id": "455364", "postDate": "01/13/2019 17:08:13", "content": "I think the vast majority of people share their work early on because it promotes progressive learning. I personally gained a lot in some ways from posting my kernel early because another kaggler pointed out to me in the comments that one of my suggested rule changes had already been implemented this year. Had I not posted the kernel early I may not have known that and submitted a rule change that was already in place. I also shared my work because the likelihood of winning a competition is generally quite low so I might as well just post my kernel anyway.
", "rawMarkdown": "I think the vast majority of people share their work early on because it promotes progressive learning. I personally gained a lot in some ways from posting my kernel early because another kaggler pointed out to me in the comments that one of my suggested rule changes had already been implemented this year. Had I not posted the kernel early I may not have known that and submitted a rule change that was already in place. I also shared my work because the likelihood of winning a competition is generally quite low so I might as well just post my kernel anyway.", "votes": null }, { "id": "455470", "postDate": "01/14/2019 01:03:38", "content": "Incentivizing the fair catch was a pretty obvious (in your words, boring) starting and ending point because one could easily infer causation between returns and concussions, and static incentives are always better than penalties in terms of the flow of the game and the possibly subjective nature of proposed penalties.
\n\nThe creativity comes in determining pre-snap criteria which qualify a punt as one where the fair catch bonus is active. That bonus could very easily upset the balance of the game without the right qualification criteria in place.
\n\nRegardless, it is fun to see a bunch of smart people attacking the same business problem from different angles and having all those angles ultimately direct back to some common suggestions, each with their own unique quirks.
", "rawMarkdown": "Incentivizing the fair catch was a pretty obvious (in your words, boring) starting and ending point because one could easily infer causation between returns and concussions, and static incentives are always better than penalties in terms of the flow of the game and the possibly subjective nature of proposed penalties. \n\nThe creativity comes in determining pre-snap criteria which qualify a punt as one where the fair catch bonus is active. That bonus could very easily upset the balance of the game without the right qualification criteria in place.\n\nRegardless, it is fun to see a bunch of smart people attacking the same business problem from different angles and having all those angles ultimately direct back to some common suggestions, each with their own unique quirks.", "votes": null }, { "id": "455513", "postDate": "01/14/2019 04:52:22", "content": "+1 to everything Colin said. Also I don’t see why a “boring” rule change a bad thing for this competition? In my opinion proposing something that can reduce injuries with as minimal impact to the culture or integrity of the game is key. This game is woven into the culture of the US from the NFL, college, high school, all the way down to peewee football. Proposing a drastic change requires drastic support for how it would truly reduce injuries better than other “boring” changes.
\n\nAnother thing to remember is that the votes right now mean nothing until the final results are determined. So we should probably wait until then before jumping to any conclusions. I know for the Kaggle Survey challenge the winning kernel had very few views and votes before being announced as the winner (it was an excellent kernel nonetheless). But that just goes to show the kaggle team did review every kernel and didn’t rely on votes to make their decisions.
\n\nI had an idea for my rule change proposals fairly early on and was pretty devisdtated when I saw many others with the same ideas. But in the end, just like you said it’s pretty cool to see people coming to the same conclusions from different angles.
", "rawMarkdown": "1 to everything Colin said. Also I don’t see why a “boring” rule change a bad thing for this competition? In my opinion proposing something that can reduce injuries with as minimal impact to the culture or integrity of the game is key. This game is woven into the culture of the US from the NFL, college, high school, all the way down to peewee football. Proposing a drastic change requires drastic support for how it would truly reduce injuries better than other “boring” changes.\n\nAnother thing to remember is that the votes right now mean nothing until the final results are determined. So we should probably wait until then before jumping to any conclusions. I know for the Kaggle Survey challenge the winning kernel had very few views and votes before being announced as the winner (it was an excellent kernel nonetheless). But that just goes to show the kaggle team did review every kernel and didn’t rely on votes to make their decisions.\n\nI had an idea for my rule change proposals fairly early on and was pretty devisdtated when I saw many others with the same ideas. But in the end, just like you said it’s pretty cool to see people coming to the same conclusions from different angles.", "votes": null }, { "id": "455816", "postDate": "01/14/2019 16:38:56", "content": "\"But that just goes to show the kaggle team did review every kernel and didn’t rely on votes to make their decisions.\"
\n\nThank you for saying that.
