While checking the discussion, I found an important comment from Competition Host and Kaggle Staff.
\nHere is a summary of them.
Q.
\n1% on public LB - is this a record?
\nA.
\n\nThe 1% is an artifact of ignoring the rows that don't have sales in the test time period. You can safely assume the the actual Public leaderboard is calculated from 5-15% of the scored rows.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/305986#1680404
\n
Q.
\n[Resolved] Question regarding evaluation metric
\nA.
\n\nYou can also walk through this code:
\n
\n https://github.com/benhamner/Metrics/blob/master/Python/ml_metrics/average_precision.py
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/306007#1680513
\n
Q.
\nWhat does \"7-day period immediately after the training data ends\" exactly mean?
\nA.
\n\n7 day period after the latest date found in the training data. The test week is the same for all customers, not one individual week per customer based on their latest training sample.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/306380#1682561
\n
Q.
\nData Questions - Missing Transactions
A.
\n\n\nHi!
\n
\n 1 - The unit of price isn't any \"currency/unit\" as we chose to not disclose the real values.
\n 2 - Yes, that is correct.
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/306016#1680549
\nQ.
\n I have a question I hope I didn't miss it anywhere. Should we consider predicting items that a user has NOT purchased before? Or both items purchased and not purchased? I feel typically the idea is to show the user new content but they can certainly re-purchase the same item 🤔
A.
\n\n\nFor this competition we don't require that it should be new content that you provide as recommendations. Therefore you may recommend items that the customer already has bought.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/305952#1684468
\nQ.
\nQuestion about the variables:
\nFN, Active - What are these?
\nSales channel ID - is is right to assume 2 is in person and 1 is online?
\nPostal code - 1.2 mil is a lot of postal codes (approx every country you sell in) is there some else in these hashes?
\nwhat percent of the customers are new in the validation set (any idea helps)?
\nThanks
\nA.
\n\nFN is if a customer get Fashion News newsletter, Active is if the customer is active for communication, sales channel id, 2 is online and 1 store.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/305952#1684481
\nQ.
\nI have a question to the data:
\nIn the data we have a table of transactions. Those are in principle sold items. However, this article from 2019 shows that that up to 30% - 40% of cloths and shoes bought online are being returned.
Is the transactions table already cleaned from the returns?
\nA.
\n\n\nThe transaction table holds all transactions that happened whether returned later or not.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/305952#1699535
\nQ.
\n i have one question. As i checked, H&M is a SWEDEN based company. So this data come from SWEDEN customer and cities(for offline transactions) ?
\nIf it is correct, can i use sweden external data (CPI, economy, …) for this competition ?
A.
\n\n\nH&M Group exist worldwide. We don't disclosure what market this data comes from due to privacy reasons.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/307001#1699540
\n
\nI will regularly update it.
While checking the discussion, I found an important comment from Competition Host and Kaggle Staff.
\nHere is a summary of them.
Q.
\n1% on public LB - is this a record?
\nA.
\n\nThe 1% is an artifact of ignoring the rows that don't have sales in the test time period. You can safely assume the the actual Public leaderboard is calculated from 5-15% of the scored rows.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/305986#1680404
\n
Q.
\n[Resolved] Question regarding evaluation metric
\nA.
\n\nYou can also walk through this code:
\n
\n https://github.com/benhamner/Metrics/blob/master/Python/ml_metrics/average_precision.py
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/306007#1680513
\n
Q.
\nWhat does \"7-day period immediately after the training data ends\" exactly mean?
\nA.
\n\n7 day period after the latest date found in the training data. The test week is the same for all customers, not one individual week per customer based on their latest training sample.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/306380#1682561
\n
Q.
\nData Questions - Missing Transactions
A.
\n\n\nHi!
\n
\n 1 - The unit of price isn't any \"currency/unit\" as we chose to not disclose the real values.
\n 2 - Yes, that is correct.
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/306016#1680549
\nQ.
\n I have a question I hope I didn't miss it anywhere. Should we consider predicting items that a user has NOT purchased before? Or both items purchased and not purchased? I feel typically the idea is to show the user new content but they can certainly re-purchase the same item 🤔
A.
\n\n\nFor this competition we don't require that it should be new content that you provide as recommendations. Therefore you may recommend items that the customer already has bought.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/305952#1684468
\nQ.
\nQuestion about the variables:
\nFN, Active - What are these?
\nSales channel ID - is is right to assume 2 is in person and 1 is online?
\nPostal code - 1.2 mil is a lot of postal codes (approx every country you sell in) is there some else in these hashes?
\nwhat percent of the customers are new in the validation set (any idea helps)?
\nThanks
\nA.
