{
  "id": 319891,
  "title": "3 week challenge | update everyday",
  "url": "/competitions/h-and-m-personalized-fashion-recommendations/discussion/319891",
  "author_name": "",
  "post_date": "2022-04-19T09:23:18.946168500Z",
  "votes": 18,
  "comment_count": 5,
  "views": 0,
  "content": "<h2>finally rank 48th,  i will do some summary later,  and thanks for this comp,  learned a lot about recommendation.</h2>\n<p>There are three weeks left in this comp. Time is tight, so i will update what i do everyday to push myself 😄. This is a very interesting comp, hope to enjoy it with everyone.</p>\n<p>------- 19 Aril -------</p>\n<ol>\n<li>dive into public EDA &amp; baseline;</li>\n<li>search CV method.</li>\n</ol>\n<p>------- 20 Aril -------<br>\nno leisure … 😭</p>\n<p>------- 21 Aril -------</p>\n<ol>\n<li>Data Analysis;</li>\n<li>Process 3-fold ;<br>\n</li>\n<li>Learn recall model likes :  youtubeDNN DIN DSSM</li>\n<li>try some new recall method;</li>\n<li>implement evaluation function for recall &amp; rank. </li>\n</ol>\n<p>------- 22 Aril -------</p>\n<ol>\n<li>Test all public recall cv map@12, recall ,  precision;</li>\n<li>build sequence base recall model.</li>\n</ol>\n<p>------- 23-24 Aril -------<br>\nhave rest ..</p>\n<p>------- 25 Aril -------<br>\nModify bert model/data pipeline/ trainer,  try recall model baseline</p>\n<p>-------- 26 Aril -------<br>\nTransformer Model did not work. …. even worse than GRU4rec ,  cv only  0.01.<br>\nTry add more information to pytorch-baseline code.</p>\n<p>-------- 27 Aril -------<br>\nAdd feature to pytorch-baseline model,  got cv 0.027,  will try do submision today<br>\nbaseline cv 0.023, lb 0.0191<br>\nAdd feat cv 0.027, lb0.022</p>\n<p>-------- 28 Aril -------<br>\nTrain article embedding with some labeled and unlabel task.<br>\nA bit cv improve.</p>\n<p>-------- 29 Aril -------<br>\nNo time do kaggle 😓</p>\n<p>-------- 30 Aril -------</p>\n<ol>\n<li>Define record dataframe format.</li>\n<li>Combine recall data (me and my teamate)</li>\n<li>Rank Model baseline</li>\n</ol>\n<p>-------- 1 May -------<br>\nNo time do kaggle 💔</p>\n<p>-------- 2 May -------</p>\n<ol>\n<li>Ranker boost cv 0.001 when just use pytorch-baseline recall model;</li>\n<li>combine multiple-recall and  train ranker</li>\n</ol>\n<p>-------- 4 May -------<br>\nfinetune ranker  recall parameter,  cv improve to 0.0294</p>\n<p>-------- 5 May -------<br>\nMerge 5 recall method, cv got 0.0303 but cv decrease ..</p>\n<p>-------- 6 May -------<br>\nTry new customer and article recall,  feature engineering in rank model</p>\n<p>-------- 7 May -------<br>\nteamates feature engineering work,  try DeepFm ranker.<br>\ncontinue work for cold start recall</p>",
  "messages": [
    {
      "id": "1760425",
      "postDate": "04/19/2022 09:23:18",
      "content": "<h2>finally rank 48th,  i will do some summary later,  and thanks for this comp,  learned a lot about recommendation.</h2>\n<p>There are three weeks left in this comp. Time is tight, so i will update what i do everyday to push myself 😄. This is a very interesting comp, hope to enjoy it with everyone.</p>\n<p>------- 19 Aril -------</p>\n<ol>\n<li>dive into public EDA &amp; baseline;</li>\n<li>search CV method.</li>\n</ol>\n<p>------- 20 Aril -------<br>\nno leisure … 😭</p>\n<p>------- 21 Aril -------</p>\n<ol>\n<li>Data Analysis;</li>\n<li>Process 3-fold ;<br>\n</li>\n<li>Learn recall model likes :  youtubeDNN DIN DSSM</li>\n<li>try some new recall method;</li>\n<li>implement evaluation function for recall &amp; rank. </li>\n</ol>\n<p>------- 22 Aril -------</p>\n<ol>\n<li>Test all public recall cv map@12, recall ,  precision;</li>\n<li>build sequence base recall model.</li>\n</ol>\n<p>------- 23-24 Aril -------<br>\nhave rest ..</p>\n<p>------- 25 Aril -------<br>\nModify bert model/data pipeline/ trainer,  try recall model baseline</p>\n<p>-------- 26 Aril -------<br>\nTransformer Model did not work. …. even worse than GRU4rec ,  cv only  0.01.<br>\nTry add more information to pytorch-baseline code.