", "rawMarkdown": "\"But that just goes to show the kaggle team did review every kernel and didn’t rely on votes to make their decisions.\"\n\nThank you for saying that.", "votes": null }, { "id": "455840", "postDate": "01/14/2019 17:17:11", "content": "Great points, all. This was my first competition and, really, one of my first analyses of any kind. It seemed clear to me that the organizers were trying to push us in a more realistic direction when it came to recommendations. Lots of discussion of the integrity of existing rules and an emphasis on the impact of proposed rule changes to gameplay. All of which is to say that more boring answers were not surprising and I also think probably approximate the realities of data analysis in a corporate environment.
\n\nLooking forward to reading through a lot of these kernels to see what I could have done better.
", "rawMarkdown": "Great points, all. This was my first competition and, really, one of my first analyses of any kind. It seemed clear to me that the organizers were trying to push us in a more realistic direction when it came to recommendations. Lots of discussion of the integrity of existing rules and an emphasis on the impact of proposed rule changes to gameplay. All of which is to say that more boring answers were not surprising and I also think probably approximate the realities of data analysis in a corporate environment.\n\nLooking forward to reading through a lot of these kernels to see what I could have done better.", "votes": null }, { "id": "455844", "postDate": "01/14/2019 17:20:11", "content": "Responding to the topic of posting kernels early, in prediction competitions there's an objective scoring metric that can be improved incrementally over the course of the competition, so there's not much risk in sharing code early on. You can share some starer code or a general approach without giving away your best model that is objectively better than the ones you share.
", "rawMarkdown": "Responding to the topic of posting kernels early, in prediction competitions there's an objective scoring metric that can be improved incrementally over the course of the competition, so there's not much risk in sharing code early on. You can share some starer code or a general approach without giving away your best model that is objectively better than the ones you share.", "votes": null }, { "id": "456386", "postDate": "01/15/2019 18:01:15", "content": "I really enjoy about how big is the field of the data science. Everyday we learn something new and even if we are just beginners is absolutely amazing what we can do right now. I just want to talk to my future me and ask him about his knowledge.
\n\nAll experts were once a beginners.
", "rawMarkdown": "I really enjoy about how big is the field of the data science. Everyday we learn something new and even if we are just beginners is absolutely amazing what we can do right now. I just want to talk to my future me and ask him about his knowledge.\n\nAll experts were once a beginners.", "votes": null }, { "id": "456469", "postDate": "01/15/2019 21:49:32", "content": "@ericfreeman Thanks for the feedback and sorry that wasn't more clear from the start
", "rawMarkdown": "ericfreeman Thanks for the feedback and sorry that wasn't more clear from the start", "votes": null }, { "id": "456797", "postDate": "01/16/2019 15:03:00", "content": "Really great points by all. I don't think there's anything wrong with a boring but impactful rule change. However, I believe if the kernels were published and shared more democratically prior to the end of the competition, it would have implored more \"outside-the-box\" thinking. This would have lead to a more variety of solutions for the judging committee and ultimately the NFL to review.
\n\nEither way, I really enjoyed the competition and hope to work on more like it in the future.
", "rawMarkdown": "Really great points by all. I don't think there's anything wrong with a boring but impactful rule change. However, I believe if the kernels were published and shared more democratically prior to the end of the competition, it would have implored more \"outside-the-box\" thinking. This would have lead to a more variety of solutions for the judging committee and ultimately the NFL to review. \n\nEither way, I really enjoyed the competition and hope to work on more like it in the future.", "votes": null }, { "id": "456803", "postDate": "01/16/2019 15:11:33", "content": "Not a big deal. It's the first time doing this kind of thing. It didn't cause me any problems because I was so absorbed in the analysis, I hadn't started the presentation.
\n\nIn my personal lessons learned, I put \"don't want until the last week to work on the presentation\"
", "rawMarkdown": "Not a big deal. It's the first time doing this kind of thing. It didn't cause me any problems because I was so absorbed in the analysis, I hadn't started the presentation.\n\nIn my personal lessons learned, I put \"don't want until the last week to work on the presentation\"", "votes": null } ], "comments": [ { "id": 455008, "author_name": "ericfreeman", "author_url": "", "post_date": "01/12/2019 18:34:22", "content": "How we were to split between the kernel and presentation was unclear to me until Chris clarified it on the last weekend. That would have been great to have at the beginning.
", "votes": null, "replies": [ { "id": 456469, "author_name": "crawford", "author_url": "", "post_date": "01/15/2019 21:49:32", "content": "@ericfreeman Thanks for the feedback and sorry that wasn't more clear from the start
", "votes": null, "replies": [] }, { "id": 456803, "author_name": "ericfreeman", "author_url": "", "post_date": "01/16/2019 15:11:33", "content": "Not a big deal. It's the first time doing this kind of thing. It didn't cause me any problems because I was so absorbed in the analysis, I hadn't started the presentation.
