What a place, Kaggle. No matter how many competitions you have completed here you still get surprised by new challenges and formats.
\n\nI find this new comp. format really interesting, with a good and not so good part:
\n\nThe good part: It is more open than most other competitions in Kaggle. No defined metric to optimize, no clear and cut subset of reality to crunch with our boosting machines. \nJust some data related to the task at hand and all freedom available to \"do data science\".
The not so good part: A natural consequence of the nature of the competition -ideas and approaches will be competing here, not .csv files- With some exceptions, I see almost nothing of the usual exchange of ideas and kernels of other Kaggle competitions. It is just normal, probably will be solved after the end of competition when all solution kernels are made public. But in the meantime it makes the comp. threads and kernels much less active than usual.
Let's hope the lack of activity is just a consequence of the competition format and there will be much more exchange of ideas when this ends.
", "messages": [ { "id": "445355", "postDate": "12/26/2018 09:14:17", "content": "What a place, Kaggle. No matter how many competitions you have completed here you still get surprised by new challenges and formats.
\n\nI find this new comp. format really interesting, with a good and not so good part:
\n\nThe good part: It is more open than most other competitions in Kaggle. No defined metric to optimize, no clear and cut subset of reality to crunch with our boosting machines. \nJust some data related to the task at hand and all freedom available to \"do data science\".
The not so good part: A natural consequence of the nature of the competition -ideas and approaches will be competing here, not .csv files- With some exceptions, I see almost nothing of the usual exchange of ideas and kernels of other Kaggle competitions. It is just normal, probably will be solved after the end of competition when all solution kernels are made public. But in the meantime it makes the comp. threads and kernels much less active than usual.
Let's hope the lack of activity is just a consequence of the competition format and there will be much more exchange of ideas when this ends.
", "rawMarkdown": "What a place, Kaggle. No matter how many competitions you have completed here you still get surprised by new challenges and formats.\n\nI find this new comp. format really interesting, with a good and not so good part:\n\n- The good part: It is more open than most other competitions in Kaggle. No defined metric to optimize, no clear and cut subset of reality to crunch with our boosting machines. \nJust some data related to the task at hand and all freedom available to \"do data science\".\n\n- The not so good part: A natural consequence of the nature of the competition -ideas and approaches will be competing here, not .csv files- With some exceptions, I see almost nothing of the usual exchange of ideas and kernels of other Kaggle competitions. It is just normal, probably will be solved after the end of competition when all solution kernels are made public. But in the meantime it makes the comp. threads and kernels much less active than usual.\n\nLet's hope the lack of activity is just a consequence of the competition format and there will be much more exchange of ideas when this ends.", "votes": null }, { "id": "445470", "postDate": "12/26/2018 14:33:54", "content": "I agree that the sharing of Kernels has definitely been much smaller in this competition in comparison to the ML competitions which is unfortunate. Despite this I hope Kaggle continue these analytics based competitions as it doesn't require expert domain knowledge in machine learning and leaves the competition open to a much wider audience. I would like to know how many people have entered the competition just out of interest but they don't give any information about that.
", "rawMarkdown": "I agree that the sharing of Kernels has definitely been much smaller in this competition in comparison to the ML competitions which is unfortunate. Despite this I hope Kaggle continue these analytics based competitions as it doesn't require expert domain knowledge in machine learning and leaves the competition open to a much wider audience. I would like to know how many people have entered the competition just out of interest but they don't give any information about that.", "votes": null }, { "id": "445488", "postDate": "12/26/2018 15:06:37", "content": "It's good for a novice like me in the fact that i feel like my \"solution\" can actually make a difference to the game without completely understanding the math to path of how i get there. Peace everyone.
", "rawMarkdown": "It's good for a novice like me in the fact that i feel like my \"solution\" can actually make a difference to the game without completely understanding the math to path of how i get there. Peace everyone.", "votes": null }, { "id": "445528", "postDate": "12/26/2018 16:45:41", "content": "I feel like a big part of it is that a project like this has high barrier to entry in terms of time and knowledge. It requires more time and effort to write a report and make a presentation than it does to crank data through a boosted decision tree. You can't sit down and have a submission to a project like this ready in a couple of hours. This project would also be difficult to tackle if you didn't know much about the NFL.
