My current model only utilizes the previous unit sales, promotions and nothing else. Is there anyone who uses oil price, holidays etc or external data like weather and get significant improvement? I dont want to learn what you use exactly ofc but I am just curious if these really help.
", "messages": [ { "id": "255699", "postDate": "12/09/2017 20:46:47", "content": "My current model only utilizes the previous unit sales, promotions and nothing else. Is there anyone who uses oil price, holidays etc or external data like weather and get significant improvement? I dont want to learn what you use exactly ofc but I am just curious if these really help.
", "rawMarkdown": "My current model only utilizes the previous unit sales, promotions and nothing else. Is there anyone who uses oil price, holidays etc or external data like weather and get significant improvement? I dont want to learn what you use exactly ofc but I am just curious if these really help.", "votes": null }, { "id": "255854", "postDate": "12/10/2017 10:51:13", "content": "There is a kernel which uses the median of some averages which gets- maybe?- a small improvement by using holidays. https://www.kaggle.com/paulorzp/log-ma-with-special-days-lb-0-529
", "rawMarkdown": "There is a kernel which uses the median of some averages which gets- maybe?- a small improvement by using holidays. https://www.kaggle.com/paulorzp/log-ma-with-special-days-lb-0-529", "votes": null }, { "id": "255878", "postDate": "12/10/2017 12:48:10", "content": "Holidays are confusing. On one hand because they are incomplete(1) and also because stores do not close of holidays as someone might expect(2). Also, there is only one holiday in the test period(3).
\n\nProbably the best, more reliable and targeted information is already in the training dataset.
\n\n(1) We do not have holiday info for the cities of: \"Daule\", \"Babahoyo\" and \"Playas\" and the states of: \"Pichincha\", \"Chimborazo\", \"Bolivar\", \"Pastaza\", \"Tungurahua\", \"Guayas\", \"Los Rios\", \"Azuay\", \"El Oro\" and \"Manabi\".
\n\n(2) They only close: twice a year (all together 25/12 and 01/01), for works (one shop n successive days) and before they exist. Holidays may alter behavior but do not close shops. Exception: Shop 25 is the only shop opening on 01/01.
\n\n(3) There is only one holiday in the test period: (Holiday, Local, Ambato, Fundacion de Ambato) There are 2 stores in Ambato 23 and 50.
", "rawMarkdown": "Holidays are confusing. On one hand because they are incomplete(1) and also because stores do **not** close of holidays as someone might expect(2). Also, there is only one holiday in the test period(3).\n\nProbably the best, more reliable and targeted information is already in the training dataset.\n\n----\n\n(1) We do not have holiday info for the cities of: \"Daule\", \"Babahoyo\" and \"Playas\" and the states of: \"Pichincha\", \"Chimborazo\", \"Bolivar\", \"Pastaza\", \"Tungurahua\", \"Guayas\", \"Los Rios\", \"Azuay\", \"El Oro\" and \"Manabi\".\n\n(2) They only close: twice a year (all together 25/12 and 01/01), for works (one shop n successive days) and before they exist. Holidays may alter behavior but do not close shops. Exception: Shop 25 is the only shop opening on 01/01.\n\n(3) There is only one holiday in the test period: (Holiday, Local, Ambato, Fundacion de Ambato) There are 2 stores in Ambato 23 and 50.", "votes": null }, { "id": "256869", "postDate": "12/12/2017 21:24:02", "content": "My group is new to kaggle, but so far I can confirm qualitatively the comment in the holiday data description about pay day being important. This date of when people are payed in the data does influences how many products are sold. The lag and lead though is something we need to look at.
", "rawMarkdown": "My group is new to kaggle, but so far I can confirm qualitatively the comment in the holiday data description about pay day being important. This date of when people are payed in the data does influences how many products are sold. The lag and lead though is something we need to look at.", "votes": null }, { "id": "261802", "postDate": "12/24/2017 02:27:17", "content": "I use oils and google trends that seem not useful.
", "rawMarkdown": "I use oils and google trends that seem not useful.", "votes": null } ], "comments": [ { "id": 255854, "author_name": "jeveromy", "author_url": "", "post_date": "12/10/2017 10:51:13", "content": "There is a kernel which uses the median of some averages which gets- maybe?- a small improvement by using holidays. https://www.kaggle.com/paulorzp/log-ma-with-special-days-lb-0-529
", "votes": null, "replies": [] }, { "id": 255878, "author_name": "batangas", "author_url": "", "post_date": "12/10/2017 12:48:10", "content": "Holidays are confusing. On one hand because they are incomplete(1) and also because stores do not close of holidays as someone might expect(2). Also, there is only one holiday in the test period(3).
\n\nProbably the best, more reliable and targeted information is already in the training dataset.
\n\n(1) We do not have holiday info for the cities of: \"Daule\", \"Babahoyo\" and \"Playas\" and the states of: \"Pichincha\", \"Chimborazo\", \"Bolivar\", \"Pastaza\", \"Tungurahua\", \"Guayas\", \"Los Rios\", \"Azuay\", \"El Oro\" and \"Manabi\".
\n\n(2) They only close: twice a year (all together 25/12 and 01/01), for works (one shop n successive days) and before they exist. Holidays may alter behavior but do not close shops. Exception: Shop 25 is the only shop opening on 01/01.
\n\n(3) There is only one holiday in the test period: (Holiday, Local, Ambato, Fundacion de Ambato) There are 2 stores in Ambato 23 and 50.
", "votes": null, "replies": [] }, { "id": 256869, "author_name": "skyrmion", "author_url": "", "post_date": "12/12/2017 21:24:02", "content": "My group is new to kaggle, but so far I can confirm qualitatively the comment in the holiday data description about pay day being important. This date of when people are payed in the data does influences how many products are sold. The lag and lead though is something we need to look at.
", "votes": null, "replies": [] }, { "id": 261802, "author_name": "viczyf", "author_url": "", "post_date": "12/24/2017 02:27:17", "content": "I use oils and google trends that seem not useful.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "255699": "My current model only utilizes the previous unit sales, promotions and nothing else. Is there anyone who uses oil price, holidays etc or external data like weather and get significant improvement? I dont want to learn what you use exactly ofc but I am just curious if these really help.", "255854": "There is a kernel which uses the median of some averages which gets- maybe?- a small improvement by using holidays. https://www.kaggle.com/paulorzp/log-ma-with-special-days-lb-0-529", "255878": "Holidays are confusing. On one hand because they are incomplete(1) and also because stores do **not** close of holidays as someone might expect(2). Also, there is only one holiday in the test period(3).\n\nProbably the best, more reliable and targeted information is already in the training dataset.\n\n----\n\n(1) We do not have holiday info for the cities of: \"Daule\", \"Babahoyo\" and \"Playas\" and the states of: \"Pichincha\", \"Chimborazo\", \"Bolivar\", \"Pastaza\", \"Tungurahua\", \"Guayas\", \"Los Rios\", \"Azuay\", \"El Oro\" and \"Manabi\".\n\n(2) They only close: twice a year (all together 25/12 and 01/01), for works (one shop n successive days) and before they exist. Holidays may alter behavior but do not close shops. Exception: Shop 25 is the only shop opening on 01/01.\n\n(3) There is only one holiday in the test period: (Holiday, Local, Ambato, Fundacion de Ambato) There are 2 stores in Ambato 23 and 50.", "256869": "My group is new to kaggle, but so far I can confirm qualitatively the comment in the holiday data description about pay day being important. This date of when people are payed in the data does influences how many products are sold. The lag and lead though is something we need to look at.", "261802": "I use oils and google trends that seem not useful." }, "source": "meta" }