Hello,everyone!\nI have found that almost every card_id have an established buying habits.That is, basically all transactions for a card_id are A or non-A(category_3's value = 'A' means installments =0 and category_3's value ='B' or 'C' means installments >0).It seems that a cardholder either basically purchase in installments or basically no installments.I don't know how to insert images,but I think one of you may have found out.I try to divide the train sets into two parts and predict them separately.But it dosen't help to improve LB score. \nDo you have further discoveries or better solutions?\nMy English is not very good, I hope you can understand.Thanks!
", "messages": [ { "id": "443865", "postDate": "12/22/2018 16:09:21", "content": "Hello,everyone!\nI have found that almost every card_id have an established buying habits.That is, basically all transactions for a card_id are A or non-A(category_3's value = 'A' means installments =0 and category_3's value ='B' or 'C' means installments >0).It seems that a cardholder either basically purchase in installments or basically no installments.I don't know how to insert images,but I think one of you may have found out.I try to divide the train sets into two parts and predict them separately.But it dosen't help to improve LB score. \nDo you have further discoveries or better solutions?\nMy English is not very good, I hope you can understand.Thanks!
", "rawMarkdown": "Hello,everyone!\nI have found that almost every card_id have an established buying habits.That is, basically all transactions for a card_id are A or non-A(category_3's value = 'A' means installments =0 and category_3's value ='B' or 'C' means installments >0).It seems that a cardholder either basically purchase in installments or basically no installments.I don't know how to insert images,but I think one of you may have found out.I try to divide the train sets into two parts and predict them separately.But it dosen't help to improve LB score. \nDo you have further discoveries or better solutions?\nMy English is not very good, I hope you can understand.Thanks!", "votes": null }, { "id": "444086", "postDate": "12/23/2018 06:01:54", "content": "Thank you for sharing your observations and insights.In category 'A' it is 0 or -1, but if check category 'B' it is only 1, or -1. and for 'C' it is more than one or -1. What does -1, or 999 mean in 'instalments'?
", "rawMarkdown": "Thank you for sharing your observations and insights.In category 'A' it is 0 or -1, but if check category 'B' it is only 1, or -1. and for 'C' it is more than one or -1. What does -1, or 999 mean in 'instalments'?", "votes": null }, { "id": "444104", "postDate": "12/23/2018 07:18:49", "content": "I did this as you say, got a very worse score. but if we divide the data into different part, we can find the purchase amount's mean is different hugely.
", "rawMarkdown": "I did this as you say, got a very worse score. but if we divide the data into different part, we can find the purchase amount's mean is different hugely.", "votes": null }, { "id": "444171", "postDate": "12/23/2018 12:13:12", "content": "Hi,Salamat!\nYou are right! I continue to analyze what you said and find the card_id which had -1 installment record has better loyalty scores(46301 cards with -0.236824 on average) than the average(-0.393636).It seems that the anomalous record has no negative effect on the loyalty score,even they get better loyalty score instead.In addition,the card_id which had 999 installments record has less loyalty scores(104 cards with -0.84444 on average).And Only 6 operations are allowed(authorized_flag = Y).More interestingly,the 6 allowed operations have some similarities(merchant_category_id is 840 and state_id is 9).But these six records are generated in four merchants , the merchant whose merchant_category_id is 840 is less than 1 percent of all merchants.I don't think it's a coincidence.Personally,I think the purchase way may be associated with merchant information.I will next analyse merchants.csv and try to find if there is an intrinsic link between them and find some really unusual transactions.Thanks!
", "rawMarkdown": "Hi,Salamat!\nYou are right! I continue to analyze what you said and find the card_id which had -1 installment record has better loyalty scores(46301 cards with -0.236824 on average) than the average(-0.393636).It seems that the anomalous record has no negative effect on the loyalty score,even they get better loyalty score instead.In addition,the card_id which had 999 installments record has less loyalty scores(104 cards with -0.84444 on average).And Only 6 operations are allowed(authorized_flag = Y).More interestingly,the 6 allowed operations have some similarities(merchant_category_id is 840 and state_id is 9).But these six records are generated in four merchants , the merchant whose merchant_category_id is 840 is less than 1 percent of all merchants.I don't think it's a coincidence.Personally,I think the purchase way may be associated with merchant information.I will next analyse merchants.csv and try to find if there is an intrinsic link between them and find some really unusual transactions.Thanks!", "votes": null }, { "id": "444176", "postDate": "12/23/2018 12:32:41", "content": "Yeah,not only the purchase amount'mean is different hugely,but also other features are also very different under different category_3 values such as merchant_category_id.I will continue to find more regularities.Good luck!
", "rawMarkdown": "Yeah,not only the purchase amount'mean is different hugely,but also other features are also very different under different category_3 values such as merchant_category_id.I will continue to find more regularities.Good luck!", "votes": null }, { "id": "444493", "postDate": "12/24/2018 05:53:55", "content": "so what kind of aggregation gives u the best boost?\nCategory-3 seems to be a better choice considering your observation.
", "rawMarkdown": "so what kind of aggregation gives u the best boost?\nCategory-3 seems to be a better choice considering your observation.", "votes": null }, { "id": "465745", "postDate": "02/03/2019 22:26:38", "content": "@火石桥电鳗\nDo you know the difference between -1 and 0?\nFrom your comment, I am guessing -1 means that the customers did not opt for installment. But in fact, their paying pattern is the same as the ones with 0, correct?