", "rawMarkdown": "I feel like a big part of it is that a project like this has high barrier to entry in terms of time and knowledge. It requires more time and effort to write a report and make a presentation than it does to crank data through a boosted decision tree. You can't sit down and have a submission to a project like this ready in a couple of hours. This project would also be difficult to tackle if you didn't know much about the NFL.", "votes": null }, { "id": "445975", "postDate": "12/27/2018 09:41:07", "content": "About bigger time/knowledge required, I wouldnt say so... I mean, if you think of other comps best kernels and solutions, many of them are perfecly formated, high quality presentations.
\n\nAlso, about necessity of domain knowledge, my view is like in normal competitions, domain knowledge is a good hedge. But usually proper stats knowledge plus powerful tools proves to be what is needed. I have achieved decent results in many comps without prior domain knowledge and for sure the opposite can be true, domain knowledge without data science best practices dont take you too far.
\n\nBut we'll see, really curious about what final set of skills will prove to be useful in this one, we will find out soon :-)
", "rawMarkdown": "About bigger time/knowledge required, I wouldnt say so... I mean, if you think of other comps best kernels and solutions, many of them are perfecly formated, high quality presentations. \n\nAlso, about necessity of domain knowledge, my view is like in normal competitions, domain knowledge is a good hedge. But usually proper stats knowledge plus powerful tools proves to be what is needed. I have achieved decent results in many comps without prior domain knowledge and for sure the opposite can be true, domain knowledge without data science best practices dont take you too far.\n\nBut we'll see, really curious about what final set of skills will prove to be useful in this one, we will find out soon :-)", "votes": null }, { "id": "446134", "postDate": "12/27/2018 15:18:13", "content": "You are focusing on the best kernels and solutions whereas my comment was about barriers to entry. Of course the best submissions in any competition are likely to involve significant time and effort. My point is that the minimum amount of time and domain knowledge required to put together a submission is higher for a project like this than it is for the typical prediction competition.
", "rawMarkdown": "You are focusing on the best kernels and solutions whereas my comment was about barriers to entry. Of course the best submissions in any competition are likely to involve significant time and effort. My point is that the minimum amount of time and domain knowledge required to put together a submission is higher for a project like this than it is for the typical prediction competition.", "votes": null }, { "id": "446205", "postDate": "12/27/2018 17:58:52", "content": "Agreed, from that point of view of minimal barriers it can be true. (I was not thinking of minimal case, usually spend a lot of time tinkering with comps, even in the many cases that I just play with without ever submitting...) but yeah, mimimal effort for basic submission is probably bigger now.
", "rawMarkdown": "Agreed, from that point of view of minimal barriers it can be true. (I was not thinking of minimal case, usually spend a lot of time tinkering with comps, even in the many cases that I just play with without ever submitting...) but yeah, mimimal effort for basic submission is probably bigger now.", "votes": null }, { "id": "446398", "postDate": "12/28/2018 02:03:37", "content": "I thought this competition was a good idea around the holidays and lead up to the SuperBowl, one for the armchair coaches and refs, possibly to have a team that is not all kernel coders, one for the powerpoint presentation fans (not me!) Anyways certainly paying more attention to punts in the game lately!
\n\nPeople have shared the rulebook \nhttps://operations.nfl.com/media/3277/2018-nfl-rulebook_final-version.pdf
\n\nIn another post it was mentioned another competition by the NFL using NGS data:\nmore data for big data bowl\nhttps://operations.nfl.com/the-game/big-data-bowl/
\n\nPossibly not so much to share in kernels as that may align too closely to the submission. But would be interesting to know more on people's approach to analysis or what they found interesting in the data. Perhaps in post-comp wrap.
