I come across with a problem in one particular item of the train dataset. There is a gap of roughly 3 years of information. In particular there is reasonable amount of information for the year 2013, and few information for the current year 2017, even in this year there are gaps between days in a month. As you can see in the file attached.\nIn general, the must recent year could give us the best information, but the data of 2013 is more sequential with few gaps between days.
\n\nWhat do you think it would the best year to train a model model for this case ?
I come across with a problem in one particular item of the train dataset. There is a gap of roughly 3 years of information. In particular there is reasonable amount of information for the year 2013, and few information for the current year 2017, even in this year there are gaps between days in a month. As you can see in the file attached.\nIn general, the must recent year could give us the best information, but the data of 2013 is more sequential with few gaps between days.
\n\nWhat do you think it would the best year to train a model model for this case ?
Such kind of data is the Nightmare of Data Analyst, including or not is a problem!
", "rawMarkdown": "Such kind of data is the Nightmare of Data Analyst, including or not is a problem!", "votes": null }, { "id": "243389", "postDate": "11/13/2017 23:17:51", "content": "I would be inclined to dump the history prior to the sequence of 0s (the missing period) - given the overall dynamics of the dataset, what are the chances 4year old information is still relevant?
", "rawMarkdown": "I would be inclined to dump the history prior to the sequence of 0s (the missing period) - given the overall dynamics of the dataset, what are the chances 4year old information is still relevant?", "votes": null }, { "id": "243419", "postDate": "11/14/2017 01:13:17", "content": "I'm currently using the tail end of those regions to help account for the 40k items that have zero training set history.
", "rawMarkdown": "I'm currently using the tail end of those regions to help account for the 40k items that have zero training set history.", "votes": null }, { "id": "246903", "postDate": "11/21/2017 23:24:38", "content": "May I ask which item is this?
", "rawMarkdown": "May I ask which item is this?", "votes": null }, { "id": "247152", "postDate": "11/22/2017 12:59:16", "content": "The item is 96995, in particular the data of the plot is for store 1
", "rawMarkdown": "The item is 96995, in particular the data of the plot is for store 1", "votes": null }, { "id": "247153", "postDate": "11/22/2017 13:02:35", "content": "The item is 96995, in particular the data of the plot is for store 1
", "rawMarkdown": "The item is 96995, in particular the data of the plot is for store 1", "votes": null }, { "id": "247157", "postDate": "11/22/2017 13:08:09", "content": "Thanks, for the comments to be honest I'm new in time series, but I want to learn a lot of this subject with this competition. Indeed, the best results are with the most recent data, I ran some experiments in that matter just to be sure. But in this case in particular I wonder whether it would best to fill the gaps on 2017 assuming there were 0 unit sales on the missing days or simply don't fill the gaps an consider all of 2017 as a sequence independently of the gaps.
", "rawMarkdown": "Thanks, for the comments to be honest I'm new in time series, but I want to learn a lot of this subject with this competition. Indeed, the best results are with the most recent data, I ran some experiments in that matter just to be sure. But in this case in particular I wonder whether it would best to fill the gaps on 2017 assuming there were 0 unit sales on the missing days or simply don't fill the gaps an consider all of 2017 as a sequence independently of the gaps.", "votes": null } ], "comments": [ { "id": 241864, "author_name": "robinhoo1973", "author_url": "", "post_date": "11/10/2017 01:12:22", "content": "Such kind of data is the Nightmare of Data Analyst, including or not is a problem!
", "votes": null, "replies": [] }, { "id": 243389, "author_name": "konradb", "author_url": "", "post_date": "11/13/2017 23:17:51", "content": "I would be inclined to dump the history prior to the sequence of 0s (the missing period) - given the overall dynamics of the dataset, what are the chances 4year old information is still relevant?
", "votes": null, "replies": [ { "id": 243419, "author_name": "vannak", "author_url": "", "post_date": "11/14/2017 01:13:17", "content": "I'm currently using the tail end of those regions to help account for the 40k items that have zero training set history.
", "votes": null, "replies": [] }, { "id": 247157, "author_name": "daniluvatar", "author_url": "", "post_date": "11/22/2017 13:08:09", "content": "Thanks, for the comments to be honest I'm new in time series, but I want to learn a lot of this subject with this competition. Indeed, the best results are with the most recent data, I ran some experiments in that matter just to be sure. But in this case in particular I wonder whether it would best to fill the gaps on 2017 assuming there were 0 unit sales on the missing days or simply don't fill the gaps an consider all of 2017 as a sequence independently of the gaps.
", "votes": null, "replies": [] } ] }, { "id": 246903, "author_name": "rawi707", "author_url": "", "post_date": "11/21/2017 23:24:38", "content": "May I ask which item is this?
", "votes": null, "replies": [ { "id": 247153, "author_name": "daniluvatar", "author_url": "", "post_date": "11/22/2017 13:02:35", "content": "The item is 96995, in particular the data of the plot is for store 1
", "votes": null, "replies": [] } ] }, { "id": 247152, "author_name": "daniluvatar", "author_url": "", "post_date": "11/22/2017 12:59:16", "content": "The item is 96995, in particular the data of the plot is for store 1
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "241833": "I come across with a problem in one particular item of the train dataset. There is a gap of roughly 3 years of information. In particular there is reasonable amount of information for the year 2013, and few information for the current year 2017, even in this year there are gaps between days in a month. As you can see in the file attached.\nIn general, the must recent year could give us the best information, but the data of 2013 is more sequential with few gaps between days.\n\nWhat do you think it would the best year to train a model model for this case ? ![Huge Gaps on Items][1]\n\n \n\n\n [1]: https://drive.google.com/open?id=1US_lii-D_g6nw1-0IHvLzYsmMJKrzZjC", "241864": "Such kind of data is the Nightmare of Data Analyst, including or not is a problem!", "243389": "I would be inclined to dump the history prior to the sequence of 0s (the missing period) - given the overall dynamics of the dataset, what are the chances 4year old information is still relevant?", "243419": "I'm currently using the tail end of those regions to help account for the 40k items that have zero training set history.", "246903": "May I ask which item is this?", "247152": "The item is 96995, in particular the data of the plot is for store 1", "247153": "The item is 96995, in particular the data of the plot is for store 1", "247157": "Thanks, for the comments to be honest I'm new in time series, but I want to learn a lot of this subject with this competition. Indeed, the best results are with the most recent data, I ran some experiments in that matter just to be sure. But in this case in particular I wonder whether it would best to fill the gaps on 2017 assuming there were 0 unit sales on the missing days or simply don't fill the gaps an consider all of 2017 as a sequence independently of the gaps." }, "source": "meta" }