{
  "id": 318270,
  "title": "H&M: Plan for modeling (easy stage)",
  "url": "/competitions/h-and-m-personalized-fashion-recommendations/discussion/318270",
  "author_name": "",
  "post_date": "2022-04-11T13:32:00.603044500Z",
  "votes": 2,
  "comment_count": 7,
  "views": 0,
  "content": "<p><strong>Hi everyone!</strong><br>\nHere I will describe my plan for the first (simple) stage of modeling and chat about possible approaches to dive deeper. I welcome any criticism and comments. Write to me!</p>\n<p>My primary way of modeling based on <a href=\"https://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again\" target=\"_blank\">this EDA notebook</a>. <strong>Thanks <a href=\"https://www.kaggle.com/lichtlab\" target=\"_blank\">@lichtlab</a>!</strong> 🔥<br>\n<strong>In first step</strong>, I'm going to take last 3-4 weeks to make first dummy prediction: predict things that a person bought recently. <br>\nIf a person has made less than 12 purchases lately, then fill in the empty spaces with zeros.<br>\nPadding with zeros at the end does not degrade MAP@12.🤥</p>\n<p>Based on <a href=\"https://www.kaggle.com/lichtlab\" target=\"_blank\">@lichtlab</a> 's analysis, I'm guessing it will take nonzero score :)</p>\n<p><strong>In second step</strong>, im going to fill the missing fields from previous step.<br>\nI guess i need to look for the most similar to already purchased products. <br>\nThis can be done using simple algorithms that minimize the distance (cos) between objects in some space. (LGBM, Xgboost and etc.) But here I feel insecure and ask You for advice.<br>\nWhat is the best way to find similar products?</p>\n<p>What problems did you face? Do you like my plan? Write a comment to me! 😆</p>",
  "messages": [
    {
      "id": "1752156",
      "postDate": "04/11/2022 13:32:00",
      "content": "<p><strong>Hi everyone!</strong><br>\nHere I will describe my plan for the first (simple) stage of modeling and chat about possible approaches to dive deeper. I welcome any criticism and comments. Write to me!</p>\n<p>My primary way of modeling based on <a href=\"https://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again\" target=\"_blank\">this EDA notebook</a>. <strong>Thanks <a href=\"https://www.kaggle.com/lichtlab\" target=\"_blank\">@lichtlab</a>!</strong> 🔥<br>\n<strong>In first step</strong>, I'm going to take last 3-4 weeks to make first dummy prediction: predict things that a person bought recently. <br>\nIf a person has made less than 12 purchases lately, then fill in the empty spaces with zeros.<br>\nPadding with zeros at the end does not degrade MAP@12.🤥</p>\n<p>Based on <a href=\"https://www.kaggle.com/lichtlab\" target=\"_blank\">@lichtlab</a> 's analysis, I'm guessing it will take nonzero score :)</p>\n<p><strong>In second step</strong>, im going to fill the missing fields from previous step.<br>\nI guess i need to look for the most similar to already purchased products. <br>\nThis can be done using simple algorithms that minimize the distance (cos) between objects in some space. (LGBM, Xgboost and etc.) But here I feel insecure and ask You for advice.<br>\nWhat is the best way to find similar products?</p>\n<p>What problems did you face? Do you like my plan? Write a comment to me! 😆</p>",
      "rawMarkdown": "**Hi everyone!**\nHere I will describe my plan for the first (simple) stage of modeling and chat about possible approaches to dive deeper. I welcome any criticism and comments. Write to me!\n\nMy primary way of modeling based on [this EDA notebook](https://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again). **Thanks @lichtlab!** 🔥\n**In first step**, I'm going to take last 3-4 weeks to make first dummy prediction: predict things that a person bought recently. \nIf a person has made less than 12 purchases lately, then fill in the empty spaces with zeros.\nPadding with zeros at the end does not degrade MAP@12.🤥\n\nBased on @lichtlab 's analysis, I'm guessing it will take nonzero score :)\n\n**In second step**, im going to fill the missing fields from previous step.\nI guess i need to look for the most similar to already purchased products. \nThis can be done using simple algorithms that minimize the distance (cos) between objects in some space. (LGBM, Xgboost and etc.) But here I feel insecure and ask You for advice.\nWhat is the best way to find similar products?\n\nWhat problems did you face? Do you like my plan? Write a comment to me! 😆",
