![\"\"](\"https://i.imgur.com/Q9Cyp6H.png\")
I was wondering if someone has managed to find the listing price of the products. My first guess would be to take the mode of the various prices, but this would pose a problem when we have products with different price ranges. For clarity, a \"price range\" is, in my opinion, a price associated to a lot of transactions, and which is usually \"fixed\".
\nTake, for example, the price distribution of this product (sorry for the x axis but I was not able to show only some dates):
\n
We can see there are different price ranges. I have various hypotesis:
\n![\"\"](\"https://i.imgur.com/1yvJL3p.png\")
We see that we have 3 transactions which are the maximum value possible: according to what I have said before, this may be the effective price of the product. In this case, recovering the original price would be easily done through a max function over all the transactions.
\nAnyway, whichever is the hypotesis, there are still problem related to product with a low number of transaction, in which we cannot see a frequent selling price. Also, it is possible that a product was never bought at the original price!
\nMy question is: how do you deal with this situation? Having the listing price would be a very precious information for some customer analysis. For now, I've decided to tackle the problem by \"approximating\" the real price with a mean operation, but I know this maybe a way too biased estimator of the real price.
", "messages": [ { "id": "1713059", "postDate": "03/05/2022 16:22:01", "content": "I was wondering if someone has managed to find the listing price of the products. My first guess would be to take the mode of the various prices, but this would pose a problem when we have products with different price ranges. For clarity, a \"price range\" is, in my opinion, a price associated to a lot of transactions, and which is usually \"fixed\".
\nTake, for example, the price distribution of this product (sorry for the x axis but I was not able to show only some dates):
\n
We can see there are different price ranges. I have various hypotesis:
\n
We see that we have 3 transactions which are the maximum value possible: according to what I have said before, this may be the effective price of the product. In this case, recovering the original price would be easily done through a max function over all the transactions.
\nAnyway, whichever is the hypotesis, there are still problem related to product with a low number of transaction, in which we cannot see a frequent selling price. Also, it is possible that a product was never bought at the original price!
\nMy question is: how do you deal with this situation? Having the listing price would be a very precious information for some customer analysis. For now, I've decided to tackle the problem by \"approximating\" the real price with a mean operation, but I know this maybe a way too biased estimator of the real price.
", "rawMarkdown": "I was wondering if someone has managed to find the listing price of the products. My first guess would be to take the mode of the various prices, but this would pose a problem when we have products with different price ranges. For clarity, a \"price range\" is, in my opinion, a price associated to a lot of transactions, and which is usually \"fixed\". \n\nTake, for example, the price distribution of this product (sorry for the x axis but I was not able to show only some dates):\n\n\n\nWe can see there are different price ranges. I have various hypotesis:\n1. Different price ranges are associated to different geographical regions. For example, even if the dataset is referred only to USA, my guess is that we may have different price ranges for each country. If this is the case, there is (to my understanding) no possibility to retrieve the effective price of the product, before any discount is applied.\n2. Different price ranges are only due to discounts applied to product in different regions, but the original listing price is unique for each H&M product. This is a plausible explanation of the fact that the majority of products have a price range which is rather frequent and there are no outliers that are greater than this price (and if there are it would be possible that the product is most of the time on sale). For example, considering another product price distribution:\n\n\n\nWe see that we have 3 transactions which are the maximum value possible: according to what I have said before, this may be the effective price of the product. In this case, recovering the original price would be easily done through a max function over all the transactions.\n\nAnyway, whichever is the hypotesis, there are still problem related to product with a low number of transaction, in which we cannot see a frequent selling price. Also, it is possible that a product was never bought at the original price!\n\nMy question is: how do you deal with this situation? Having the listing price would be a very precious information for some customer analysis. For now, I've decided to tackle the problem by \"approximating\" the real price with a mean operation, but I know this maybe a way too biased estimator of the real price.", "votes": null }, { "id": "1713093", "postDate": "03/05/2022 17:04:45", "content": "See here. The price is the dataframe appears to be the true price divided by 590
", "rawMarkdown": "See [here][1]. The price is the dataframe appears to be the true price divided by 590\n\n[1]: https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/310496", "votes": null }, { "id": "1713610", "postDate": "03/06/2022 07:50:35", "content": "Thanks for sharing.
\nThe amount of postal codes (352899) should invalidate your hypotesis #1.
