{
  "id": 42694,
  "title": "Cross Validation (CV), Creating Model Strategy",
  "url": "/competitions/favorita-grocery-sales-forecasting/discussion/42694",
  "author_name": "",
  "post_date": "2017-11-03T07:48:17.108200200Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Could anyone please mentioned what could be the Cross Validation strategy </p>\n\n<ol>\n<li>Which days/weeks/months needs to choose for the training and which needs to be for the test </li>\n<li>Since we need to predict for the month of Aug'2017 (16 days), whether the idea is to create and train model (including CV) on Aug'2016 ? If we need to consider the \"Rolling Forecast Origin\", whether the <strong>training</strong> set would be for Aug'2015 , 1 day / 1 week for the <strong>Testing</strong> going forward</li>\n<li>When we go for the model against store, we need to forecast for that store in the Test (for submission)</li>\n<li>I have gone thorough the \"<strong><a href=\"https://www.kaggle.com/c/rossmann-store-sales\">Rossmann Store Sales</a></strong>\" competition, thanks to <a href=\"https://www.kaggle.com/dongxu027\"><strong>DSEverything</strong></a> for pointing this out, where few folks have mentioned the CV strategy is referred mentioned by <strong><a href=\"https://www.otexts.org/fpp/2/5\">Rob J Hyndman</a></strong> in his book , however the above questions (1) till (4) is raised after been referring</li>\n</ol>\n\n<p>Thanks in advance</p>\n\n<p>Cheers\nSushanta</p>",
  "messages": [
    {
      "id": "239330",
      "postDate": "11/03/2017 07:48:17",
      "content": "<p>Could anyone please mentioned what could be the Cross Validation strategy </p>\n\n<ol>\n<li>Which days/weeks/months needs to choose for the training and which needs to be for the test </li>\n<li>Since we need to predict for the month of Aug'2017 (16 days), whether the idea is to create and train model (including CV) on Aug'2016 ? If we need to consider the \"Rolling Forecast Origin\", whether the <strong>training</strong> set would be for Aug'2015 , 1 day / 1 week for the <strong>Testing</strong> going forward</li>\n<li>When we go for the model against store, we need to forecast for that store in the Test (for submission)</li>\n<li>I have gone thorough the \"<strong><a href=\"https://www.kaggle.com/c/rossmann-store-sales\">Rossmann Store Sales</a></strong>\" competition, thanks to <a href=\"https://www.kaggle.com/dongxu027\"><strong>DSEverything</strong></a> for pointing this out, where few folks have mentioned the CV strategy is referred mentioned by <strong><a href=\"https://www.otexts.org/fpp/2/5\">Rob J Hyndman</a></strong> in his book , however the above questions (1) till (4) is raised after been referring</li>\n</ol>\n\n<p>Thanks in advance</p>\n\n<p>Cheers\nSushanta</p>",
      "rawMarkdown": "Could anyone please mentioned what could be the Cross Validation strategy \n\n 1. Which days/weeks/months needs to choose for the training and which needs to be for the test \n 2. Since we need to predict for the month of Aug'2017 (16 days), whether the idea is to create and train model (including CV) on Aug'2016 ? If we need to consider the \"Rolling Forecast Origin\", whether the **training** set would be for Aug'2015 , 1 day / 1 week for the **Testing** going forward\n 3. When we go for the model against store, we need to forecast for that store in the Test (for submission)\n 4. I have gone thorough the \"**[Rossmann Store Sales][1]**\" competition, thanks to [**DSEverything**][2] for pointing this out, where few folks have mentioned the CV strategy is referred mentioned by **[Rob J Hyndman][3]** in his book , however the above questions (1) till (4) is raised after been referring\n \nThanks in advance\n\nCheers\nSushanta\n\n\n  [1]: https://www.kaggle.com/c/rossmann-store-sales\n  [2]: https://www.kaggle.com/dongxu027\n  [3]: https://www.otexts.org/fpp/2/5",
      "votes": null
    },
    {
      "id": "247384",
      "postDate": "11/22/2017 23:15:12",
      "content": "<p>Cross validation (CV) is usually used when the data size is small, hence, to use or not depend on how are you going to structure your data and ho are you going to define the other parameters of the regressor. Not sure if CV would help, or how much would it help, as you will end up with models equal to the number of CV folds.</p>",
      "rawMarkdown": "Cross validation (CV) is usually used when the data size is small, hence, to use or not depend on how are you going to structure your data and ho are you going to define the other parameters of the regressor. Not sure if CV would help, or how much would it help, as you will end up with models equal to the number of CV folds.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 247384,
      "author_name": "rawi707",
      "author_url": "",
      "post_date": "11/22/2017 23:15:12",
      "content": "<p>Cross validation (CV) is usually used when the data size is small, hence, to use or not depend on how are you going to structure your data and ho are you going to define the other parameters of the regressor. Not sure if CV would help, or how much would it help, as you will end up with models equal to the number of CV folds.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "239330": "Could anyone please mentioned what could be the Cross Validation strategy \n\n 1. Which days/weeks/months needs to choose for the training and which needs to be for the test \n 2. Since we need to predict for the month of Aug'2017 (16 days), whether the idea is to create and train model (including CV) on Aug'2016 ? If we need to consider the \"Rolling Forecast Origin\", whether the **training** set would be for Aug'2015 , 1 day / 1 week for the **Testing** going forward\n 3. When we go for the model against store, we need to forecast for that store in the Test (for submission)\n 4. I have gone thorough the \"**[Rossmann Store Sales][1]**\" competition, thanks to [**DSEverything**][2] for pointing this out, where few folks have mentioned the CV strategy is referred mentioned by **[Rob J Hyndman][3]** in his book , however the above questions (1) till (4) is raised after been referring\n \nThanks in advance\n\nCheers\nSushanta\n\n\n  [1]: https://www.kaggle.com/c/rossmann-store-sales\n  [2]: https://www.kaggle.com/dongxu027\n  [3]: https://www.otexts.org/fpp/2/5",
    "247384": "Cross validation (CV) is usually used when the data size is small, hence, to use or not depend on how are you going to structure your data and ho are you going to define the other parameters of the regressor. Not sure if CV would help, or how much would it help, as you will end up with models equal to the number of CV folds."
  },
  "source": "meta"
}