\n\nIn my personal lessons learned, I put \"don't want until the last week to work on the presentation\"
", "votes": null, "replies": [] } ] }, { "id": 455030, "author_name": "sp007c", "author_url": "", "post_date": "01/12/2019 19:32:45", "content": "I loved pretty much everything about this opportunity to apply my data science skills to my favorite sport. But I do have these issues:\n- I found the Kaggle editor unreliable, so I ended up using a local notebook\n- A large majority of highly upvoted Kernels propose the same boring rule - incentivize a fair catch. So in my opinion, the popularity of Kernels seems to be based on style over substance.
", "votes": null, "replies": [ { "id": 455221, "author_name": "miguelpm", "author_url": "", "post_date": "01/13/2019 10:00:28", "content": "Well, this comp format implied like 150 kernels being published at the same time, what makes it difficult for all of us to digest them, or even read them all... (in fact I find it a problem of the format, we dont share before and we dont discuss much afterwards because 1) all opinions are already strongly expressed in final kernels and 2) we are sooo tired.)
\n\nThat said, I think the path to a conclussion of analysis is as important as the conclussion itself. Especially true if analytics are, like in this case, supporting measures to be shared with a wide audience. Data Science is also about communication. So style matters.
\n\nImo substance is also there, in a good part of the Kernels, popular and not so popular. Recommendations should be \"effective\", or \"possibly useful\", what is already a lot considering data available.
\n\nI guess the comment on \"boring\" measures refers to making the game less dynamic, but that has to do with actionability of measure and will be decided by judges of competition. I focused in other aspects but from a Data Science point of view nothing boring about that measures as long as they are founded and can help to reduce concussions.
", "votes": null, "replies": [] }, { "id": 455281, "author_name": "garlsham", "author_url": "", "post_date": "01/13/2019 13:27:29", "content": "I agree with your first point about the lack of sharing kernels. Having been one of a small few to publish a kernel with suggested rule changes / general EDA quite early (around two weeks in) I can understand why people didn't because I and a few others had their kernels copied entirely and reposted as part of other kernels. This was the only reason why I didn't provide regular updates to my own kernel until the very end.
", "votes": null, "replies": [] }, { "id": 455328, "author_name": "ericfreeman", "author_url": "", "post_date": "01/13/2019 16:01:49", "content": "I took \"boring\" rule to mean it was a pretty obvious rule and it didn't take a lot of creativity to come up with it.
\n\nI agree that style matters. I haven't seen a popular kernel with poor substance. They all had good substance, so the ones that stood out had better visuals and a better presentation.
\n\nAs a kaggle noob, I'm curious why people publish kernels early in other competitions? Is it that they are so far past that kernel, it's no longer competitive, so they share it? Or are they trying to get rewarded for sharing kernels?
", "votes": null, "replies": [] }, { "id": 455364, "author_name": "garlsham", "author_url": "", "post_date": "01/13/2019 17:08:13", "content": "I think the vast majority of people share their work early on because it promotes progressive learning. I personally gained a lot in some ways from posting my kernel early because another kaggler pointed out to me in the comments that one of my suggested rule changes had already been implemented this year. Had I not posted the kernel early I may not have known that and submitted a rule change that was already in place. I also shared my work because the likelihood of winning a competition is generally quite low so I might as well just post my kernel anyway.
", "votes": null, "replies": [] }, { "id": 455470, "author_name": "colin33", "author_url": "", "post_date": "01/14/2019 01:03:38", "content": "Incentivizing the fair catch was a pretty obvious (in your words, boring) starting and ending point because one could easily infer causation between returns and concussions, and static incentives are always better than penalties in terms of the flow of the game and the possibly subjective nature of proposed penalties.
\n\nThe creativity comes in determining pre-snap criteria which qualify a punt as one where the fair catch bonus is active. That bonus could very easily upset the balance of the game without the right qualification criteria in place.
\n\nRegardless, it is fun to see a bunch of smart people attacking the same business problem from different angles and having all those angles ultimately direct back to some common suggestions, each with their own unique quirks.
", "votes": null, "replies": [] }, { "id": 455513, "author_name": "robikscube", "author_url": "", "post_date": "01/14/2019 04:52:22", "content": "+1 to everything Colin said. Also I don’t see why a “boring” rule change a bad thing for this competition? In my opinion proposing something that can reduce injuries with as minimal impact to the culture or integrity of the game is key. This game is woven into the culture of the US from the NFL, college, high school, all the way down to peewee football. Proposing a drastic change requires drastic support for how it would truly reduce injuries better than other “boring” changes.