", "rawMarkdown": "I thought this competition was a good idea around the holidays and lead up to the SuperBowl, one for the armchair coaches and refs, possibly to have a team that is not all kernel coders, one for the powerpoint presentation fans (not me!) Anyways certainly paying more attention to punts in the game lately!\n\nPeople have shared the rulebook \nhttps://operations.nfl.com/media/3277/2018-nfl-rulebook_final-version.pdf\n\nIn another post it was mentioned another competition by the NFL using NGS data:\nmore data for big data bowl\nhttps://operations.nfl.com/the-game/big-data-bowl/\n\nPossibly not so much to share in kernels as that may align too closely to the submission. But would be interesting to know more on people's approach to analysis or what they found interesting in the data. Perhaps in post-comp wrap.", "votes": null } ], "comments": [ { "id": 445470, "author_name": "garlsham", "author_url": "", "post_date": "12/26/2018 14:33:54", "content": "I agree that the sharing of Kernels has definitely been much smaller in this competition in comparison to the ML competitions which is unfortunate. Despite this I hope Kaggle continue these analytics based competitions as it doesn't require expert domain knowledge in machine learning and leaves the competition open to a much wider audience. I would like to know how many people have entered the competition just out of interest but they don't give any information about that.
", "votes": null, "replies": [] }, { "id": 445488, "author_name": "riobarr", "author_url": "", "post_date": "12/26/2018 15:06:37", "content": "It's good for a novice like me in the fact that i feel like my \"solution\" can actually make a difference to the game without completely understanding the math to path of how i get there. Peace everyone.
", "votes": null, "replies": [] }, { "id": 445528, "author_name": "hamelg", "author_url": "", "post_date": "12/26/2018 16:45:41", "content": "I feel like a big part of it is that a project like this has high barrier to entry in terms of time and knowledge. It requires more time and effort to write a report and make a presentation than it does to crank data through a boosted decision tree. You can't sit down and have a submission to a project like this ready in a couple of hours. This project would also be difficult to tackle if you didn't know much about the NFL.
", "votes": null, "replies": [ { "id": 445975, "author_name": "miguelpm", "author_url": "", "post_date": "12/27/2018 09:41:07", "content": "About bigger time/knowledge required, I wouldnt say so... I mean, if you think of other comps best kernels and solutions, many of them are perfecly formated, high quality presentations.
\n\nAlso, about necessity of domain knowledge, my view is like in normal competitions, domain knowledge is a good hedge. But usually proper stats knowledge plus powerful tools proves to be what is needed. I have achieved decent results in many comps without prior domain knowledge and for sure the opposite can be true, domain knowledge without data science best practices dont take you too far.
\n\nBut we'll see, really curious about what final set of skills will prove to be useful in this one, we will find out soon :-)
", "votes": null, "replies": [] }, { "id": 446134, "author_name": "hamelg", "author_url": "", "post_date": "12/27/2018 15:18:13", "content": "You are focusing on the best kernels and solutions whereas my comment was about barriers to entry. Of course the best submissions in any competition are likely to involve significant time and effort. My point is that the minimum amount of time and domain knowledge required to put together a submission is higher for a project like this than it is for the typical prediction competition.
", "votes": null, "replies": [] }, { "id": 446205, "author_name": "miguelpm", "author_url": "", "post_date": "12/27/2018 17:58:52", "content": "Agreed, from that point of view of minimal barriers it can be true. (I was not thinking of minimal case, usually spend a lot of time tinkering with comps, even in the many cases that I just play with without ever submitting...) but yeah, mimimal effort for basic submission is probably bigger now.
", "votes": null, "replies": [] } ] }, { "id": 446398, "author_name": "something4kag", "author_url": "", "post_date": "12/28/2018 02:03:37", "content": "I thought this competition was a good idea around the holidays and lead up to the SuperBowl, one for the armchair coaches and refs, possibly to have a team that is not all kernel coders, one for the powerpoint presentation fans (not me!) Anyways certainly paying more attention to punts in the game lately!
