I am probably being too cautious but here it is anyway...
\n\nI contacted a fellow contestant yesterday asking if they would like to team up with me. He/She replied today and decided to share the following (see below in quotes) in his/her reply to my email. I believe he/she did not mean any harm but I felt it should be shared publicly.
\n\n> ...I could share with you some of my solutions. For me, there's no magic tricks to get score boost. A larger image size 224x224 can help you simply get a lb 0.98 high score. Another huge improvement comes from choosing cv2.INTER_AREA interpolation :) The rest are simply tuning for architecture and augmentations. My current score is also a single model and my training hasn't ended yet. I'm sure it's quite easy for you to boost your lb score.
\n\n@sohier, please respond to let me know if I am in any trouble for this thing that is not my fault. The sequence of events today is, I read the email -> came here to make this post -> then I replied to his email with a link to this post -> copied @sohier in the email. Thanks
\n\nShare your comments below.
", "messages": [ { "id": 767582, "postDate": "2020-03-09T20:45:37.310Z", "content": "I am probably being too cautious but here it is anyway...
\n\nI contacted a fellow contestant yesterday asking if they would like to team up with me. He/She replied today and decided to share the following (see below in quotes) in his/her reply to my email. I believe he/she did not mean any harm but I felt it should be shared publicly.
\n\n> ...I could share with you some of my solutions. For me, there's no magic tricks to get score boost. A larger image size 224x224 can help you simply get a lb 0.98 high score. Another huge improvement comes from choosing cv2.INTER_AREA interpolation :) The rest are simply tuning for architecture and augmentations. My current score is also a single model and my training hasn't ended yet. I'm sure it's quite easy for you to boost your lb score.
\n\n@sohier, please respond to let me know if I am in any trouble for this thing that is not my fault. The sequence of events today is, I read the email -> came here to make this post -> then I replied to his email with a link to this post -> copied @sohier in the email. Thanks
\n\nShare your comments below.
", "rawMarkdown": "I am probably being too cautious but here it is anyway...\n\nI contacted a fellow contestant yesterday asking if they would like to team up with me. He/She replied today and decided to share the following (see below in quotes) in his/her reply to my email. I believe he/she did not mean any harm but I felt it should be shared publicly.\n\n> ...I could share with you some of my solutions. For me, there's no magic tricks to get score boost. A larger image size 224x224 can help you simply get a lb 0.98 high score. Another huge improvement comes from choosing cv2.INTER_AREA interpolation :) The rest are simply tuning for architecture and augmentations. My current score is also a single model and my training hasn't ended yet. I'm sure it's quite easy for you to boost your lb score.\n\n@sohier, please respond to let me know if I am in any trouble for this thing that is not my fault. The sequence of events today is, I read the email -> came here to make this post -> then I replied to his email with a link to this post -> copied @sohier in the email. Thanks\n\nShare your comments below.\n", "votes": 22 }, { "id": 768001, "postDate": "2020-03-10T10:58:58.287Z", "content": "You found a very effective way of fighting private sharing: public reshare. Well done.
", "rawMarkdown": "You found a very effective way of fighting private sharing: public reshare. Well done.", "votes": 13, "replies": [ { "id": 768236, "postDate": "2020-03-10T15:09:39.690Z", "content": "Thanks @cpmpml. Doing my best to worth belonging to the responsible Kaggle community:-)
", "rawMarkdown": "Thanks @cpmpml. Doing my best to worth belonging to the responsible Kaggle community:-)", "votes": 4 } ] }, { "id": 768280, "postDate": "2020-03-10T15:49:20.587Z", "content": "i don't think it is any problem at all.
\n\nsometimes, i also received invitation e-mail with brief description of his/her methods. I think the intention is for me to judge if my method is similar or not (hence would there be gain in teaming up/ensembling/ etc...). Sometimes poeple also discloses their gpu or computer resources, ideas, etc...
\n\nthe intention is not for private sharing but rather to provide information for decision for team forming or not.
\n\nsince the intention is ok, i think it is no problem
\n\non a side note, being in the field of computer-vision for a long time, my feeling is that implementation details are probably more important then the idea. so actually nothing is actually disclosed at all.
", "rawMarkdown": "i don't think it is any problem at all.\n\nsometimes, i also received invitation e-mail with brief description of his/her methods. I think the intention is for me to judge if my method is similar or not (hence would there be gain in teaming up/ensembling/ etc...). Sometimes poeple also discloses their gpu or computer resources, ideas, etc...\n\nthe intention is not for private sharing but rather to provide information for decision for team forming or not.\n\nsince the intention is ok, i think it is no problem\n\n---\n\non a side note, being in the field of computer-vision for a long time, my feeling is that implementation details are probably more important then the idea. so actually nothing is actually disclosed at all.", "votes": 13 }, { "id": 767613, "postDate": "2020-03-09T22:18:25.020Z", "content": "@sheriytm we generally don't comment on specific cases, especially before the close of a competition. What I can say is that we know a certain amount of high level discussion of methods is necessary and appropriate for discussing whether to form teams.