</p>\n<p>-------- 27 Aril -------<br>\nAdd feature to pytorch-baseline model,  got cv 0.027,  will try do submision today<br>\nbaseline cv 0.023, lb 0.0191<br>\nAdd feat cv 0.027, lb0.022</p>\n<p>-------- 28 Aril -------<br>\nTrain article embedding with some labeled and unlabel task.<br>\nA bit cv improve.</p>\n<p>-------- 29 Aril -------<br>\nNo time do kaggle 😓</p>\n<p>-------- 30 Aril -------</p>\n<ol>\n<li>Define record dataframe format.</li>\n<li>Combine recall data (me and my teamate)</li>\n<li>Rank Model baseline</li>\n</ol>\n<p>-------- 1 May -------<br>\nNo time do kaggle 💔</p>\n<p>-------- 2 May -------</p>\n<ol>\n<li>Ranker boost cv 0.001 when just use pytorch-baseline recall model;</li>\n<li>combine multiple-recall and  train ranker</li>\n</ol>\n<p>-------- 4 May -------<br>\nfinetune ranker  recall parameter,  cv improve to 0.0294</p>\n<p>-------- 5 May -------<br>\nMerge 5 recall method, cv got 0.0303 but cv decrease ..</p>\n<p>-------- 6 May -------<br>\nTry new customer and article recall,  feature engineering in rank model</p>\n<p>-------- 7 May -------<br>\nteamates feature engineering work,  try DeepFm ranker.<br>\ncontinue work for cold start recall</p>",
      "rawMarkdown": "finally rank 48th,  i will do some summary later,  and thanks for this comp,  learned a lot about recommendation.\n-------- \n\nThere are three weeks left in this comp. Time is tight, so i will update what i do everyday to push myself 😄. This is a very interesting comp, hope to enjoy it with everyone.\n\n------- 19 Aril -------\n1. dive into public EDA & baseline;\n2. search CV method.\n\n------- 20 Aril -------\nno leisure ... 😭\n\n------- 21 Aril -------\n1. Data Analysis;\n2. Process 3-fold ;\n~~3. understand and rebuild public recall method.~~\n3. Learn recall model likes :  youtubeDNN DIN DSSM\n4. try some new recall method;\n5. implement evaluation function for recall & rank. \n\n------- 22 Aril -------\n1. Test all public recall cv map@12, recall ,  precision;\n2. build sequence base recall model.\n\n------- 23-24 Aril -------\nhave rest ..\n\n------- 25 Aril -------\nModify bert model/data pipeline/ trainer,  try recall model baseline\n\n-------- 26 Aril -------\nTransformer Model did not work. .... even worse than GRU4rec ,  cv only  0.01.\nTry add more information to pytorch-baseline code.\n\n-------- 27 Aril -------\nAdd feature to pytorch-baseline model,  got cv 0.027,  will try do submision today\nbaseline cv 0.023, lb 0.0191\nAdd feat cv 0.027, lb0.022\n\n-------- 28 Aril -------\nTrain article embedding with some labeled and unlabel task.\nA bit cv improve.\n\n-------- 29 Aril -------\nNo time do kaggle 😓\n\n-------- 30 Aril -------\n1. Define record dataframe format.\n2. Combine recall data (me and my teamate)\n3. Rank Model baseline\n\n-------- 1 May -------\nNo time do kaggle 💔\n\n-------- 2 May -------\n1. Ranker boost cv 0.001 when just use pytorch-baseline recall model;\n2. combine multiple-recall and  train ranker\n\n-------- 4 May -------\nfinetune ranker  recall parameter,  cv improve to 0.0294\n\n-------- 5 May -------\nMerge 5 recall method, cv got 0.0303 but cv decrease ..\n\n-------- 6 May -------\nTry new customer and article recall,  feature engineering in rank model\n\n-------- 7 May -------\nteamates feature engineering work,  try DeepFm ranker.\ncontinue work for cold start recall",
      "votes": null
    },
    {
      "id": "1761819",
      "postDate": "04/20/2022 07:57:36",
      "content": "<p>Haha, it's also pushing others to work harder. Thanks my friend😄</p>",
      "rawMarkdown": "Haha, it's also pushing others to work harder. Thanks my friend😄",
      "votes": null
    },
    {
      "id": "1765322",
      "postDate": "04/23/2022 10:43:04",
      "content": "<p>Exciting challenge</p>",
      "rawMarkdown": "Exciting challenge",
      "votes": null
    },
    {
      "id": "1766509",
      "postDate": "04/24/2022 15:53:05",
      "content": "<p>This is a very interesting competition. It's great to see people sharing their methods and working together to improve. I'm looking forward to seeing the final results!</p>",
      "rawMarkdown": "This is a very interesting competition. It's great to see people sharing their methods and working together to improve. I'm looking forward to seeing the final results!",
