In this competition, our task is to sort each ad_id by its probability, right?\nThen, I feel little strange and curious.
\n\nBecause outbrain must be most interested to know which ad will be clicked, more than orders of them.
\n\nSo, why didn't they ask kagglers to predict which one will be clicked or not?\nOr, what do you think about the reason why did they task us to order them?
\n\nSorry, my english might be unclear, I hope you get it.
\n\nWaiting for deep insight from pro-kagglers!!
", "messages": [ { "id": "147254", "postDate": "11/29/2016 18:42:39", "content": "In this competition, our task is to sort each ad_id by its probability, right?\nThen, I feel little strange and curious.
\n\nBecause outbrain must be most interested to know which ad will be clicked, more than orders of them.
\n\nSo, why didn't they ask kagglers to predict which one will be clicked or not?\nOr, what do you think about the reason why did they task us to order them?
\n\nSorry, my english might be unclear, I hope you get it.
\n\nWaiting for deep insight from pro-kagglers!!
", "rawMarkdown": "In this competition, our task is to sort each ad_id by its probability, right?\r\nThen, I feel little strange and curious. \r\n\r\nBecause outbrain must be most interested to know which ad will be clicked, more than orders of them.\r\n \r\nSo, why didn't they ask kagglers to predict which one will be clicked or not?\r\nOr, what do you think about the reason why did they task us to order them?\r\n\r\nSorry, my english might be unclear, I hope you get it.\r\n\r\nWaiting for deep insight from pro-kagglers!!", "votes": null }, { "id": "147270", "postDate": "11/29/2016 20:26:41", "content": "I don't know about business but yeah, increasing click through rate should be the final goal. One thing is to relocate ads based on predicted ordering.
", "rawMarkdown": "I don't know about business but yeah, increasing click through rate should be the final goal. One thing is to relocate ads based on predicted ordering.", "votes": null }, { "id": "147290", "postDate": "11/29/2016 21:32:00", "content": "We are asked to predict whether an ad is clicked indeed. The order has only to do with the ads we include in each submission, not the order the ads were clicked. This order matters for the evaluation, it gives more points to the submissions that had the correct ad clicked at the top of the 12 ads that it is possible to submit.
\n\nThe evaluation metric sounds unnatural at first, but it has its logic, which is explained here:\nMean Average Prediction
\n\nSo, for this problem we should come up with 12 ads per row of the test set. If the user had clicked only one ad and we had that ad as n.1 on our list of 12, we get a full point. If it was n.2 in our list we get only 0.5 points, etc. It is a bit more complicated if the user had clicked more than one ad, but the overall logic is the same: the highest on our list we have the correct answers, the highest our score will be.
", "rawMarkdown": "We are asked to predict whether an ad is clicked indeed. The order has only to do with the ads we include in each submission, not the order the ads were clicked. This order matters for the evaluation, it gives more points to the submissions that had the correct ad clicked at the top of the 12 ads that it is possible to submit.\r\n\r\nThe evaluation metric sounds unnatural at first, but it has its logic, which is explained here:\r\n[Mean Average Prediction][1]\r\n\r\n\r\nSo, for this problem we should come up with 12 ads per row of the test set. If the user had clicked only one ad and we had that ad as n.1 on our list of 12, we get a full point. If it was n.2 in our list we get only 0.5 points, etc. It is a bit more complicated if the user had clicked more than one ad, but the overall logic is the same: the highest on our list we have the correct answers, the highest our score will be.\r\n\r\n\r\n [1]: https://www.kaggle.com/wiki/MeanAveragePrecision", "votes": null }, { "id": "147452", "postDate": "11/30/2016 17:51:42", "content": "@rcarson
\n\nThanks for replying me!
\n\n@Panos
\n\nI see, I was misunderstanding. Now it's make sense. So I had to double check evaluation and metrics.
", "rawMarkdown": "rcarson \r\n\r\nThanks for replying me!\r\n\r\n@Panos \r\n\r\nI see, I was misunderstanding. Now it's make sense. So I had to double check evaluation and metrics.", "votes": null }, { "id": "148478", "postDate": "12/05/2016 07:22:23", "content": "@Psittacus The problem is to predict the probability of clicking an add, for each display_id. \nSorting makes sense because for each display_id you have a set if adds (ad_id), the \"recommendation\" is to sort by probability of getting clicked. \nI think the image with the example of their page makes it clear, at least from a conceptual standpoint.
