I can't stop thinking about the reasons for this competition.\nReally. I mean what is the point of predicting something abstract and unknown?\nWhat they say about target:
\n\nSo, the company already has some formula to calculate abstract score representing loyalty.\nMaybe, it's based on spendings, monthly spendings, sliding average or smth similar.\nLet's represent this equation: (x+y)/z = t (where t- target, and xyz - some spendings, active days, amount of purchases, etc)
\n\nMore naturally we'd predict X, Y and Z rather than T.
\n\nOtherwise, it's rather trying to guess already known formula than a predictive modeling
\n\nYour thoughts?
", "messages": [ { "id": 432911, "postDate": "2018-12-04T13:09:38.110Z", "content": "I can't stop thinking about the reasons for this competition.\nReally. I mean what is the point of predicting something abstract and unknown?\nWhat they say about target:
\n\nSo, the company already has some formula to calculate abstract score representing loyalty.\nMaybe, it's based on spendings, monthly spendings, sliding average or smth similar.\nLet's represent this equation: (x+y)/z = t (where t- target, and xyz - some spendings, active days, amount of purchases, etc)
\n\nMore naturally we'd predict X, Y and Z rather than T.
\n\nOtherwise, it's rather trying to guess already known formula than a predictive modeling
\n\nYour thoughts?
", "rawMarkdown": "I can't stop thinking about the reasons for this competition.\nReally. I mean what is the point of predicting something abstract and unknown?\nWhat they say about target:\n\n - 'Loyalty numerical score calculated 2 months after historical and evaluation period'\n - 'You are predicting a loyalty score for each card_id represented in test.csv and sample_submission.csv.'\n\nSo, the company already has some formula to calculate abstract score representing loyalty.\nMaybe, it's based on spendings, monthly spendings, sliding average or smth similar.\nLet's represent this equation: (x+y)/z = t (where t- target, and xyz - some spendings, active days, amount of purchases, etc)\n\nMore naturally we'd predict X, Y and Z rather than T. \n\nOtherwise, it's rather trying to guess already known formula than a predictive modeling\n\nYour thoughts?\n", "votes": 6 }, { "id": 432948, "postDate": "2018-12-04T14:10:10.093Z", "content": "We are predicting a loyalty score. But loyal to what? To ELO, to the merchants, to the products, to the city?
", "rawMarkdown": "We are predicting a loyalty score. But loyal to what? To ELO, to the merchants, to the products, to the city?", "votes": 3, "replies": [ { "id": 433003, "postDate": "2018-12-04T14:58:11.037Z", "content": "Yes, this has been my question for a week now. Loyal to whom? Can people have multiple cards so that the rows are no longer independent?
", "rawMarkdown": "Yes, this has been my question for a week now. Loyal to whom? Can people have multiple cards so that the rows are no longer independent? ", "votes": 5 }, { "id": 433009, "postDate": "2018-12-04T15:00:32.263Z", "content": "Ha!\nreally the case, i didn't think about it :)
", "rawMarkdown": "Ha!\nreally the case, i didn't think about it :)", "votes": 1 }, { "id": 433047, "postDate": "2018-12-04T15:56:41.083Z", "content": "Loyalty implies repeated customer to the merchant. If the customers buy, using another non-Elo card then, it is not seen in Elo's register. The issue is that you don't have information about the promotions, which is the main question in the competition. You can have some customers with high loyality value, regardless of the promotion, i.e. they will take anything anyway. Others, who don't take anything and most of them, in-between. However, where is the data about the promotion, regardless if the customer took the offer or not? I went to see other equivalent data, outside Kaggle, and there is NO way to know it without further desciption by Elo.
\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/73213
", "rawMarkdown": "Loyalty implies repeated customer to the merchant. If the customers buy, using another non-Elo card then, it is not seen in Elo's register. The issue is that you don't have information about the promotions, which is the main question in the competition. You can have some customers with high loyality value, regardless of the promotion, i.e. they will take anything anyway. Others, who don't take anything and most of them, in-between. However, where is the data about the promotion, regardless if the customer took the offer or not? I went to see other equivalent data, outside Kaggle, and there is NO way to know it without further desciption by Elo.\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/73213", "votes": 3 } ] }, { "id": 435379, "postDate": "2018-12-07T23:39:01.870Z", "content": "completely agreed. my personal experience with the (x+y)/z = t situation is to predict x, y, z rather than t.\nthat way we can minimize the prediction standard error of t_hat..
", "rawMarkdown": "completely agreed. my personal experience with the (x+y)/z = t situation is to predict x, y, z rather than t.\nthat way we can minimize the prediction standard error of t_hat.." }, { "id": 432942, "postDate": "2018-12-04T14:04:21.340Z", "content": "I agree with you. It is puzzling. One reason of the competicion could be to find target values before evaluating period.
\n\nImagine it is necessary this period with its transactions to calculate a past loyalty. To predict loyalty would make sense. Are you agree? But nothing is clear.