\n\nFN is if a customer get Fashion News newsletter, Active is if the customer is active for communication, sales channel id, 2 is online and 1 store.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/305952#1684481
\nQ.
\nI have a question to the data:
\nIn the data we have a table of transactions. Those are in principle sold items. However, this article from 2019 shows that that up to 30% - 40% of cloths and shoes bought online are being returned.
Is the transactions table already cleaned from the returns?
\nA.
\n\n\nThe transaction table holds all transactions that happened whether returned later or not.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/305952#1699535
\nQ.
\n i have one question. As i checked, H&M is a SWEDEN based company. So this data come from SWEDEN customer and cities(for offline transactions) ?
\nIf it is correct, can i use sweden external data (CPI, economy, …) for this competition ?
A.
\n\n\nH&M Group exist worldwide. We don't disclosure what market this data comes from due to privacy reasons.
\n
Link : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/307001#1699540
\n
\nI will regularly update it.
I was thinking about this question which was still open:
\n\n\nOnline transactions for the month of April 2020 is missing. What is a possible explanation for this? This is important because this can have an impact on the customer behaviour analysis. See notebook for the performed analysis.
\n
And I saw also this in the last answer currently in this post:
\n\n\nsales channel id, 2 is online and 1 store
\n
I created a micro notebook https://www.kaggle.com/pietromaldini1/fast-experiment in which i checked your first statement. And we can see that it is the opposite.
\nThe offline transactions are missing for that month.
\n(I'm sorry that the notebook is really basic and bad looking)
Once i discovered this i would give you a partial answer to why the data may be missing.
\nMy current possible explanations:
\n1) In some states around the world for Covid there was partial or complete lockdown, the data could come from places where people could not physically go to stores, so no offline transactions(or not enough to preserve privacy).
2) It gave out too much information which created risk of privacy violations, always connected to the lockdowns. If some frequently buying users stops buying in that period you could find out their geographical location, omitting that part of data completely reduces this risk ( there is still a risk if the reduction in offline shopping is counterbalanced by a sudden increase in online shopping by that user , but without the offline buying behaviour you can't know if it is due to lockdown or just to a sudden change in buying behaviour).
", "rawMarkdown": "I was thinking about this question which was still open: \n> Online transactions for the month of April 2020 is missing. What is a possible explanation for this? This is important because this can have an impact on the customer behaviour analysis. See notebook for the performed analysis.\n\nAnd I saw also this in the last answer currently in this post:\n> sales channel id, 2 is online and 1 store\n\nI created a micro notebook https://www.kaggle.com/pietromaldini1/fast-experiment in which i checked your first statement. And we can see that it is the opposite. \nThe offline transactions are missing for that month.\n(I'm sorry that the notebook is really basic and bad looking)\n\nOnce i discovered this i would give you a partial answer to why the data may be missing.\n\nMy current possible explanations:\n1) In some states around the world for Covid there was partial or complete lockdown, the data could come from places where people could not physically go to stores, so no offline transactions(or not enough to preserve privacy).\n\n\n 2) It gave out too much information which created risk of privacy violations, always connected to the lockdowns. If some frequently buying users stops buying in that period you could find out their geographical location, omitting that part of data completely reduces this risk ( there is still a risk if the reduction in offline shopping is counterbalanced by a sudden increase in online shopping by that user , but without the offline buying behaviour you can't know if it is due to lockdown or just to a sudden change in buying behaviour).\n\n \n", "votes": 24, "replies": [ { "id": 1688128, "postDate": "2022-02-13T11:59:40.077Z", "content": "I agree with your explanation. I found a news article about H&M in June, 2020.
\n\n", "rawMarkdown": "I agree with your explanation. I found [a news article](https://www.theretailbulletin.com/fashion/hm-sales-halve-during-lockdown-16-06-2020/) about H&M in June, 2020.\n\n> By mid-April, around 90% of the retailer’s stores were shuttered.\n> Meanwhile, online sales increased by 36% during the period.\n> Although stores began reopening in a number of markets from the end of April, H&M said recovery rates within the various markets have been uneven. Some 900 of the retailer’s 5,058 stores are still temporarily closed but ecommerce operations are open in 48 of its 51 online markets.", "votes": 8 }, { "id": 1690503, "postDate": "2022-02-15T01:09:57.980Z", "content": "By mid-April, around 90% of the retailer’s stores were shuttered.
\n
\n Meanwhile, online sales increased by 36% during the period.
\n Although stores began reopening in a number of markets from the end of April, H&M said recovery rates within the various markets have been uneven. Some 900 of the retailer’s 5,058 stores are still temporarily closed but ecommerce operations are open in 48 of its 51 online markets.
Thanks for the comment and sharing the Notebook.
\nI agree with your opinion.
