It is being said that \"For each customer_id observed in the training data, you may predict up to 12 labels for the article_id, which is the predicted items a customer will buy in the next 7-day period after the training time period.\"
\nThe max of training time period is 22nd Sep 2020. but not all the customer has done the transaction on 22nd Sep. What will be the prediction time period.
\nIt is being said that \"For each customer_id observed in the training data, you may predict up to 12 labels for the article_id, which is the predicted items a customer will buy in the next 7-day period after the training time period.\"
\nThe max of training time period is 22nd Sep 2020. but not all the customer has done the transaction on 22nd Sep. What will be the prediction time period.
\nThe same prediction time period 23rd Sep2020 to 29th Sep2020 for all customers
", "rawMarkdown": "The same prediction time period 23rd Sep2020 to 29th Sep2020 for all customers", "votes": null }, { "id": "1731614", "postDate": "03/22/2022 14:38:59", "content": "So we do not need to include the date/time of the predicted transactions. We only supply the customer_ID and the article_ids that the customer will buy. The order of the article_ids does not matter, and the max number we can include is 12.
", "rawMarkdown": "So we do not need to include the date/time of the predicted transactions. We only supply the customer_ID and the article_ids that the customer will buy. The order of the article_ids does not matter, and the max number we can include is 12.", "votes": null }, { "id": "1731716", "postDate": "03/22/2022 16:12:16", "content": "Order of the article ids does matter
", "rawMarkdown": "Order of the article ids does matter", "votes": null }, { "id": "1732142", "postDate": "03/23/2022 04:07:53", "content": "Order of the article_id is very much important. please see the evaluation
", "rawMarkdown": "Order of the article_id is very much important. please see the [evaluation](https://www.kaggle.com/competitions/h-and-m-personalized-fashion-recommendations/overview/evaluation)", "votes": null }, { "id": "1733682", "postDate": "03/24/2022 14:10:43", "content": "I have read the evaluation and I also looked at this notebook for MAP:
\nhttps://www.kaggle.com/code/debarshichanda/understanding-mean-average-precision/notebook
\nAre you sure that order matters when computing MAP, could you explain how or refer me to something that explains it more clearly.
", "rawMarkdown": "I have read the evaluation and I also looked at this notebook for MAP:\n\nhttps://www.kaggle.com/code/debarshichanda/understanding-mean-average-precision/notebook\n\nAre you sure that order matters when computing MAP, could you explain how or refer me to something that explains it more clearly.", "votes": null }, { "id": "1733870", "postDate": "03/24/2022 17:18:42", "content": "Order does not matter, if a customer bought 12 articles and all of your 12 predictions are right for this customer.
\nLet's assume the customer bought 12 articles and only one of your 12 predictions is wrong.
\nIf your very first prediction is wrong your prediction will be penalized for every article:
\nP@1=0, P@2=0.5, P@3=0.66, …, P@12=0.92
If only your last prediction is wrong you get a penalization only for the last article:
\nP@1=1, P@2=1, P@3=1, …, P@12=0.92
Basically, if the customer bought m articles(ground truth) (m>0), then the first n predictions(n=min(m,12)), should be match the m articles in any order. If any of first n articles are incorrect, your score will be less than it would/could have been.
\nTo give additional perspective, for a customer buying 12 or more items,
\nThe same prediction time period 23rd Sep2020 to 29th Sep2020 for all customers
", "votes": null, "replies": [] }, { "id": 1731614, "author_name": "peterwarren", "author_url": "", "post_date": "03/22/2022 14:38:59", "content": "So we do not need to include the date/time of the predicted transactions. We only supply the customer_ID and the article_ids that the customer will buy. The order of the article_ids does not matter, and the max number we can include is 12.
", "votes": null, "replies": [ { "id": 1731716, "author_name": "atulverma", "author_url": "", "post_date": "03/22/2022 16:12:16", "content": "Order of the article ids does matter
", "votes": null, "replies": [] }, { "id": 1732142, "author_name": "pankajkumar", "author_url": "", "post_date": "03/23/2022 04:07:53", "content": "Order of the article_id is very much important. please see the evaluation
", "votes": null, "replies": [] }, { "id": 1733682, "author_name": "peterwarren", "author_url": "", "post_date": "03/24/2022 14:10:43", "content": "I have read the evaluation and I also looked at this notebook for MAP:
\nhttps://www.kaggle.com/code/debarshichanda/understanding-mean-average-precision/notebook
\nAre you sure that order matters when computing MAP, could you explain how or refer me to something that explains it more clearly.
", "votes": null, "replies": [] }, { "id": 1733870, "author_name": "tbierhance", "author_url": "", "post_date": "03/24/2022 17:18:42", "content": "Order does not matter, if a customer bought 12 articles and all of your 12 predictions are right for this customer.
\nLet's assume the customer bought 12 articles and only one of your 12 predictions is wrong.
\nIf your very first prediction is wrong your prediction will be penalized for every article:
\nP@1=0, P@2=0.5, P@3=0.66, …, P@12=0.92
If only your last prediction is wrong you get a penalization only for the last article:
\nP@1=1, P@2=1, P@3=1, …, P@12=0.92
Basically, if the customer bought m articles(ground truth) (m>0), then the first n predictions(n=min(m,12)), should be match the m articles in any order. If any of first n articles are incorrect, your score will be less than it would/could have been.
\nTo give additional perspective, for a customer buying 12 or more items,
\n