      "votes": null
    },
    {
      "id": "1752293",
      "postDate": "04/11/2022 15:39:14",
      "content": "<p>Good plan, waiting for the results</p>",
      "rawMarkdown": "Good plan, waiting for the results",
      "votes": null
    },
    {
      "id": "1752470",
      "postDate": "04/11/2022 18:50:25",
      "content": "<p><a href=\"https://www.kaggle.com/matveyspiridonov\" target=\"_blank\">@matveyspiridonov</a> How do you add the empty space with zero.As it takes too much time for even cudf apply to do it.</p>",
      "rawMarkdown": "matveyspiridonov How do you add the empty space with zero.As it takes too much time for even cudf apply to do it.",
      "votes": null
    },
    {
      "id": "1752932",
      "postDate": "04/12/2022 10:04:49",
      "content": "<p>It looks like LGBMRanker is the best model for this competition, since everyone is talking about it and no one is sharing the code :D<br>\nI am currently trying to do something similar to what you are planning. I take the similarity from the product description. I'm guessing it would be a good idea to multiply the distances by the popularity of the product in that week/month. Some products have not been sold for a while (maybe they are out of stock, or are seasonal) it might be worth considering that as well.</p>",
      "rawMarkdown": "It looks like LGBMRanker is the best model for this competition, since everyone is talking about it and no one is sharing the code :D\nI am currently trying to do something similar to what you are planning. I take the similarity from the product description. I'm guessing it would be a good idea to multiply the distances by the popularity of the product in that week/month. Some products have not been sold for a while (maybe they are out of stock, or are seasonal) it might be worth considering that as well.",
      "votes": null
    },
    {
      "id": "1753076",
      "postDate": "04/12/2022 13:07:57",
      "content": "<p>Hi)<br>\nI have not think about capacity yet. Im going to try it today and update discussion after.</p>",
      "rawMarkdown": "Hi)\nI have not think about capacity yet. Im going to try it today and update discussion after.",
      "votes": null
    },
    {
      "id": "1753086",
      "postDate": "04/12/2022 13:17:46",
      "content": "<p>Yeah…<br>\nWeighted the distance with popularity of the product in last weeks its a good idea.</p>",
      "rawMarkdown": "Yeah...\nWeighted the distance with popularity of the product in last weeks its a good idea.",
      "votes": null
    },
    {
      "id": "1754786",
      "postDate": "04/14/2022 02:40:04",
      "content": "<p>I may be completely off but I don't understand why is everybody so reliant on \"customers buy same products again\"? This is garments, not groceries.</p>",
      "rawMarkdown": "I may be completely off but I don't understand why is everybody so reliant on \"customers buy same products again\"? This is garments, not groceries.",
      "votes": null
    },
    {
      "id": "1755623",
      "postDate": "04/14/2022 19:11:09",
      "content": "<p>Hi) <br>\nYou can read this article to understand why I think so.<br>\n<a href=\"https://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again\" target=\"_blank\">https://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again</a></p>",
      "rawMarkdown": "Hi) \nYou can read this article to understand why I think so.\nhttps://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1752293,
      "author_name": "ahmadalkhateeb123",
      "author_url": "",
      "post_date": "04/11/2022 15:39:14",
      "content": "<p>Good plan, waiting for the results</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1752470,
      "author_name": "devanshchowdhury",
      "author_url": "",
      "post_date": "04/11/2022 18:50:25",
      "content": "<p><a href=\"https://www.kaggle.com/matveyspiridonov\" target=\"_blank\">@matveyspiridonov</a> How do you add the empty space with zero.As it takes too much time for even cudf apply to do it.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1753076,
          "author_name": "matveyspiridonov",
          "author_url": "",
          "post_date": "04/12/2022 13:07:57",