\n\nAnother thing to remember is that the votes right now mean nothing until the final results are determined. So we should probably wait until then before jumping to any conclusions. I know for the Kaggle Survey challenge the winning kernel had very few views and votes before being announced as the winner (it was an excellent kernel nonetheless). But that just goes to show the kaggle team did review every kernel and didn’t rely on votes to make their decisions.
\n\nI had an idea for my rule change proposals fairly early on and was pretty devisdtated when I saw many others with the same ideas. But in the end, just like you said it’s pretty cool to see people coming to the same conclusions from different angles.
", "votes": null, "replies": [] }, { "id": 455816, "author_name": "sp007c", "author_url": "", "post_date": "01/14/2019 16:38:56", "content": "\"But that just goes to show the kaggle team did review every kernel and didn’t rely on votes to make their decisions.\"
\n\nThank you for saying that.
", "votes": null, "replies": [] }, { "id": 455844, "author_name": "hamelg", "author_url": "", "post_date": "01/14/2019 17:20:11", "content": "Responding to the topic of posting kernels early, in prediction competitions there's an objective scoring metric that can be improved incrementally over the course of the competition, so there's not much risk in sharing code early on. You can share some starer code or a general approach without giving away your best model that is objectively better than the ones you share.
", "votes": null, "replies": [] } ] }, { "id": 455840, "author_name": "rkkamp", "author_url": "", "post_date": "01/14/2019 17:17:11", "content": "Great points, all. This was my first competition and, really, one of my first analyses of any kind. It seemed clear to me that the organizers were trying to push us in a more realistic direction when it came to recommendations. Lots of discussion of the integrity of existing rules and an emphasis on the impact of proposed rule changes to gameplay. All of which is to say that more boring answers were not surprising and I also think probably approximate the realities of data analysis in a corporate environment.
\n\nLooking forward to reading through a lot of these kernels to see what I could have done better.
", "votes": null, "replies": [] }, { "id": 456386, "author_name": "joseconomy", "author_url": "", "post_date": "01/15/2019 18:01:15", "content": "I really enjoy about how big is the field of the data science. Everyday we learn something new and even if we are just beginners is absolutely amazing what we can do right now. I just want to talk to my future me and ask him about his knowledge.
\n\nAll experts were once a beginners.
", "votes": null, "replies": [] }, { "id": 456797, "author_name": "mtodisco10", "author_url": "", "post_date": "01/16/2019 15:03:00", "content": "Really great points by all. I don't think there's anything wrong with a boring but impactful rule change. However, I believe if the kernels were published and shared more democratically prior to the end of the competition, it would have implored more \"outside-the-box\" thinking. This would have lead to a more variety of solutions for the judging committee and ultimately the NFL to review.
\n\nEither way, I really enjoyed the competition and hope to work on more like it in the future.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "455002": "This was Kaggle's first analytics competition and my first attempt at competing in a competition. The first time you do anything, you learn a lot. I made my own personal lessons learned list, and I created this topic to give Kaggle feedback on the competition.\n\nWhat worked? What didn't? What would want to see different next time?", "455008": "How we were to split between the kernel and presentation was unclear to me until Chris clarified it on the last weekend. That would have been great to have at the beginning.", "455030": "I loved pretty much everything about this opportunity to apply my data science skills to my favorite sport. But I do have these issues:\n- I found the Kaggle editor unreliable, so I ended up using a local notebook\n- A large majority of highly upvoted Kernels propose the same boring rule - incentivize a fair catch. So in my opinion, the popularity of Kernels seems to be based on style over substance.", "455221": "Well, this comp format implied like 150 kernels being published at the same time, what makes it difficult for all of us to digest them, or even read them all... (in fact I find it a problem of the format, we dont share before and we dont discuss much afterwards because 1) all opinions are already strongly expressed in final kernels and 2) we are sooo tired.)\n\nThat said, I think the path to a conclussion of analysis is as important as the conclussion itself. Especially true if analytics are, like in this case, supporting measures to be shared with a wide audience. Data Science is also about communication. So style matters.\n\nImo substance is also there, in a good part of the Kernels, popular and not so popular. Recommendations should be \"effective\", or \"possibly useful\", what is already a lot considering data available.