\nIt may be better to treat the April 2020 data as an outlier.
Thank you for sharing this post. It is very convenient to have all the informations on the same place.
", "rawMarkdown": "Thank you for sharing this post. It is very convenient to have all the informations on the same place. ", "votes": 1 }, { "id": 1688202, "postDate": "2022-02-13T13:23:34.267Z", "content": "@zakopur0 great summary and thanks for sharing!
", "rawMarkdown": "@zakopur0 great summary and thanks for sharing!", "votes": 1 }, { "id": 1773493, "postDate": "2022-05-01T08:32:20.870Z", "content": "@zakopur0 Thanks for sharing!
\n\n\nFor this competition we don't require that it should be new content that you provide as recommendations. Therefore you may recommend items that the customer already has bought.
\n
Does this mean that we should delete articles purchased in the next week that are not present in the customer's history?
\nBecause these articles are positive samples for the customer.
@paweljankiewicz @zakopur0 is this right?
", "rawMarkdown": "@zakopur0 Thanks for sharing!\n\n> For this competition we don't require that it should be new content that you provide as recommendations. Therefore you may recommend items that the customer already has bought.\n\nDoes this mean that we should delete articles purchased in the next week that are not present in the customer's history?\nBecause these articles are positive samples for the customer.\n\n@paweljankiewicz @zakopur0 is this right?" }, { "id": 1713205, "postDate": "2022-03-05T18:45:48.137Z", "content": "how we can distinguish between a pruchase or a return?
", "rawMarkdown": "how we can distinguish between a pruchase or a return? " }, { "id": 1699249, "postDate": "2022-02-21T03:50:20.997Z", "content": "@zakopur0 i have one question. As i checked, H&M is a SWEDEN based company. So this data come from SWEDEN customer and cities(for offline transactions) ?
\nIf it is correct, can i use sweden external data (CPI, economy, …) for this competition ?
H&M Group exist worldwide. We don't disclosure what market this data comes from due to privacy reasons.
", "rawMarkdown": "H&M Group exist worldwide. We don't disclosure what market this data comes from due to privacy reasons. ", "votes": 4 } ] }, { "id": 1729855, "postDate": "2022-03-20T15:53:32.640Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1687971, "postDate": "2022-02-13T09:21:35.990Z", "content": "Thank you!!
", "rawMarkdown": "Thank you!!", "votes": 1 }, { "id": 1687639, "postDate": "2022-02-13T03:03:27.523Z", "content": "super helpful!
\nThank you!
Excellent thread @zakopur0 . Thanks
", "rawMarkdown": "Excellent thread @zakopur0 . Thanks", "votes": 1 }, { "id": 1734058, "postDate": "2022-03-24T23:35:59.653Z", "content": "thanks a lot
", "rawMarkdown": "thanks a lot" }, { "id": 1731049, "postDate": "2022-03-22T00:25:31.823Z", "content": "thanks a lot, its really helpful!
", "rawMarkdown": "thanks a lot, its really helpful!" }, { "id": 1722216, "postDate": "2022-03-14T09:52:24.297Z", "content": "Great thanks!
", "rawMarkdown": "Great thanks!" }, { "id": 1711813, "postDate": "2022-03-04T10:29:07.857Z", "content": "Thank You!
", "rawMarkdown": "Thank You!" }, { "id": 1699768, "postDate": "2022-02-21T12:31:31.557Z", "content": "Thank you for teaching me!
", "rawMarkdown": "Thank you for teaching me!" } ], "comments": [ { "id": 1687052, "author_name": "Pietro Maldini", "author_url": "", "post_date": "2022-02-12T15:33:31.087000", "content": "I was thinking about this question which was still open:
\n\n\nOnline transactions for the month of April 2020 is missing. What is a possible explanation for this? This is important because this can have an impact on the customer behaviour analysis. See notebook for the performed analysis.
\n
And I saw also this in the last answer currently in this post:
\n\n\nsales channel id, 2 is online and 1 store
\n
I created a micro notebook https://www.kaggle.com/pietromaldini1/fast-experiment in which i checked your first statement. And we can see that it is the opposite.
\nThe offline transactions are missing for that month.
\n(I'm sorry that the notebook is really basic and bad looking)
Once i discovered this i would give you a partial answer to why the data may be missing.
\nMy current possible explanations:
\n1) In some states around the world for Covid there was partial or complete lockdown, the data could come from places where people could not physically go to stores, so no offline transactions(or not enough to preserve privacy).
2) It gave out too much information which created risk of privacy violations, always connected to the lockdowns. If some frequently buying users stops buying in that period you could find out their geographical location, omitting that part of data completely reduces this risk ( there is still a risk if the reduction in offline shopping is counterbalanced by a sudden increase in online shopping by that user , but without the offline buying behaviour you can't know if it is due to lockdown or just to a sudden change in buying behaviour).