          "content": "<p>Hi)<br>\nI have not think about capacity yet. Im going to try it today and update discussion after.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1752932,
      "author_name": "grzegorzzawadzki",
      "author_url": "",
      "post_date": "04/12/2022 10:04:49",
      "content": "<p>It looks like LGBMRanker is the best model for this competition, since everyone is talking about it and no one is sharing the code :D<br>\nI am currently trying to do something similar to what you are planning. I take the similarity from the product description. I'm guessing it would be a good idea to multiply the distances by the popularity of the product in that week/month. Some products have not been sold for a while (maybe they are out of stock, or are seasonal) it might be worth considering that as well.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1753086,
          "author_name": "matveyspiridonov",
          "author_url": "",
          "post_date": "04/12/2022 13:17:46",
          "content": "<p>Yeah…<br>\nWeighted the distance with popularity of the product in last weeks its a good idea.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1754786,
      "author_name": "atulverma",
      "author_url": "",
      "post_date": "04/14/2022 02:40:04",
      "content": "<p>I may be completely off but I don't understand why is everybody so reliant on \"customers buy same products again\"? This is garments, not groceries.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1755623,
          "author_name": "matveyspiridonov",
          "author_url": "",
          "post_date": "04/14/2022 19:11:09",
          "content": "<p>Hi) <br>\nYou can read this article to understand why I think so.<br>\n<a href=\"https://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again\" target=\"_blank\">https://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1752156": "**Hi everyone!**\nHere I will describe my plan for the first (simple) stage of modeling and chat about possible approaches to dive deeper. I welcome any criticism and comments. Write to me!\n\nMy primary way of modeling based on [this EDA notebook](https://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again). **Thanks @lichtlab!** 🔥\n**In first step**, I'm going to take last 3-4 weeks to make first dummy prediction: predict things that a person bought recently. \nIf a person has made less than 12 purchases lately, then fill in the empty spaces with zeros.\nPadding with zeros at the end does not degrade MAP@12.🤥\n\nBased on @lichtlab 's analysis, I'm guessing it will take nonzero score :)\n\n**In second step**, im going to fill the missing fields from previous step.\nI guess i need to look for the most similar to already purchased products. \nThis can be done using simple algorithms that minimize the distance (cos) between objects in some space. (LGBM, Xgboost and etc.) But here I feel insecure and ask You for advice.\nWhat is the best way to find similar products?\n\nWhat problems did you face? Do you like my plan? Write a comment to me! 😆",
    "1752293": "Good plan, waiting for the results",
    "1752470": "matveyspiridonov How do you add the empty space with zero.As it takes too much time for even cudf apply to do it.",
    "1752932": "It looks like LGBMRanker is the best model for this competition, since everyone is talking about it and no one is sharing the code :D\nI am currently trying to do something similar to what you are planning. I take the similarity from the product description. I'm guessing it would be a good idea to multiply the distances by the popularity of the product in that week/month. Some products have not been sold for a while (maybe they are out of stock, or are seasonal) it might be worth considering that as well.",
    "1753076": "Hi)\nI have not think about capacity yet. Im going to try it today and update discussion after.",
    "1753086": "Yeah...\nWeighted the distance with popularity of the product in last weeks its a good idea.",
    "1754786": "I may be completely off but I don't understand why is everybody so reliant on \"customers buy same products again\"? This is garments, not groceries.",
    "1755623": "Hi) \nYou can read this article to understand why I think so.\nhttps://www.kaggle.com/code/lichtlab/do-customers-buy-the-same-products-again"
  },
  "source": "meta"
}