", "rawMarkdown": "@sheriytm we generally don't comment on specific cases, especially before the close of a competition. What I can say is that we know a certain amount of high level discussion of methods is necessary and appropriate for discussing whether to form teams.", "votes": 5, "replies": [ { "id": 767614, "postDate": "2020-03-09T22:21:31.683Z", "content": "Thanks @sohier. This is nothing I did not know but I just felt uncomfortable hence the post:-)
", "rawMarkdown": "Thanks @sohier. This is nothing I did not know but I just felt uncomfortable hence the post:-)", "votes": 3 }, { "id": 767618, "postDate": "2020-03-09T22:40:06.907Z", "content": "\nWhat I can say is that we know a certain amount of high level discussion of methods is necessary and appropriate for discussing whether to form teams.\n
It's good to know. \nI was thinking we had to share some amount of information about the model in kaggle forum first in order to talk about the model for team merge.
", "rawMarkdown": "```\nWhat I can say is that we know a certain amount of high level discussion of methods is necessary and appropriate for discussing whether to form teams.\n```\n\nIt's good to know. \nI was thinking we had to share some amount of information about the model in kaggle forum first in order to talk about the model for team merge. ", "votes": 2 }, { "id": 767861, "postDate": "2020-03-10T07:15:18.127Z", "content": "@sohier This comment is really vague. What does that mean? What does high level mean? The example from above obviously is not high level and goes quite into detail. To me it sounds like private sharing.
", "rawMarkdown": "@sohier This comment is really vague. What does that mean? What does high level mean? The example from above obviously is not high level and goes quite into detail. To me it sounds like private sharing.", "votes": 2 }, { "id": 767954, "postDate": "2020-03-10T09:46:24.953Z", "content": "@philippsinger, I thought it is vague too. I did what I had to do in this situation with the hope that people can learn a little bit more about what constitutes private sharing.
", "rawMarkdown": "@philippsinger, I thought it is vague too. I did what I had to do in this situation with the hope that people can learn a little bit more about what constitutes private sharing.", "votes": 1 }, { "id": 767995, "postDate": "2020-03-10T10:55:39.657Z", "content": "Well, it is not private sharing anymore now ;)
", "rawMarkdown": "Well, it is not private sharing anymore now ;)\n", "votes": 4 } ] }, { "id": 780689, "postDate": "2020-03-20T14:17:54.460Z", "content": "Thanks everyone for chiming in and I can confirm now that neither of us is in trouble as a result of the subject matter posted.
", "rawMarkdown": "Thanks everyone for chiming in and I can confirm now that neither of us is in trouble as a result of the subject matter posted." }, { "id": 770223, "postDate": "2020-03-12T17:43:46.333Z", "content": "It is interesting to know that resizing with INTER_AREA interpolation makes that much of a difference...
", "rawMarkdown": "It is interesting to know that resizing with INTER_AREA interpolation makes that much of a difference...", "replies": [ { "id": 770726, "postDate": "2020-03-13T10:21:17.093Z", "content": "If it really does..
", "rawMarkdown": "If it really does..", "votes": 2 } ] }, { "id": 769613, "postDate": "2020-03-12T04:30:29.380Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 768001, "author_name": "CPMP", "author_url": "", "post_date": "2020-03-10T10:58:58.287000", "content": "You found a very effective way of fighting private sharing: public reshare. Well done.
", "votes": 13, "replies": [ { "id": 768236, "author_name": "YaGana Sheriff-Hussaini", "author_url": "", "post_date": "2020-03-10T15:09:39.690000", "content": "Thanks @cpmpml. Doing my best to worth belonging to the responsible Kaggle community:-)
", "votes": 4, "replies": [] } ] }, { "id": 768280, "author_name": "hengck23", "author_url": "", "post_date": "2020-03-10T15:49:20.587000", "content": "i don't think it is any problem at all.
\n\nsometimes, i also received invitation e-mail with brief description of his/her methods. I think the intention is for me to judge if my method is similar or not (hence would there be gain in teaming up/ensembling/ etc...). Sometimes poeple also discloses their gpu or computer resources, ideas, etc...
\n\nthe intention is not for private sharing but rather to provide information for decision for team forming or not.
\n\nsince the intention is ok, i think it is no problem
\n\non a side note, being in the field of computer-vision for a long time, my feeling is that implementation details are probably more important then the idea. so actually nothing is actually disclosed at all.