", "votes": 24, "replies": [ { "id": 1688128, "author_name": "tomoo inubushi", "author_url": "", "post_date": "2022-02-13T11:59:40.077000", "content": "I agree with your explanation. I found a news article about H&M in June, 2020.
\n\n", "votes": 8, "replies": [] }, { "id": 1690503, "author_name": "zakopuro", "author_url": "", "post_date": "2022-02-15T01:09:57.980000", "content": "By mid-April, around 90% of the retailer’s stores were shuttered.
\n
\n Meanwhile, online sales increased by 36% during the period.
\n Although stores began reopening in a number of markets from the end of April, H&M said recovery rates within the various markets have been uneven. Some 900 of the retailer’s 5,058 stores are still temporarily closed but ecommerce operations are open in 48 of its 51 online markets.
Thanks for the comment and sharing the Notebook.
\nI agree with your opinion.
\nIt may be better to treat the April 2020 data as an outlier.
Thank you for sharing this post. It is very convenient to have all the informations on the same place.
", "votes": 1, "replies": [] }, { "id": 1688202, "author_name": "Jie Wu", "author_url": "", "post_date": "2022-02-13T13:23:34.267000", "content": "@zakopur0 great summary and thanks for sharing!
", "votes": 1, "replies": [] }, { "id": 1773493, "author_name": "Amed", "author_url": "", "post_date": "2022-05-01T08:32:20.870000", "content": "@zakopur0 Thanks for sharing!
\n\n\nFor this competition we don't require that it should be new content that you provide as recommendations. Therefore you may recommend items that the customer already has bought.
\n
Does this mean that we should delete articles purchased in the next week that are not present in the customer's history?
\nBecause these articles are positive samples for the customer.
@paweljankiewicz @zakopur0 is this right?
", "votes": 0, "replies": [] }, { "id": 1713205, "author_name": "EnricRovira", "author_url": "", "post_date": "2022-03-05T18:45:48.137000", "content": "how we can distinguish between a pruchase or a return?
", "votes": 0, "replies": [] }, { "id": 1699249, "author_name": "Nguyentuananh", "author_url": "", "post_date": "2022-02-21T03:50:20.997000", "content": "@zakopur0 i have one question. As i checked, H&M is a SWEDEN based company. So this data come from SWEDEN customer and cities(for offline transactions) ?
\nIf it is correct, can i use sweden external data (CPI, economy, …) for this competition ?
H&M Group exist worldwide. We don't disclosure what market this data comes from due to privacy reasons.
", "votes": 4, "replies": [] } ] }, { "id": 1729855, "author_name": "", "author_url": "", "post_date": "2022-03-20T15:53:32.640000", "content": "", "votes": 0, "replies": [] }, { "id": 1687971, "author_name": "morixx", "author_url": "", "post_date": "2022-02-13T09:21:35.990000", "content": "Thank you!!
", "votes": 1, "replies": [] }, { "id": 1687639, "author_name": "Clear n' Simple", "author_url": "", "post_date": "2022-02-13T03:03:27.523000", "content": "super helpful!
\nThank you!
Excellent thread @zakopur0 . Thanks
", "votes": 1, "replies": [] }, { "id": 1734058, "author_name": "stallone", "author_url": "", "post_date": "2022-03-24T23:35:59.653000", "content": "thanks a lot
", "votes": 0, "replies": [] }, { "id": 1731049, "author_name": "newbieclass", "author_url": "", "post_date": "2022-03-22T00:25:31.823000", "content": "thanks a lot, its really helpful!
", "votes": 0, "replies": [] }, { "id": 1722216, "author_name": "Darron Kwon", "author_url": "", "post_date": "2022-03-14T09:52:24.297000", "content": "Great thanks!
", "votes": 0, "replies": [] }, { "id": 1711813, "author_name": "allen0402", "author_url": "", "post_date": "2022-03-04T10:29:07.857000", "content": "Thank You!
", "votes": 0, "replies": [] }, { "id": 1699768, "author_name": "tonsuke", "author_url": "", "post_date": "2022-02-21T12:31:31.557000", "content": "Thank you for teaching me!
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1686313": "While checking the discussion, I found an important comment from Competition Host and Kaggle Staff.\nHere is a summary of them.\n\n**Q.**\n1% on public LB - is this a record?\n**A.**\n> The 1% is an artifact of ignoring the rows that don't have sales in the test time period. You can safely assume the the actual Public leaderboard is calculated from 5-15% of the scored rows. \n \nLink : https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/305986#1680404 \n