I had a new idea this year to hopefully simplify being able to create histograms that show your own submission compared to everybody else's for each matchup - the massive Google Sheet I was using previously was pretty hacky and required a fair bit of manual work during the tourney. It's a little early to be talking about this so may have to try to bump this come tourney time, but thought I'd share in case anybody else has ideas/concerns around this.
\n\nThe idea is that we basically have an opt-in shared Google Drive folder that anybody who wants histograms can upload their submission(s) to. Then you would be able to download this Drive folder with hopefully a decent sample size of submissions to your computer, unzip it, and run the python script I have in there to generate all the histograms.
\n\nThe python script just needs tweaked with the competition gender and year, the name(s) of your submission file(s) and some other optional tweaks (colors, number of bins, etc). I already created the shared Drive with random submissions to prove the concept. You can check it out here: \nhttps://drive.google.com/drive/folders/1m_MRW4xwdqyX5_LJ40tdZOcOBQOoNjoW?usp=sharing
\n\nI figure this keeps it completely opt-in which should be ok by Kaggle's policies, I hope. Then as long as nobody is too big of a troll, we should be able to get a reasonable idea which team we should be rooting for at game time!
", "messages": [ { "id": "766875", "postDate": "03/08/2020 21:59:55", "content": "I had a new idea this year to hopefully simplify being able to create histograms that show your own submission compared to everybody else's for each matchup - the massive Google Sheet I was using previously was pretty hacky and required a fair bit of manual work during the tourney. It's a little early to be talking about this so may have to try to bump this come tourney time, but thought I'd share in case anybody else has ideas/concerns around this.
\n\nThe idea is that we basically have an opt-in shared Google Drive folder that anybody who wants histograms can upload their submission(s) to. Then you would be able to download this Drive folder with hopefully a decent sample size of submissions to your computer, unzip it, and run the python script I have in there to generate all the histograms.
\n\nThe python script just needs tweaked with the competition gender and year, the name(s) of your submission file(s) and some other optional tweaks (colors, number of bins, etc). I already created the shared Drive with random submissions to prove the concept. You can check it out here: \nhttps://drive.google.com/drive/folders/1m_MRW4xwdqyX5_LJ40tdZOcOBQOoNjoW?usp=sharing
\n\nI figure this keeps it completely opt-in which should be ok by Kaggle's policies, I hope. Then as long as nobody is too big of a troll, we should be able to get a reasonable idea which team we should be rooting for at game time!
", "rawMarkdown": "I had a new idea this year to hopefully simplify being able to create histograms that show your own submission compared to everybody else's for each matchup - the massive Google Sheet I was using previously was pretty hacky and required a fair bit of manual work during the tourney. It's a little early to be talking about this so may have to try to ***bump*** this come tourney time, but thought I'd share in case anybody else has ideas/concerns around this.\n\nThe idea is that we basically have an opt-in shared Google Drive folder that anybody who wants histograms can upload their submission(s) to. Then you would be able to download this Drive folder with hopefully a decent sample size of submissions to your computer, unzip it, and run the python script I have in there to generate all the histograms.\n\nThe python script just needs tweaked with the competition gender and year, the name(s) of your submission file(s) and some other optional tweaks (colors, number of bins, etc). I already created the shared Drive with random submissions to prove the concept. You can check it out here: \nhttps://drive.google.com/drive/folders/1m_MRW4xwdqyX5_LJ40tdZOcOBQOoNjoW?usp=sharing\n\nI figure this keeps it completely opt-in which should be ok by Kaggle's policies, I hope. Then as long as nobody is too big of a troll, we should be able to get a reasonable idea which team we should be rooting for at game time!", "votes": null }, { "id": "768497", "postDate": "03/10/2020 21:46:55", "content": "This sounds like an interesting idea, though I don't know that showing individual scores would really give that much insight if they were kept anonymous anyway. What really matters is if your prediction is different from the crowd. I wonder if the Kaggle team would be willing to generate the average and standard deviation of all submissions for each game (in the standard submission format). This would be useful for following along and would still ensure that everyone's entry is kept anonymous.
