Out of curiosity, what's the best score that anybody has achieved with unpersonalized recommendations (i.e. exactly the same prediction for every customer)? My first such attempt scored 0.0074.
\n(I appreciate that this competition is all about the personalization, but I was wondering how far you could get without it.)
", "messages": [ { "id": "1731757", "postDate": "03/22/2022 16:52:15", "content": "Out of curiosity, what's the best score that anybody has achieved with unpersonalized recommendations (i.e. exactly the same prediction for every customer)? My first such attempt scored 0.0074.
\n(I appreciate that this competition is all about the personalization, but I was wondering how far you could get without it.)
", "rawMarkdown": "Out of curiosity, what's the best score that anybody has achieved with **un**personalized recommendations (i.e. exactly the same prediction for every customer)? My first such attempt scored 0.0074.\n\n(I appreciate that this competition is all about the personalization, but I was wondering how far you could get without it.)", "votes": null }, { "id": "1732008", "postDate": "03/22/2022 23:25:23", "content": "It is an interesting idea. Not sure how far and where it can be taken, given the constraints and lure of other options.
", "rawMarkdown": "It is an interesting idea. Not sure how far and where it can be taken, given the constraints and lure of other options.", "votes": null }, { "id": "1732357", "postDate": "03/23/2022 09:53:11", "content": "I appreciate it won't be useful for winning the the competition - but I suppose it might be useful in generating predictions for customers with no purchase history. (Indeed - starting with the approach used by all the people who've scored .0225 and then substituting my best general prediction so far for the customers with no history, gives a boost to .0227 despite their only being ~10k such customers.)
", "rawMarkdown": "I appreciate it won't be useful for winning the the competition - but I suppose it might be useful in generating predictions for customers with no purchase history. (Indeed - starting with the approach used by all the people who've scored .0225 and then substituting my best general prediction so far for the customers with no history, gives a boost to .0227 despite their only being ~10k such customers.)", "votes": null }, { "id": "1733959", "postDate": "03/24/2022 20:09:34", "content": "See what the scores might mean - https://www.kaggle.com/competitions/h-and-m-personalized-fashion-recommendations/discussion/314192#1733955
\nAlso, check this notebook which is build on about 10 notebooks and somewhere in there, that idea you mentioned is used. https://www.kaggle.com/code/atulverma/h-m-ensembling-with-lstm
", "rawMarkdown": "See what the scores might mean - https://www.kaggle.com/competitions/h-and-m-personalized-fashion-recommendations/discussion/314192#1733955\n\nAlso, check this notebook which is build on about 10 notebooks and somewhere in there, that idea you mentioned is used. https://www.kaggle.com/code/atulverma/h-m-ensembling-with-lstm", "votes": null } ], "comments": [ { "id": 1732008, "author_name": "atulverma", "author_url": "", "post_date": "03/22/2022 23:25:23", "content": "It is an interesting idea. Not sure how far and where it can be taken, given the constraints and lure of other options.
", "votes": null, "replies": [ { "id": 1732357, "author_name": "andrewrrose", "author_url": "", "post_date": "03/23/2022 09:53:11", "content": "I appreciate it won't be useful for winning the the competition - but I suppose it might be useful in generating predictions for customers with no purchase history. (Indeed - starting with the approach used by all the people who've scored .0225 and then substituting my best general prediction so far for the customers with no history, gives a boost to .0227 despite their only being ~10k such customers.)
", "votes": null, "replies": [] }, { "id": 1733959, "author_name": "atulverma", "author_url": "", "post_date": "03/24/2022 20:09:34", "content": "See what the scores might mean - https://www.kaggle.com/competitions/h-and-m-personalized-fashion-recommendations/discussion/314192#1733955
\nAlso, check this notebook which is build on about 10 notebooks and somewhere in there, that idea you mentioned is used. https://www.kaggle.com/code/atulverma/h-m-ensembling-with-lstm
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1731757": "Out of curiosity, what's the best score that anybody has achieved with **un**personalized recommendations (i.e. exactly the same prediction for every customer)? My first such attempt scored 0.0074.\n\n(I appreciate that this competition is all about the personalization, but I was wondering how far you could get without it.)", "1732008": "It is an interesting idea. Not sure how far and where it can be taken, given the constraints and lure of other options.", "1732357": "I appreciate it won't be useful for winning the the competition - but I suppose it might be useful in generating predictions for customers with no purchase history. (Indeed - starting with the approach used by all the people who've scored .0225 and then substituting my best general prediction so far for the customers with no history, gives a boost to .0227 despite their only being ~10k such customers.)", "1733959": "See what the scores might mean - https://www.kaggle.com/competitions/h-and-m-personalized-fashion-recommendations/discussion/314192#1733955\n\nAlso, check this notebook which is build on about 10 notebooks and somewhere in there, that idea you mentioned is used. https://www.kaggle.com/code/atulverma/h-m-ensembling-with-lstm" }, "source": "meta" }