We have released a blog post on this Kaggle challenge, which contains more contexts and brief descriptions of the two benchmarks. I am posting it here so more people can see it:
\n\nhttp://engineeringblog.yelp.com/2015/12/yelp-restaurant-photo-classification-kaggle.html
\n\nYou are also welcomed to take a look at the other engineering posts we published to see the projects we are working on here at Yelp. For example here is a recent post on how we applied deep learning to understand Yelp photos:
\n\n", "messages": [ { "id": "103717", "postDate": "01/05/2016 19:39:34", "content": "We have released a blog post on this Kaggle challenge, which contains more contexts and brief descriptions of the two benchmarks. I am posting it here so more people can see it:
\n\nhttp://engineeringblog.yelp.com/2015/12/yelp-restaurant-photo-classification-kaggle.html
\n\nYou are also welcomed to take a look at the other engineering posts we published to see the projects we are working on here at Yelp. For example here is a recent post on how we applied deep learning to understand Yelp photos:
\n\n", "rawMarkdown": "We have released a blog post on this Kaggle challenge, which contains more contexts and brief descriptions of the two benchmarks. I am posting it here so more people can see it:\r\n\r\nhttp://engineeringblog.yelp.com/2015/12/yelp-restaurant-photo-classification-kaggle.html\r\n\r\nYou are also welcomed to take a look at the other engineering posts we published to see the projects we are working on here at Yelp. For example here is a recent post on how we applied deep learning to understand Yelp photos:\r\n\r\nhttp://engineeringblog.yelp.com/2015/10/how-we-use-deep-learning-to-classify-business-photos-at-yelp.html", "votes": null }, { "id": "103738", "postDate": "01/06/2016 01:27:52", "content": "Thank you!
", "rawMarkdown": "Thank you!", "votes": null }, { "id": "104624", "postDate": "01/14/2016 17:58:49", "content": "Could you share the code for the baseline models?
", "rawMarkdown": "Could you share the code for the baseline models?", "votes": null }, { "id": "104652", "postDate": "01/14/2016 23:27:53", "content": "Hi Zach the ideas of the baseline models have been explained in the post so we are not planning releasing the code at this time unless there is a strong reason for that.
", "rawMarkdown": "Hi Zach the ideas of the baseline models have been explained in the post so we are not planning releasing the code at this time unless there is a strong reason for that.", "votes": null }, { "id": "104725", "postDate": "01/15/2016 21:36:20", "content": "[quote=Wei-Hong Chuang;104652]
\n\nHi Zach the ideas of the baseline models have been explained in the post so we are not planning releasing the code at this time unless there is a strong reason for that.
\n\n[/quote]\nIt might be a good idea of releasing some baseline code and encourage more participants, considering this competition has no money prize but needs much GPU/coding work, while the 2nd data science bowl which uses similar techniques has 200k dollars cash.
", "rawMarkdown": "[quote=Wei-Hong Chuang;104652]\r\n\r\nHi Zach the ideas of the baseline models have been explained in the post so we are not planning releasing the code at this time unless there is a strong reason for that.\r\n\r\n[/quote]\r\nIt might be a good idea of releasing some baseline code and encourage more participants, considering this competition has no money prize but needs much GPU/coding work, while the 2nd data science bowl which uses similar techniques has 200k dollars cash.", "votes": null }, { "id": "105560", "postDate": "01/24/2016 19:36:09", "content": "@phunter, Agree, currently, I can not figure out how to map the predicted labels to the business id, even though locally the mean F1 seems good
", "rawMarkdown": "phunter, Agree, currently, I can not figure out how to map the predicted labels to the business id, even though locally the mean F1 seems good", "votes": null }, { "id": "105701", "postDate": "01/25/2016 22:46:20", "content": "Hi, although we decided not to release the baseline code (and many participants have surpassed it), we can try to help if you have very specific questions.