", "votes": 13, "replies": [] }, { "id": 767613, "author_name": "Sohier Dane", "author_url": "", "post_date": "2020-03-09T22:18:25.020000", "content": "@sheriytm we generally don't comment on specific cases, especially before the close of a competition. What I can say is that we know a certain amount of high level discussion of methods is necessary and appropriate for discussing whether to form teams.
", "votes": 5, "replies": [ { "id": 767614, "author_name": "YaGana Sheriff-Hussaini", "author_url": "", "post_date": "2020-03-09T22:21:31.683000", "content": "Thanks @sohier. This is nothing I did not know but I just felt uncomfortable hence the post:-)
", "votes": 3, "replies": [] }, { "id": 767618, "author_name": "Appian", "author_url": "", "post_date": "2020-03-09T22:40:06.907000", "content": "\nWhat I can say is that we know a certain amount of high level discussion of methods is necessary and appropriate for discussing whether to form teams.\n
It's good to know. \nI was thinking we had to share some amount of information about the model in kaggle forum first in order to talk about the model for team merge.
", "votes": 2, "replies": [] }, { "id": 767861, "author_name": "Psi", "author_url": "", "post_date": "2020-03-10T07:15:18.127000", "content": "@sohier This comment is really vague. What does that mean? What does high level mean? The example from above obviously is not high level and goes quite into detail. To me it sounds like private sharing.
", "votes": 2, "replies": [] }, { "id": 767954, "author_name": "YaGana Sheriff-Hussaini", "author_url": "", "post_date": "2020-03-10T09:46:24.953000", "content": "@philippsinger, I thought it is vague too. I did what I had to do in this situation with the hope that people can learn a little bit more about what constitutes private sharing.
", "votes": 1, "replies": [] }, { "id": 767995, "author_name": "CPMP", "author_url": "", "post_date": "2020-03-10T10:55:39.657000", "content": "Well, it is not private sharing anymore now ;)
", "votes": 4, "replies": [] } ] }, { "id": 780689, "author_name": "YaGana Sheriff-Hussaini", "author_url": "", "post_date": "2020-03-20T14:17:54.460000", "content": "Thanks everyone for chiming in and I can confirm now that neither of us is in trouble as a result of the subject matter posted.
", "votes": 0, "replies": [] }, { "id": 770223, "author_name": "YL", "author_url": "", "post_date": "2020-03-12T17:43:46.333000", "content": "It is interesting to know that resizing with INTER_AREA interpolation makes that much of a difference...
", "votes": 0, "replies": [ { "id": 770726, "author_name": "Vlad Vaduva", "author_url": "", "post_date": "2020-03-13T10:21:17.093000", "content": "If it really does..
", "votes": 2, "replies": [] } ] }, { "id": 769613, "author_name": "", "author_url": "", "post_date": "2020-03-12T04:30:29.380000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "767582": "I am probably being too cautious but here it is anyway...\n\nI contacted a fellow contestant yesterday asking if they would like to team up with me. He/She replied today and decided to share the following (see below in quotes) in his/her reply to my email. I believe he/she did not mean any harm but I felt it should be shared publicly.\n\n> ...I could share with you some of my solutions. For me, there's no magic tricks to get score boost. A larger image size 224x224 can help you simply get a lb 0.98 high score. Another huge improvement comes from choosing cv2.INTER_AREA interpolation :) The rest are simply tuning for architecture and augmentations. My current score is also a single model and my training hasn't ended yet. I'm sure it's quite easy for you to boost your lb score.\n\n@sohier, please respond to let me know if I am in any trouble for this thing that is not my fault. The sequence of events today is, I read the email -> came here to make this post -> then I replied to his email with a link to this post -> copied @sohier in the email. Thanks\n\nShare your comments below.\n", "768001": "You found a very effective way of fighting private sharing: public reshare. Well done.", "768280": "i don't think it is any problem at all.\n\nsometimes, i also received invitation e-mail with brief description of his/her methods. I think the intention is for me to judge if my method is similar or not (hence would there be gain in teaming up/ensembling/ etc...). Sometimes poeple also discloses their gpu or computer resources, ideas, etc...\n\nthe intention is not for private sharing but rather to provide information for decision for team forming or not.\n\nsince the intention is ok, i think it is no problem\n\n---\n\non a side note, being in the field of computer-vision for a long time, my feeling is that implementation details are probably more important then the idea. so actually nothing is actually disclosed at all.", "767613": "@sheriytm we generally don't comment on specific cases, especially before the close of a competition. What I can say is that we know a certain amount of high level discussion of methods is necessary and appropriate for discussing whether to form teams.", "780689": "Thanks everyone for chiming in and I can confirm now that neither of us is in trouble as a result of the subject matter posted.", "770223": "It is interesting to know that resizing with INTER_AREA interpolation makes that much of a difference...", "769613": "" } }