", "rawMarkdown": "Psittacus The problem is to predict the probability of clicking an add, for each display_id. \r\nSorting makes sense because for each display_id you have a set if adds (ad_id), the \"recommendation\" is to sort by probability of getting clicked. \r\nI think the image with the example of their page makes it clear, at least from a conceptual standpoint.", "votes": null } ], "comments": [ { "id": 147270, "author_name": "jiweiliu", "author_url": "", "post_date": "11/29/2016 20:26:41", "content": "I don't know about business but yeah, increasing click through rate should be the final goal. One thing is to relocate ads based on predicted ordering.
", "votes": null, "replies": [] }, { "id": 147290, "author_name": "panosc", "author_url": "", "post_date": "11/29/2016 21:32:00", "content": "We are asked to predict whether an ad is clicked indeed. The order has only to do with the ads we include in each submission, not the order the ads were clicked. This order matters for the evaluation, it gives more points to the submissions that had the correct ad clicked at the top of the 12 ads that it is possible to submit.
\n\nThe evaluation metric sounds unnatural at first, but it has its logic, which is explained here:\nMean Average Prediction
\n\nSo, for this problem we should come up with 12 ads per row of the test set. If the user had clicked only one ad and we had that ad as n.1 on our list of 12, we get a full point. If it was n.2 in our list we get only 0.5 points, etc. It is a bit more complicated if the user had clicked more than one ad, but the overall logic is the same: the highest on our list we have the correct answers, the highest our score will be.
", "votes": null, "replies": [] }, { "id": 147452, "author_name": "mymkyt", "author_url": "", "post_date": "11/30/2016 17:51:42", "content": "@rcarson
\n\nThanks for replying me!
\n\n@Panos
\n\nI see, I was misunderstanding. Now it's make sense. So I had to double check evaluation and metrics.
", "votes": null, "replies": [] }, { "id": 148478, "author_name": "marbel", "author_url": "", "post_date": "12/05/2016 07:22:23", "content": "@Psittacus The problem is to predict the probability of clicking an add, for each display_id. \nSorting makes sense because for each display_id you have a set if adds (ad_id), the \"recommendation\" is to sort by probability of getting clicked. \nI think the image with the example of their page makes it clear, at least from a conceptual standpoint.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "147254": "In this competition, our task is to sort each ad_id by its probability, right?\r\nThen, I feel little strange and curious. \r\n\r\nBecause outbrain must be most interested to know which ad will be clicked, more than orders of them.\r\n \r\nSo, why didn't they ask kagglers to predict which one will be clicked or not?\r\nOr, what do you think about the reason why did they task us to order them?\r\n\r\nSorry, my english might be unclear, I hope you get it.\r\n\r\nWaiting for deep insight from pro-kagglers!!", "147270": "I don't know about business but yeah, increasing click through rate should be the final goal. One thing is to relocate ads based on predicted ordering.", "147290": "We are asked to predict whether an ad is clicked indeed. The order has only to do with the ads we include in each submission, not the order the ads were clicked. This order matters for the evaluation, it gives more points to the submissions that had the correct ad clicked at the top of the 12 ads that it is possible to submit.\r\n\r\nThe evaluation metric sounds unnatural at first, but it has its logic, which is explained here:\r\n[Mean Average Prediction][1]\r\n\r\n\r\nSo, for this problem we should come up with 12 ads per row of the test set. If the user had clicked only one ad and we had that ad as n.1 on our list of 12, we get a full point. If it was n.2 in our list we get only 0.5 points, etc. It is a bit more complicated if the user had clicked more than one ad, but the overall logic is the same: the highest on our list we have the correct answers, the highest our score will be.\r\n\r\n\r\n [1]: https://www.kaggle.com/wiki/MeanAveragePrecision", "147452": "rcarson \r\n\r\nThanks for replying me!\r\n\r\n@Panos \r\n\r\nI see, I was misunderstanding. Now it's make sense. So I had to double check evaluation and metrics.", "148478": "Psittacus The problem is to predict the probability of clicking an add, for each display_id. \r\nSorting makes sense because for each display_id you have a set if adds (ad_id), the \"recommendation\" is to sort by probability of getting clicked. \r\nI think the image with the example of their page makes it clear, at least from a conceptual standpoint." }, "source": "meta" }