Sadly, even multiplying for 590 does not allow me to obtain the listing price. The existence of multiple frequent price range still represents a major problem in assessing the real price. Maybe it would be useful to consider multiple price ranges as the true listing prices for different regions?
", "rawMarkdown": "Sadly, even multiplying for 590 does not allow me to obtain the listing price. The existence of multiple frequent price range still represents a major problem in assessing the real price. Maybe it would be useful to consider multiple price ranges as the true listing prices for different regions?", "votes": null }, { "id": "1713620", "postDate": "03/06/2022 08:00:42", "content": "I don't think so. It would be interesting to cluster the postal codes based on the price ranges identified for the products: this would give us an insight on how near are the cities related to the postal codes (for example, same price means city in the same region?)
", "rawMarkdown": "I don't think so. It would be interesting to cluster the postal codes based on the price ranges identified for the products: this would give us an insight on how near are the cities related to the postal codes (for example, same price means city in the same region?)", "votes": null } ], "comments": [ { "id": 1713093, "author_name": "cdeotte", "author_url": "", "post_date": "03/05/2022 17:04:45", "content": "See here. The price is the dataframe appears to be the true price divided by 590
", "votes": null, "replies": [ { "id": 1713618, "author_name": "kingpowa", "author_url": "", "post_date": "03/06/2022 07:58:20", "content": "Sadly, even multiplying for 590 does not allow me to obtain the listing price. The existence of multiple frequent price range still represents a major problem in assessing the real price. Maybe it would be useful to consider multiple price ranges as the true listing prices for different regions?
", "votes": null, "replies": [] } ] }, { "id": 1713610, "author_name": "a45632", "author_url": "", "post_date": "03/06/2022 07:50:35", "content": "Thanks for sharing.
\nThe amount of postal codes (352899) should invalidate your hypotesis #1.
I don't think so. It would be interesting to cluster the postal codes based on the price ranges identified for the products: this would give us an insight on how near are the cities related to the postal codes (for example, same price means city in the same region?)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1713059": "I was wondering if someone has managed to find the listing price of the products. My first guess would be to take the mode of the various prices, but this would pose a problem when we have products with different price ranges. For clarity, a \"price range\" is, in my opinion, a price associated to a lot of transactions, and which is usually \"fixed\". \n\nTake, for example, the price distribution of this product (sorry for the x axis but I was not able to show only some dates):\n\n\n\nWe can see there are different price ranges. I have various hypotesis:\n1. Different price ranges are associated to different geographical regions. For example, even if the dataset is referred only to USA, my guess is that we may have different price ranges for each country. If this is the case, there is (to my understanding) no possibility to retrieve the effective price of the product, before any discount is applied.\n2. Different price ranges are only due to discounts applied to product in different regions, but the original listing price is unique for each H&M product. This is a plausible explanation of the fact that the majority of products have a price range which is rather frequent and there are no outliers that are greater than this price (and if there are it would be possible that the product is most of the time on sale). For example, considering another product price distribution:\n\n\n\nWe see that we have 3 transactions which are the maximum value possible: according to what I have said before, this may be the effective price of the product. In this case, recovering the original price would be easily done through a max function over all the transactions.\n\nAnyway, whichever is the hypotesis, there are still problem related to product with a low number of transaction, in which we cannot see a frequent selling price. Also, it is possible that a product was never bought at the original price!\n\nMy question is: how do you deal with this situation? Having the listing price would be a very precious information for some customer analysis. For now, I've decided to tackle the problem by \"approximating\" the real price with a mean operation, but I know this maybe a way too biased estimator of the real price.", "1713093": "See [here][1]. The price is the dataframe appears to be the true price divided by 590\n\n[1]: https://www.kaggle.com/c/h-and-m-personalized-fashion-recommendations/discussion/310496", "1713610": "Thanks for sharing.\nThe amount of postal codes (352899) should invalidate your hypotesis #1.", "1713618": "Sadly, even multiplying for 590 does not allow me to obtain the listing price. The existence of multiple frequent price range still represents a major problem in assessing the real price. Maybe it would be useful to consider multiple price ranges as the true listing prices for different regions?", "1713620": "I don't think so. It would be interesting to cluster the postal codes based on the price ranges identified for the products: this would give us an insight on how near are the cities related to the postal codes (for example, same price means city in the same region?)" }, "source": "meta" }