", "rawMarkdown": "This sounds like an interesting idea, though I don't know that showing individual scores would really give that much insight if they were kept anonymous anyway. What really matters is if your prediction is different from the crowd. I wonder if the Kaggle team would be willing to generate the average and standard deviation of all submissions for each game (in the standard submission format). This would be useful for following along and would still ensure that everyone's entry is kept anonymous.", "votes": null }, { "id": "768552", "postDate": "03/11/2020 00:26:11", "content": "Not sure what you mean by individual scores, but this is what my idea generates which is pretty much what we generate every year in one form or another:\nhttps://drive.google.com/drive/folders/1Lpw-M0SOsJNghrqb_P1VmqzJxk3BIQ1j
\n\nIt obviously requires user participation, but the pro is that it doesn't require last minute manual work for our amazing hosts - which with two competitions now (mens and womens) is reasonable to assume they have lots going on come tournament time.
\n\nAs far as anonymity, I personally don't really care, and I suspect many others also wouldn't. Though I think in the past when all submissions were shared, people were able to essentially eliminate people from being able to win as they go which some people may have not liked. So this allows people who want to, to be anonymous but still get histograms.
\n\nEDIT: To clarify the above - you can be as anonymous as you want in this format - anywhere from completely transparent to a major troll (ideally at least translucent though :) ). Your submission file doesn't have to have your name or anything on it. You could even tweak your predictions slightly to ensure nobody can match your file directly to anything on the leaderboard.
", "rawMarkdown": "Not sure what you mean by individual scores, but this is what my idea generates which is pretty much what we generate every year in one form or another:\nhttps://drive.google.com/drive/folders/1Lpw-M0SOsJNghrqb_P1VmqzJxk3BIQ1j\n\nIt obviously requires user participation, but the pro is that it doesn't require last minute manual work for our amazing hosts - which with two competitions now (mens and womens) is reasonable to assume they have lots going on come tournament time.\n\nAs far as anonymity, I personally don't really care, and I suspect many others also wouldn't. Though I think in the past when all submissions were shared, people were able to essentially eliminate people from being able to win as they go which some people may have not liked. So this allows people who want to, to be anonymous but still get histograms.\n\nEDIT: To clarify the above - you can be as anonymous as you want in this format - anywhere from completely transparent to a major troll (ideally at least translucent though :) ). Your submission file doesn't have to have your name or anything on it. You could even tweak your predictions slightly to ensure nobody can match your file directly to anything on the leaderboard.", "votes": null }, { "id": "768556", "postDate": "03/11/2020 00:52:52", "content": "Was just thinking that we might be able to represent each distribution as a modeled distribution as opposed to showing the true submission distributions just so that it might not need to be opt-in. You're right though, that would be more work for the hosts.
\n\nEither way, count me in!
", "rawMarkdown": "Was just thinking that we might be able to represent each distribution as a modeled distribution as opposed to showing the true submission distributions just so that it might not need to be opt-in. You're right though, that would be more work for the hosts.\n\nEither way, count me in!", "votes": null } ], "comments": [ { "id": 768497, "author_name": "rsa013", "author_url": "", "post_date": "03/10/2020 21:46:55", "content": "This sounds like an interesting idea, though I don't know that showing individual scores would really give that much insight if they were kept anonymous anyway. What really matters is if your prediction is different from the crowd. I wonder if the Kaggle team would be willing to generate the average and standard deviation of all submissions for each game (in the standard submission format). This would be useful for following along and would still ensure that everyone's entry is kept anonymous.
", "votes": null, "replies": [ { "id": 768552, "author_name": "brandonhoeksema", "author_url": "", "post_date": "03/11/2020 00:26:11", "content": "Not sure what you mean by individual scores, but this is what my idea generates which is pretty much what we generate every year in one form or another:\nhttps://drive.google.com/drive/folders/1Lpw-M0SOsJNghrqb_P1VmqzJxk3BIQ1j
\n\nIt obviously requires user participation, but the pro is that it doesn't require last minute manual work for our amazing hosts - which with two competitions now (mens and womens) is reasonable to assume they have lots going on come tournament time.