\n\n I guess the comment on \"boring\" measures refers to making the game less dynamic, but that has to do with actionability of measure and will be decided by judges of competition. I focused in other aspects but from a Data Science point of view nothing boring about that measures as long as they are founded and can help to reduce concussions.", "455281": "I agree with your first point about the lack of sharing kernels. Having been one of a small few to publish a kernel with suggested rule changes / general EDA quite early (around two weeks in) I can understand why people didn't because I and a few others had their kernels copied entirely and reposted as part of other kernels. This was the only reason why I didn't provide regular updates to my own kernel until the very end.", "455328": "I took \"boring\" rule to mean it was a pretty obvious rule and it didn't take a lot of creativity to come up with it. \n\nI agree that style matters. I haven't seen a popular kernel with poor substance. They all had good substance, so the ones that stood out had better visuals and a better presentation. \n\nAs a kaggle noob, I'm curious why people publish kernels early in other competitions? Is it that they are so far past that kernel, it's no longer competitive, so they share it? Or are they trying to get rewarded for sharing kernels?", "455364": "I think the vast majority of people share their work early on because it promotes progressive learning. I personally gained a lot in some ways from posting my kernel early because another kaggler pointed out to me in the comments that one of my suggested rule changes had already been implemented this year. Had I not posted the kernel early I may not have known that and submitted a rule change that was already in place. I also shared my work because the likelihood of winning a competition is generally quite low so I might as well just post my kernel anyway.", "455470": "Incentivizing the fair catch was a pretty obvious (in your words, boring) starting and ending point because one could easily infer causation between returns and concussions, and static incentives are always better than penalties in terms of the flow of the game and the possibly subjective nature of proposed penalties. \n\nThe creativity comes in determining pre-snap criteria which qualify a punt as one where the fair catch bonus is active. That bonus could very easily upset the balance of the game without the right qualification criteria in place.\n\nRegardless, it is fun to see a bunch of smart people attacking the same business problem from different angles and having all those angles ultimately direct back to some common suggestions, each with their own unique quirks.", "455513": "1 to everything Colin said. Also I don’t see why a “boring” rule change a bad thing for this competition? In my opinion proposing something that can reduce injuries with as minimal impact to the culture or integrity of the game is key. This game is woven into the culture of the US from the NFL, college, high school, all the way down to peewee football. Proposing a drastic change requires drastic support for how it would truly reduce injuries better than other “boring” changes.\n\nAnother thing to remember is that the votes right now mean nothing until the final results are determined. So we should probably wait until then before jumping to any conclusions. I know for the Kaggle Survey challenge the winning kernel had very few views and votes before being announced as the winner (it was an excellent kernel nonetheless). But that just goes to show the kaggle team did review every kernel and didn’t rely on votes to make their decisions.\n\nI had an idea for my rule change proposals fairly early on and was pretty devisdtated when I saw many others with the same ideas. But in the end, just like you said it’s pretty cool to see people coming to the same conclusions from different angles.", "455816": "\"But that just goes to show the kaggle team did review every kernel and didn’t rely on votes to make their decisions.\"\n\nThank you for saying that.", "455840": "Great points, all. This was my first competition and, really, one of my first analyses of any kind. It seemed clear to me that the organizers were trying to push us in a more realistic direction when it came to recommendations. Lots of discussion of the integrity of existing rules and an emphasis on the impact of proposed rule changes to gameplay. All of which is to say that more boring answers were not surprising and I also think probably approximate the realities of data analysis in a corporate environment.\n\nLooking forward to reading through a lot of these kernels to see what I could have done better.", "455844": "Responding to the topic of posting kernels early, in prediction competitions there's an objective scoring metric that can be improved incrementally over the course of the competition, so there's not much risk in sharing code early on. You can share some starer code or a general approach without giving away your best model that is objectively better than the ones you share.", "456386": "I really enjoy about how big is the field of the data science. Everyday we learn something new and even if we are just beginners is absolutely amazing what we can do right now. I just want to talk to my future me and ask him about his knowledge.\n\nAll experts were once a beginners.", "456469": "ericfreeman Thanks for the feedback and sorry that wasn't more clear from the start", "456797": "Really great points by all. I don't think there's anything wrong with a boring but impactful rule change. However, I believe if the kernels were published and shared more democratically prior to the end of the competition, it would have implored more \"outside-the-box\" thinking. This would have lead to a more variety of solutions for the judging committee and ultimately the NFL to review. \n\nEither way, I really enjoyed the competition and hope to work on more like it in the future.", "456803": "Not a big deal. It's the first time doing this kind of thing. It didn't cause me any problems because I was so absorbed in the analysis, I hadn't started the presentation.\n\nIn my personal lessons learned, I put \"don't want until the last week to work on the presentation\"" }, "source": "meta" }