", "rawMarkdown": "I agree with you. It is puzzling. One reason of the competicion could be to find target values before evaluating period. \n\nImagine it is necessary this period with its transactions to calculate a past loyalty. To predict loyalty would make sense. Are you agree? But nothing is clear. ", "votes": 2, "isDeleted": true, "replies": [ { "id": 432987, "postDate": "2018-12-04T14:44:53.450Z", "content": "indeed, the reason lies here. Predict the loyalty before evaluation.\nhowever, this makes even less sense, because of errors
", "rawMarkdown": "indeed, the reason lies here. Predict the loyalty before evaluation.\nhowever, this makes even less sense, because of errors" }, { "id": 433162, "postDate": "2018-12-04T19:21:29.960Z", "content": "Errors? If the model is \"perfect\" it will predict the same results. And two months before...
", "rawMarkdown": "Errors? If the model is \"perfect\" it will predict the same results. And two months before...", "isDeleted": true } ] }, { "id": 432918, "postDate": "2018-12-04T13:16:53.693Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 432921, "postDate": "2018-12-04T13:24:48.203Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 432948, "author_name": "corner200", "author_url": "", "post_date": "2018-12-04T14:10:10.093000", "content": "We are predicting a loyalty score. But loyal to what? To ELO, to the merchants, to the products, to the city?
", "votes": 3, "replies": [ { "id": 433003, "author_name": "predictor", "author_url": "", "post_date": "2018-12-04T14:58:11.037000", "content": "Yes, this has been my question for a week now. Loyal to whom? Can people have multiple cards so that the rows are no longer independent?
", "votes": 5, "replies": [] }, { "id": 433009, "author_name": "Dmytro Samchuk", "author_url": "", "post_date": "2018-12-04T15:00:32.263000", "content": "Ha!\nreally the case, i didn't think about it :)
", "votes": 1, "replies": [] }, { "id": 433047, "author_name": "Raúl Machado", "author_url": "", "post_date": "2018-12-04T15:56:41.083000", "content": "Loyalty implies repeated customer to the merchant. If the customers buy, using another non-Elo card then, it is not seen in Elo's register. The issue is that you don't have information about the promotions, which is the main question in the competition. You can have some customers with high loyality value, regardless of the promotion, i.e. they will take anything anyway. Others, who don't take anything and most of them, in-between. However, where is the data about the promotion, regardless if the customer took the offer or not? I went to see other equivalent data, outside Kaggle, and there is NO way to know it without further desciption by Elo.
\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/73213
", "votes": 3, "replies": [] } ] }, { "id": 435379, "author_name": "Elliot", "author_url": "", "post_date": "2018-12-07T23:39:01.870000", "content": "completely agreed. my personal experience with the (x+y)/z = t situation is to predict x, y, z rather than t.\nthat way we can minimize the prediction standard error of t_hat..
", "votes": 0, "replies": [] }, { "id": 432942, "author_name": "", "author_url": "", "post_date": "2018-12-04T14:04:21.340000", "content": "I agree with you. It is puzzling. One reason of the competicion could be to find target values before evaluating period.
\n\nImagine it is necessary this period with its transactions to calculate a past loyalty. To predict loyalty would make sense. Are you agree? But nothing is clear.
", "votes": 2, "replies": [ { "id": 432987, "author_name": "Dmytro Samchuk", "author_url": "", "post_date": "2018-12-04T14:44:53.450000", "content": "indeed, the reason lies here. Predict the loyalty before evaluation.\nhowever, this makes even less sense, because of errors
", "votes": 0, "replies": [] }, { "id": 433162, "author_name": "", "author_url": "", "post_date": "2018-12-04T19:21:29.960000", "content": "Errors? If the model is \"perfect\" it will predict the same results. And two months before...
", "votes": 0, "replies": [] } ] }, { "id": 432918, "author_name": "", "author_url": "", "post_date": "2018-12-04T13:16:53.693000", "content": "", "votes": 0, "replies": [ { "id": 432921, "author_name": "", "author_url": "", "post_date": "2018-12-04T13:24:48.203000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "432911": "I can't stop thinking about the reasons for this competition.\nReally. I mean what is the point of predicting something abstract and unknown?\nWhat they say about target:\n\n - 'Loyalty numerical score calculated 2 months after historical and evaluation period'\n - 'You are predicting a loyalty score for each card_id represented in test.csv and sample_submission.csv.'\n\nSo, the company already has some formula to calculate abstract score representing loyalty.\nMaybe, it's based on spendings, monthly spendings, sliding average or smth similar.\nLet's represent this equation: (x+y)/z = t (where t- target, and xyz - some spendings, active days, amount of purchases, etc)\n\nMore naturally we'd predict X, Y and Z rather than T. \n\nOtherwise, it's rather trying to guess already known formula than a predictive modeling\n\nYour thoughts?\n", "432948": "We are predicting a loyalty score. But loyal to what? To ELO, to the merchants, to the products, to the city?", "435379": "completely agreed. my personal experience with the (x+y)/z = t situation is to predict x, y, z rather than t.\nthat way we can minimize the prediction standard error of t_hat..", "432942": "I agree with you. It is puzzling. One reason of the competicion could be to find target values before evaluating period. \n\nImagine it is necessary this period with its transactions to calculate a past loyalty. To predict loyalty would make sense. Are you agree? But nothing is clear. ", "432918": "" } }