\n\nAs far as anonymity, I personally don't really care, and I suspect many others also wouldn't. Though I think in the past when all submissions were shared, people were able to essentially eliminate people from being able to win as they go which some people may have not liked. So this allows people who want to, to be anonymous but still get histograms.
\n\nEDIT: To clarify the above - you can be as anonymous as you want in this format - anywhere from completely transparent to a major troll (ideally at least translucent though :) ). Your submission file doesn't have to have your name or anything on it. You could even tweak your predictions slightly to ensure nobody can match your file directly to anything on the leaderboard.
", "votes": null, "replies": [] }, { "id": 768556, "author_name": "rsa013", "author_url": "", "post_date": "03/11/2020 00:52:52", "content": "Was just thinking that we might be able to represent each distribution as a modeled distribution as opposed to showing the true submission distributions just so that it might not need to be opt-in. You're right though, that would be more work for the hosts.
\n\nEither way, count me in!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "766875": "I had a new idea this year to hopefully simplify being able to create histograms that show your own submission compared to everybody else's for each matchup - the massive Google Sheet I was using previously was pretty hacky and required a fair bit of manual work during the tourney. It's a little early to be talking about this so may have to try to ***bump*** this come tourney time, but thought I'd share in case anybody else has ideas/concerns around this.\n\nThe idea is that we basically have an opt-in shared Google Drive folder that anybody who wants histograms can upload their submission(s) to. Then you would be able to download this Drive folder with hopefully a decent sample size of submissions to your computer, unzip it, and run the python script I have in there to generate all the histograms.\n\nThe python script just needs tweaked with the competition gender and year, the name(s) of your submission file(s) and some other optional tweaks (colors, number of bins, etc). I already created the shared Drive with random submissions to prove the concept. You can check it out here: \nhttps://drive.google.com/drive/folders/1m_MRW4xwdqyX5_LJ40tdZOcOBQOoNjoW?usp=sharing\n\nI figure this keeps it completely opt-in which should be ok by Kaggle's policies, I hope. Then as long as nobody is too big of a troll, we should be able to get a reasonable idea which team we should be rooting for at game time!", "768497": "This sounds like an interesting idea, though I don't know that showing individual scores would really give that much insight if they were kept anonymous anyway. What really matters is if your prediction is different from the crowd. I wonder if the Kaggle team would be willing to generate the average and standard deviation of all submissions for each game (in the standard submission format). This would be useful for following along and would still ensure that everyone's entry is kept anonymous.", "768552": "Not sure what you mean by individual scores, but this is what my idea generates which is pretty much what we generate every year in one form or another:\nhttps://drive.google.com/drive/folders/1Lpw-M0SOsJNghrqb_P1VmqzJxk3BIQ1j\n\nIt obviously requires user participation, but the pro is that it doesn't require last minute manual work for our amazing hosts - which with two competitions now (mens and womens) is reasonable to assume they have lots going on come tournament time.\n\nAs far as anonymity, I personally don't really care, and I suspect many others also wouldn't. Though I think in the past when all submissions were shared, people were able to essentially eliminate people from being able to win as they go which some people may have not liked. So this allows people who want to, to be anonymous but still get histograms.\n\nEDIT: To clarify the above - you can be as anonymous as you want in this format - anywhere from completely transparent to a major troll (ideally at least translucent though :) ). Your submission file doesn't have to have your name or anything on it. You could even tweak your predictions slightly to ensure nobody can match your file directly to anything on the leaderboard.", "768556": "Was just thinking that we might be able to represent each distribution as a modeled distribution as opposed to showing the true submission distributions just so that it might not need to be opt-in. You're right though, that would be more work for the hosts.\n\nEither way, count me in!" }, "source": "meta" }