", "rawMarkdown": "Hi, although we decided not to release the baseline code (and many participants have surpassed it), we can try to help if you have very specific questions.", "votes": null }, { "id": "105913", "postDate": "01/27/2016 12:16:04", "content": "I have the same question. Say that we have already built up a multi-label ConvNets that predicts the attribute labels. Then, how to map the predicted labels to business ids? Better with a end-to-end learning fashion?
", "rawMarkdown": "I have the same question. Say that we have already built up a multi-label ConvNets that predicts the attribute labels. Then, how to map the predicted labels to business ids? Better with a end-to-end learning fashion?", "votes": null }, { "id": "106180", "postDate": "01/28/2016 23:41:35", "content": "The challenge is a multi-instance problem: each business has one set of labels and hundreds of photos. It is not a trivial task to map your single-photo predictions back the the per-business level, and it is part of the challenge to figure out the best way to do the mapping if you follow this path. There is no one single best solution here.
", "rawMarkdown": "The challenge is a multi-instance problem: each business has one set of labels and hundreds of photos. It is not a trivial task to map your single-photo predictions back the the per-business level, and it is part of the challenge to figure out the best way to do the mapping if you follow this path. There is no one single best solution here.", "votes": null }, { "id": "107558", "postDate": "02/11/2016 06:53:26", "content": "", "rawMarkdown": "", "votes": null }, { "id": "108834", "postDate": "02/20/2016 11:34:20", "content": "Hey !!!
\n\nDo we need to predict the top three labels for a business_id as in the sample_submission.csv file or any threshold for label to match with the business_id ?
\n\nbusiness_id labels\n003sg 1 2 3\nAlso , how evaluation score will be calculated ?
", "rawMarkdown": "Hey !!!\r\n\r\nDo we need to predict the top three labels for a business_id as in the sample_submission.csv file or any threshold for label to match with the business_id ? \r\n\r\n business_id labels\r\n 003sg 1 2 3\r\n\r\nAlso , how evaluation score will be calculated ?", "votes": null }, { "id": "134523", "postDate": "09/07/2016 03:02:35", "content": "Thanks for sharing. It's really useful.
", "rawMarkdown": "Thanks for sharing. It's really useful.", "votes": null } ], "comments": [ { "id": 103738, "author_name": "superds", "author_url": "", "post_date": "01/06/2016 01:27:52", "content": "Thank you!
", "votes": null, "replies": [] }, { "id": 104624, "author_name": "zachmayer", "author_url": "", "post_date": "01/14/2016 17:58:49", "content": "Could you share the code for the baseline models?
", "votes": null, "replies": [] }, { "id": 104652, "author_name": "whchuang", "author_url": "", "post_date": "01/14/2016 23:27:53", "content": "Hi Zach the ideas of the baseline models have been explained in the post so we are not planning releasing the code at this time unless there is a strong reason for that.
", "votes": null, "replies": [] }, { "id": 104725, "author_name": "phunter", "author_url": "", "post_date": "01/15/2016 21:36:20", "content": "[quote=Wei-Hong Chuang;104652]
\n\nHi Zach the ideas of the baseline models have been explained in the post so we are not planning releasing the code at this time unless there is a strong reason for that.
\n\n[/quote]\nIt might be a good idea of releasing some baseline code and encourage more participants, considering this competition has no money prize but needs much GPU/coding work, while the 2nd data science bowl which uses similar techniques has 200k dollars cash.
", "votes": null, "replies": [] }, { "id": 105560, "author_name": "usixuz", "author_url": "", "post_date": "01/24/2016 19:36:09", "content": "@phunter, Agree, currently, I can not figure out how to map the predicted labels to the business id, even though locally the mean F1 seems good
", "votes": null, "replies": [] }, { "id": 105701, "author_name": "whchuang", "author_url": "", "post_date": "01/25/2016 22:46:20", "content": "Hi, although we decided not to release the baseline code (and many participants have surpassed it), we can try to help if you have very specific questions.
