They calculated target values and put them into the train dataset. And the competition is to find the target values of the test set and measure their distance to the target values which is calculated according to the formula. So in this competition, we are trying to find the formula how they calculated the target value.
\n\nTo make a real competition, I think, it should be like try to predict when customers really leave or when they stop using that credit card.
\n\nOtherwise, there is such an assumption that this formula is perfect and we are trying to get close to the results of formula.
", "messages": [ { "id": "438476", "postDate": "12/13/2018 19:02:27", "content": "They calculated target values and put them into the train dataset. And the competition is to find the target values of the test set and measure their distance to the target values which is calculated according to the formula. So in this competition, we are trying to find the formula how they calculated the target value.
\n\nTo make a real competition, I think, it should be like try to predict when customers really leave or when they stop using that credit card.
\n\nOtherwise, there is such an assumption that this formula is perfect and we are trying to get close to the results of formula.
", "rawMarkdown": "They calculated target values and put them into the train dataset. And the competition is to find the target values of the test set and measure their distance to the target values which is calculated according to the formula. So in this competition, we are trying to find the formula how they calculated the target value.\n\nTo make a real competition, I think, it should be like try to predict when customers really leave or when they stop using that credit card. \n\nOtherwise, there is such an assumption that this formula is perfect and we are trying to get close to the results of formula.", "votes": null }, { "id": "438632", "postDate": "12/14/2018 00:57:11", "content": "I don't think so. I think the loyalty score (which the say is basically a business metric that considered both future spending and retention as main components) is calculated by other way or their business experience. They want to know how the loyalty score is related with historical transactions.
", "rawMarkdown": "I don't think so. I think the **loyalty score** (which the say is basically a business metric that considered both future spending and retention as main components) is calculated by other way or their business experience. They want to know how the **loyalty score** is related with **historical transactions**.", "votes": null }, { "id": "438654", "postDate": "12/14/2018 01:47:38", "content": "I partially agree with you in that the goal is not to give the best recommendations to the card holders, but I think what we are doing here is trying to assess when Elo's recommendations/discounts to the customers are useful and when they are not. And in what seems to be the spirit of this competition, these recommendations are kept secret. So, I still don't consider it a particularly exciting task. I've written more about it in the beginning of this kernel: https://www.kaggle.com/batalov/making-sense-of-elo-data-eda
", "rawMarkdown": "I partially agree with you in that the goal is not to give the best recommendations to the card holders, but I think what we are doing here is trying to assess when Elo's recommendations/discounts to the customers are useful and when they are not. And in what seems to be the spirit of this competition, these recommendations are kept secret. So, I still don't consider it a particularly exciting task. I've written more about it in the beginning of this kernel: https://www.kaggle.com/batalov/making-sense-of-elo-data-eda", "votes": null }, { "id": "438746", "postDate": "12/14/2018 05:53:43", "content": "But as long as they want you to ger closer to the target value in evaluation, it means that there is no question about their calculation of target score. Then why do you need this competition? :)
", "rawMarkdown": "But as long as they want you to ger closer to the target value in evaluation, it means that there is no question about their calculation of target score. Then why do you need this competition? :)", "votes": null }, { "id": "439189", "postDate": "12/14/2018 21:31:58", "content": "Besides previous comments, please note , new features you create via feature engineering will help businesses to optimize for example churn Index(formula) or Loyalty Index. Winer algorithm(s) is not always the only purpose of the competitions, business experts will also take advantage of hidden features , data analysis and EDA and more .
", "rawMarkdown": "Besides previous comments, please note , new features you create via feature engineering will help businesses to optimize for example churn Index(formula) or Loyalty Index. Winer algorithm(s) is not always the only purpose of the competitions, business experts will also take advantage of hidden features , data analysis and EDA and more .", "votes": null }, { "id": "439255", "postDate": "12/15/2018 02:47:07", "content": "I believe it may be a reverse engineering competition, but some might even argue that ML as a whole is reverse engineering. I don't know that we have the information that the target is some kind of computation of the other values, so I don't think for sure we know we are reverse engineering their formula.
