{
  "id": 378454,
  "title": "Do we need to use different co-visitation matrix for train and inference ?",
  "url": "/competitions/otto-recommender-system/discussion/378454",
  "author_name": "",
  "post_date": "2023-01-15T21:30:12.892956400Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello community, I have a question about the candidate generation for both train and test.</p>\n<p>Are your candidates generated from the same co-visitation matrices for both train and test ?</p>\n<p>Thanks</p>",
  "messages": [
    {
      "id": "2101366",
      "postDate": "01/15/2023 21:30:12",
      "content": "<p>Hello community, I have a question about the candidate generation for both train and test.</p>\n<p>Are your candidates generated from the same co-visitation matrices for both train and test ?</p>\n<p>Thanks</p>",
      "rawMarkdown": "Hello community, I have a question about the candidate generation for both train and test.\n\n\nAre your candidates generated from the same co-visitation matrices for both train and test ?\n\n\nThanks",
      "votes": null
    },
    {
      "id": "2101403",
      "postDate": "01/15/2023 22:13:39",
      "content": "<p>As we can use leakage - for train I believe should be using train and test splits for local validation (3 weeks + 1 week) and for inference - you should be using the train and test splits provided by Kaggle (4 weeks + 1 week). Effectively, you will be using different covisitation matrix for train and inference. </p>",
      "rawMarkdown": "As we can use leakage - for train I believe should be using train and test splits for local validation (3 weeks + 1 week) and for inference - you should be using the train and test splits provided by Kaggle (4 weeks + 1 week). Effectively, you will be using different covisitation matrix for train and inference.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2101403,
      "author_name": "parthpankajtiwary",
      "author_url": "",
      "post_date": "01/15/2023 22:13:39",
      "content": "<p>As we can use leakage - for train I believe should be using train and test splits for local validation (3 weeks + 1 week) and for inference - you should be using the train and test splits provided by Kaggle (4 weeks + 1 week). Effectively, you will be using different covisitation matrix for train and inference. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2101366": "Hello community, I have a question about the candidate generation for both train and test.\n\n\nAre your candidates generated from the same co-visitation matrices for both train and test ?\n\n\nThanks",
    "2101403": "As we can use leakage - for train I believe should be using train and test splits for local validation (3 weeks + 1 week) and for inference - you should be using the train and test splits provided by Kaggle (4 weeks + 1 week). Effectively, you will be using different covisitation matrix for train and inference."
  },
  "source": "meta"
}