", "rawMarkdown": "火石桥电鳗\nDo you know the difference between -1 and 0?\nFrom your comment, I am guessing -1 means that the customers did not opt for installment. But in fact, their paying pattern is the same as the ones with 0, correct?", "votes": null } ], "comments": [ { "id": 444086, "author_name": "salamatburj", "author_url": "", "post_date": "12/23/2018 06:01:54", "content": "Thank you for sharing your observations and insights.In category 'A' it is 0 or -1, but if check category 'B' it is only 1, or -1. and for 'C' it is more than one or -1. What does -1, or 999 mean in 'instalments'?
", "votes": null, "replies": [ { "id": 444171, "author_name": "wxxdepp", "author_url": "", "post_date": "12/23/2018 12:13:12", "content": "Hi,Salamat!\nYou are right! I continue to analyze what you said and find the card_id which had -1 installment record has better loyalty scores(46301 cards with -0.236824 on average) than the average(-0.393636).It seems that the anomalous record has no negative effect on the loyalty score,even they get better loyalty score instead.In addition,the card_id which had 999 installments record has less loyalty scores(104 cards with -0.84444 on average).And Only 6 operations are allowed(authorized_flag = Y).More interestingly,the 6 allowed operations have some similarities(merchant_category_id is 840 and state_id is 9).But these six records are generated in four merchants , the merchant whose merchant_category_id is 840 is less than 1 percent of all merchants.I don't think it's a coincidence.Personally,I think the purchase way may be associated with merchant information.I will next analyse merchants.csv and try to find if there is an intrinsic link between them and find some really unusual transactions.Thanks!
", "votes": null, "replies": [] }, { "id": 444493, "author_name": "subikashpal", "author_url": "", "post_date": "12/24/2018 05:53:55", "content": "so what kind of aggregation gives u the best boost?\nCategory-3 seems to be a better choice considering your observation.
", "votes": null, "replies": [] } ] }, { "id": 444104, "author_name": "codlife", "author_url": "", "post_date": "12/23/2018 07:18:49", "content": "I did this as you say, got a very worse score. but if we divide the data into different part, we can find the purchase amount's mean is different hugely.
", "votes": null, "replies": [ { "id": 444176, "author_name": "wxxdepp", "author_url": "", "post_date": "12/23/2018 12:32:41", "content": "Yeah,not only the purchase amount'mean is different hugely,but also other features are also very different under different category_3 values such as merchant_category_id.I will continue to find more regularities.Good luck!
", "votes": null, "replies": [] } ] }, { "id": 465745, "author_name": "liwenhuang", "author_url": "", "post_date": "02/03/2019 22:26:38", "content": "@火石桥电鳗\nDo you know the difference between -1 and 0?\nFrom your comment, I am guessing -1 means that the customers did not opt for installment. But in fact, their paying pattern is the same as the ones with 0, correct?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "443865": "Hello,everyone!\nI have found that almost every card_id have an established buying habits.That is, basically all transactions for a card_id are A or non-A(category_3's value = 'A' means installments =0 and category_3's value ='B' or 'C' means installments >0).It seems that a cardholder either basically purchase in installments or basically no installments.I don't know how to insert images,but I think one of you may have found out.I try to divide the train sets into two parts and predict them separately.But it dosen't help to improve LB score. \nDo you have further discoveries or better solutions?\nMy English is not very good, I hope you can understand.Thanks!", "444086": "Thank you for sharing your observations and insights.In category 'A' it is 0 or -1, but if check category 'B' it is only 1, or -1. and for 'C' it is more than one or -1. What does -1, or 999 mean in 'instalments'?", "444104": "I did this as you say, got a very worse score. but if we divide the data into different part, we can find the purchase amount's mean is different hugely.", "444171": "Hi,Salamat!\nYou are right! I continue to analyze what you said and find the card_id which had -1 installment record has better loyalty scores(46301 cards with -0.236824 on average) than the average(-0.393636).It seems that the anomalous record has no negative effect on the loyalty score,even they get better loyalty score instead.In addition,the card_id which had 999 installments record has less loyalty scores(104 cards with -0.84444 on average).And Only 6 operations are allowed(authorized_flag = Y).More interestingly,the 6 allowed operations have some similarities(merchant_category_id is 840 and state_id is 9).But these six records are generated in four merchants , the merchant whose merchant_category_id is 840 is less than 1 percent of all merchants.I don't think it's a coincidence.Personally,I think the purchase way may be associated with merchant information.I will next analyse merchants.csv and try to find if there is an intrinsic link between them and find some really unusual transactions.Thanks!", "444176": "Yeah,not only the purchase amount'mean is different hugely,but also other features are also very different under different category_3 values such as merchant_category_id.I will continue to find more regularities.Good luck!", "444493": "so what kind of aggregation gives u the best boost?\nCategory-3 seems to be a better choice considering your observation.", "465745": "火石桥电鳗\nDo you know the difference between -1 and 0?\nFrom your comment, I am guessing -1 means that the customers did not opt for installment. But in fact, their paying pattern is the same as the ones with 0, correct?" }, "source": "meta" }