{
  "id": 377608,
  "title": "May I ask how to prepar labels for a re-rank model ??",
  "url": "/competitions/otto-recommender-system/discussion/377608",
  "author_name": "",
  "post_date": "2023-01-12T00:41:35.858433Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Below is my understanding regarding how to set up labels for traning data, please correct me if I'm wrong.</p>\n<p>1 - For a click re-rank model, all the AID with clicks records should have positive label and not clicked AID should have  a negtaive label.<br>\n2 - For a carts re-rank model, all the AID with type carts should have a postive label and AID only with clicks should have negative label.<br>\n3 - For a order re-rank model, all the AID with type orders should have a postive label and AID with clicks and carts only should have negative label.</p>\n<p>One more thing, if we want to build an all-in-one model, so the label could be a implict score from different types,<br>\nLike for a clicked event we can score it as 1, for a carts event we can score it 3 and last for a order event we can score it 5.</p>\n<p>Thanks in advanced for your answer!!! 👍</p>",
  "messages": [
    {
      "id": "2096307",
      "postDate": "01/12/2023 00:41:35",
      "content": "<p>Below is my understanding regarding how to set up labels for traning data, please correct me if I'm wrong.</p>\n<p>1 - For a click re-rank model, all the AID with clicks records should have positive label and not clicked AID should have  a negtaive label.<br>\n2 - For a carts re-rank model, all the AID with type carts should have a postive label and AID only with clicks should have negative label.<br>\n3 - For a order re-rank model, all the AID with type orders should have a postive label and AID with clicks and carts only should have negative label.</p>\n<p>One more thing, if we want to build an all-in-one model, so the label could be a implict score from different types,<br>\nLike for a clicked event we can score it as 1, for a carts event we can score it 3 and last for a order event we can score it 5.</p>\n<p>Thanks in advanced for your answer!!! 👍</p>",
      "rawMarkdown": "Below is my understanding regarding how to set up labels for traning data, please correct me if I'm wrong.\n\n1 - For a click re-rank model, all the AID with clicks records should have positive label and not clicked AID should have  a negtaive label.\n2 - For a carts re-rank model, all the AID with type carts should have a postive label and AID only with clicks should have negative label.\n3 - For a order re-rank model, all the AID with type orders should have a postive label and AID with clicks and carts only should have negative label.\n\nOne more thing, if we want to build an all-in-one model, so the label could be a implict score from different types,\nLike for a clicked event we can score it as 1, for a carts event we can score it 3 and last for a order event we can score it 5.\n\nThanks in advanced for your answer!!! 👍",
      "votes": null
    },
    {
      "id": "2096369",
      "postDate": "01/12/2023 02:00:57",
      "content": "<p>Yes this is correct. For each target type, we make one dataframe of candidates. In each candidate dataframe, for each test user, you need to make <code>X</code> rows of candidates. Then for each of these rows, you add a positive or negative label. So if you have 1000 test users and 50 candidates, then your dataframe has 50,000 rows. </p>",
      "rawMarkdown": "Yes this is correct. For each target type, we make one dataframe of candidates. In each candidate dataframe, for each test user, you need to make `X` rows of candidates. Then for each of these rows, you add a positive or negative label. So if you have 1000 test users and 50 candidates, then your dataframe has 50,000 rows.",
      "votes": null
    },
    {
      "id": "2096387",
      "postDate": "01/12/2023 02:31:04",
      "content": "<p>Thanks for your reply Chris!!</p>",
      "rawMarkdown": "Thanks for your reply Chris!!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2096369,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "01/12/2023 02:00:57",
      "content": "<p>Yes this is correct. For each target type, we make one dataframe of candidates. In each candidate dataframe, for each test user, you need to make <code>X</code> rows of candidates. Then for each of these rows, you add a positive or negative label. So if you have 1000 test users and 50 candidates, then your dataframe has 50,000 rows. </p>",
      "votes": null,
      "replies": [
        {
          "id": 2096387,
          "author_name": "weiqiu",
          "author_url": "",
          "post_date": "01/12/2023 02:31:04",
          "content": "<p>Thanks for your reply Chris!!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2096307": "Below is my understanding regarding how to set up labels for traning data, please correct me if I'm wrong.\n\n1 - For a click re-rank model, all the AID with clicks records should have positive label and not clicked AID should have  a negtaive label.\n2 - For a carts re-rank model, all the AID with type carts should have a postive label and AID only with clicks should have negative label.\n3 - For a order re-rank model, all the AID with type orders should have a postive label and AID with clicks and carts only should have negative label.\n\nOne more thing, if we want to build an all-in-one model, so the label could be a implict score from different types,\nLike for a clicked event we can score it as 1, for a carts event we can score it 3 and last for a order event we can score it 5.\n\nThanks in advanced for your answer!!! 👍",
    "2096369": "Yes this is correct. For each target type, we make one dataframe of candidates. In each candidate dataframe, for each test user, you need to make `X` rows of candidates. Then for each of these rows, you add a positive or negative label. So if you have 1000 test users and 50 candidates, then your dataframe has 50,000 rows.",
    "2096387": "Thanks for your reply Chris!!"
  },
  "source": "meta"
}