", "votes": null, "replies": [] }, { "id": 105913, "author_name": "pengpai", "author_url": "", "post_date": "01/27/2016 12:16:04", "content": "I have the same question. Say that we have already built up a multi-label ConvNets that predicts the attribute labels. Then, how to map the predicted labels to business ids? Better with a end-to-end learning fashion?
", "votes": null, "replies": [] }, { "id": 106180, "author_name": "fcchou", "author_url": "", "post_date": "01/28/2016 23:41:35", "content": "The challenge is a multi-instance problem: each business has one set of labels and hundreds of photos. It is not a trivial task to map your single-photo predictions back the the per-business level, and it is part of the challenge to figure out the best way to do the mapping if you follow this path. There is no one single best solution here.
", "votes": null, "replies": [] }, { "id": 107558, "author_name": "yardstick17", "author_url": "", "post_date": "02/11/2016 06:53:26", "content": "", "votes": null, "replies": [] }, { "id": 108834, "author_name": "yardstick17", "author_url": "", "post_date": "02/20/2016 11:34:20", "content": "Hey !!!
\n\nDo we need to predict the top three labels for a business_id as in the sample_submission.csv file or any threshold for label to match with the business_id ?
\n\nbusiness_id labels\n003sg 1 2 3\nAlso , how evaluation score will be calculated ?
", "votes": null, "replies": [] }, { "id": 134523, "author_name": "zzwcsong", "author_url": "", "post_date": "09/07/2016 03:02:35", "content": "Thanks for sharing. It's really useful.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "103717": "We have released a blog post on this Kaggle challenge, which contains more contexts and brief descriptions of the two benchmarks. I am posting it here so more people can see it:\r\n\r\nhttp://engineeringblog.yelp.com/2015/12/yelp-restaurant-photo-classification-kaggle.html\r\n\r\nYou are also welcomed to take a look at the other engineering posts we published to see the projects we are working on here at Yelp. For example here is a recent post on how we applied deep learning to understand Yelp photos:\r\n\r\nhttp://engineeringblog.yelp.com/2015/10/how-we-use-deep-learning-to-classify-business-photos-at-yelp.html", "103738": "Thank you!", "104624": "Could you share the code for the baseline models?", "104652": "Hi Zach the ideas of the baseline models have been explained in the post so we are not planning releasing the code at this time unless there is a strong reason for that.", "104725": "[quote=Wei-Hong Chuang;104652]\r\n\r\nHi Zach the ideas of the baseline models have been explained in the post so we are not planning releasing the code at this time unless there is a strong reason for that.\r\n\r\n[/quote]\r\nIt might be a good idea of releasing some baseline code and encourage more participants, considering this competition has no money prize but needs much GPU/coding work, while the 2nd data science bowl which uses similar techniques has 200k dollars cash.", "105560": "phunter, Agree, currently, I can not figure out how to map the predicted labels to the business id, even though locally the mean F1 seems good", "105701": "Hi, although we decided not to release the baseline code (and many participants have surpassed it), we can try to help if you have very specific questions.", "105913": "I have the same question. Say that we have already built up a multi-label ConvNets that predicts the attribute labels. Then, how to map the predicted labels to business ids? Better with a end-to-end learning fashion?", "106180": "The challenge is a multi-instance problem: each business has one set of labels and hundreds of photos. It is not a trivial task to map your single-photo predictions back the the per-business level, and it is part of the challenge to figure out the best way to do the mapping if you follow this path. There is no one single best solution here.", "107558": "", "108834": "Hey !!!\r\n\r\nDo we need to predict the top three labels for a business_id as in the sample_submission.csv file or any threshold for label to match with the business_id ? \r\n\r\n business_id labels\r\n 003sg 1 2 3\r\n\r\nAlso , how evaluation score will be calculated ?", "134523": "Thanks for sharing. It's really useful." }, "source": "meta" }