\n\nPeople have shared the rulebook \nhttps://operations.nfl.com/media/3277/2018-nfl-rulebook_final-version.pdf
\n\nIn another post it was mentioned another competition by the NFL using NGS data:\nmore data for big data bowl\nhttps://operations.nfl.com/the-game/big-data-bowl/
\n\nPossibly not so much to share in kernels as that may align too closely to the submission. But would be interesting to know more on people's approach to analysis or what they found interesting in the data. Perhaps in post-comp wrap.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "445355": "What a place, Kaggle. No matter how many competitions you have completed here you still get surprised by new challenges and formats.\n\nI find this new comp. format really interesting, with a good and not so good part:\n\n- The good part: It is more open than most other competitions in Kaggle. No defined metric to optimize, no clear and cut subset of reality to crunch with our boosting machines. \nJust some data related to the task at hand and all freedom available to \"do data science\".\n\n- The not so good part: A natural consequence of the nature of the competition -ideas and approaches will be competing here, not .csv files- With some exceptions, I see almost nothing of the usual exchange of ideas and kernels of other Kaggle competitions. It is just normal, probably will be solved after the end of competition when all solution kernels are made public. But in the meantime it makes the comp. threads and kernels much less active than usual.\n\nLet's hope the lack of activity is just a consequence of the competition format and there will be much more exchange of ideas when this ends.", "445470": "I agree that the sharing of Kernels has definitely been much smaller in this competition in comparison to the ML competitions which is unfortunate. Despite this I hope Kaggle continue these analytics based competitions as it doesn't require expert domain knowledge in machine learning and leaves the competition open to a much wider audience. I would like to know how many people have entered the competition just out of interest but they don't give any information about that.", "445488": "It's good for a novice like me in the fact that i feel like my \"solution\" can actually make a difference to the game without completely understanding the math to path of how i get there. Peace everyone.", "445528": "I feel like a big part of it is that a project like this has high barrier to entry in terms of time and knowledge. It requires more time and effort to write a report and make a presentation than it does to crank data through a boosted decision tree. You can't sit down and have a submission to a project like this ready in a couple of hours. This project would also be difficult to tackle if you didn't know much about the NFL.", "445975": "About bigger time/knowledge required, I wouldnt say so... I mean, if you think of other comps best kernels and solutions, many of them are perfecly formated, high quality presentations. \n\nAlso, about necessity of domain knowledge, my view is like in normal competitions, domain knowledge is a good hedge. But usually proper stats knowledge plus powerful tools proves to be what is needed. I have achieved decent results in many comps without prior domain knowledge and for sure the opposite can be true, domain knowledge without data science best practices dont take you too far.\n\nBut we'll see, really curious about what final set of skills will prove to be useful in this one, we will find out soon :-)", "446134": "You are focusing on the best kernels and solutions whereas my comment was about barriers to entry. Of course the best submissions in any competition are likely to involve significant time and effort. My point is that the minimum amount of time and domain knowledge required to put together a submission is higher for a project like this than it is for the typical prediction competition.", "446205": "Agreed, from that point of view of minimal barriers it can be true. (I was not thinking of minimal case, usually spend a lot of time tinkering with comps, even in the many cases that I just play with without ever submitting...) but yeah, mimimal effort for basic submission is probably bigger now.", "446398": "I thought this competition was a good idea around the holidays and lead up to the SuperBowl, one for the armchair coaches and refs, possibly to have a team that is not all kernel coders, one for the powerpoint presentation fans (not me!) Anyways certainly paying more attention to punts in the game lately!\n\nPeople have shared the rulebook \nhttps://operations.nfl.com/media/3277/2018-nfl-rulebook_final-version.pdf\n\nIn another post it was mentioned another competition by the NFL using NGS data:\nmore data for big data bowl\nhttps://operations.nfl.com/the-game/big-data-bowl/\n\nPossibly not so much to share in kernels as that may align too closely to the submission. But would be interesting to know more on people's approach to analysis or what they found interesting in the data. Perhaps in post-comp wrap." }, "source": "meta" }