", "rawMarkdown": "I believe it may be a reverse engineering competition, but some might even argue that ML as a whole is reverse engineering. I don't know that we have the information that the target is some kind of computation of the other values, so I don't think for sure we know we are reverse engineering their formula.", "votes": null } ], "comments": [ { "id": 438632, "author_name": "jiazhuang", "author_url": "", "post_date": "12/14/2018 00:57:11", "content": "I don't think so. I think the loyalty score (which the say is basically a business metric that considered both future spending and retention as main components) is calculated by other way or their business experience. They want to know how the loyalty score is related with historical transactions.
", "votes": null, "replies": [ { "id": 438746, "author_name": "hozler", "author_url": "", "post_date": "12/14/2018 05:53:43", "content": "But as long as they want you to ger closer to the target value in evaluation, it means that there is no question about their calculation of target score. Then why do you need this competition? :)
", "votes": null, "replies": [] } ] }, { "id": 438654, "author_name": "batalov", "author_url": "", "post_date": "12/14/2018 01:47:38", "content": "I partially agree with you in that the goal is not to give the best recommendations to the card holders, but I think what we are doing here is trying to assess when Elo's recommendations/discounts to the customers are useful and when they are not. And in what seems to be the spirit of this competition, these recommendations are kept secret. So, I still don't consider it a particularly exciting task. I've written more about it in the beginning of this kernel: https://www.kaggle.com/batalov/making-sense-of-elo-data-eda
", "votes": null, "replies": [] }, { "id": 439189, "author_name": "arashnic", "author_url": "", "post_date": "12/14/2018 21:31:58", "content": "Besides previous comments, please note , new features you create via feature engineering will help businesses to optimize for example churn Index(formula) or Loyalty Index. Winer algorithm(s) is not always the only purpose of the competitions, business experts will also take advantage of hidden features , data analysis and EDA and more .
", "votes": null, "replies": [] }, { "id": 439255, "author_name": "charliethewalrus", "author_url": "", "post_date": "12/15/2018 02:47:07", "content": "I believe it may be a reverse engineering competition, but some might even argue that ML as a whole is reverse engineering. I don't know that we have the information that the target is some kind of computation of the other values, so I don't think for sure we know we are reverse engineering their formula.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "438476": "They calculated target values and put them into the train dataset. And the competition is to find the target values of the test set and measure their distance to the target values which is calculated according to the formula. So in this competition, we are trying to find the formula how they calculated the target value.\n\nTo make a real competition, I think, it should be like try to predict when customers really leave or when they stop using that credit card. \n\nOtherwise, there is such an assumption that this formula is perfect and we are trying to get close to the results of formula.", "438632": "I don't think so. I think the **loyalty score** (which the say is basically a business metric that considered both future spending and retention as main components) is calculated by other way or their business experience. They want to know how the **loyalty score** is related with **historical transactions**.", "438654": "I partially agree with you in that the goal is not to give the best recommendations to the card holders, but I think what we are doing here is trying to assess when Elo's recommendations/discounts to the customers are useful and when they are not. And in what seems to be the spirit of this competition, these recommendations are kept secret. So, I still don't consider it a particularly exciting task. I've written more about it in the beginning of this kernel: https://www.kaggle.com/batalov/making-sense-of-elo-data-eda", "438746": "But as long as they want you to ger closer to the target value in evaluation, it means that there is no question about their calculation of target score. Then why do you need this competition? :)", "439189": "Besides previous comments, please note , new features you create via feature engineering will help businesses to optimize for example churn Index(formula) or Loyalty Index. Winer algorithm(s) is not always the only purpose of the competitions, business experts will also take advantage of hidden features , data analysis and EDA and more .", "439255": "I believe it may be a reverse engineering competition, but some might even argue that ML as a whole is reverse engineering. I don't know that we have the information that the target is some kind of computation of the other values, so I don't think for sure we know we are reverse engineering their formula." }, "source": "meta" }