      "votes": null
    },
    {
      "id": "1782449",
      "postDate": "05/09/2022 15:32:32",
      "content": "<p>very impressive progression!</p>",
      "rawMarkdown": "very impressive progression!",
      "votes": null
    },
    {
      "id": "1782976",
      "postDate": "05/10/2022 03:23:50",
      "content": "<p>😂  congratulation solo silver medal </p>",
      "rawMarkdown": "😂  congratulation solo silver medal",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1761819,
      "author_name": "",
      "author_url": "",
      "post_date": "04/20/2022 07:57:36",
      "content": "<p>Haha, it's also pushing others to work harder. Thanks my friend😄</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1765322,
      "author_name": "cherrizhu",
      "author_url": "",
      "post_date": "04/23/2022 10:43:04",
      "content": "<p>Exciting challenge</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1766509,
      "author_name": "",
      "author_url": "",
      "post_date": "04/24/2022 15:53:05",
      "content": "<p>This is a very interesting competition. It's great to see people sharing their methods and working together to improve. I'm looking forward to seeing the final results!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1782449,
      "author_name": "jacob34",
      "author_url": "",
      "post_date": "05/09/2022 15:32:32",
      "content": "<p>very impressive progression!</p>",
      "votes": null,
      "replies": [
        {
          "id": 1782976,
          "author_name": "evilpsycho42",
          "author_url": "",
          "post_date": "05/10/2022 03:23:50",
          "content": "<p>😂  congratulation solo silver medal </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1760425": "finally rank 48th,  i will do some summary later,  and thanks for this comp,  learned a lot about recommendation.\n-------- \n\nThere are three weeks left in this comp. Time is tight, so i will update what i do everyday to push myself 😄. This is a very interesting comp, hope to enjoy it with everyone.\n\n------- 19 Aril -------\n1. dive into public EDA & baseline;\n2. search CV method.\n\n------- 20 Aril -------\nno leisure ... 😭\n\n------- 21 Aril -------\n1. Data Analysis;\n2. Process 3-fold ;\n~~3. understand and rebuild public recall method.~~\n3. Learn recall model likes :  youtubeDNN DIN DSSM\n4. try some new recall method;\n5. implement evaluation function for recall & rank. \n\n------- 22 Aril -------\n1. Test all public recall cv map@12, recall ,  precision;\n2. build sequence base recall model.\n\n------- 23-24 Aril -------\nhave rest ..\n\n------- 25 Aril -------\nModify bert model/data pipeline/ trainer,  try recall model baseline\n\n-------- 26 Aril -------\nTransformer Model did not work. .... even worse than GRU4rec ,  cv only  0.01.\nTry add more information to pytorch-baseline code.\n\n-------- 27 Aril -------\nAdd feature to pytorch-baseline model,  got cv 0.027,  will try do submision today\nbaseline cv 0.023, lb 0.0191\nAdd feat cv 0.027, lb0.022\n\n-------- 28 Aril -------\nTrain article embedding with some labeled and unlabel task.\nA bit cv improve.\n\n-------- 29 Aril -------\nNo time do kaggle 😓\n\n-------- 30 Aril -------\n1. Define record dataframe format.\n2. Combine recall data (me and my teamate)\n3. Rank Model baseline\n\n-------- 1 May -------\nNo time do kaggle 💔\n\n-------- 2 May -------\n1. Ranker boost cv 0.001 when just use pytorch-baseline recall model;\n2. combine multiple-recall and  train ranker\n\n-------- 4 May -------\nfinetune ranker  recall parameter,  cv improve to 0.0294\n\n-------- 5 May -------\nMerge 5 recall method, cv got 0.0303 but cv decrease ..\n\n-------- 6 May -------\nTry new customer and article recall,  feature engineering in rank model\n\n-------- 7 May -------\nteamates feature engineering work,  try DeepFm ranker.\ncontinue work for cold start recall",
    "1761819": "Haha, it's also pushing others to work harder. Thanks my friend😄",
    "1765322": "Exciting challenge",
    "1766509": "This is a very interesting competition. It's great to see people sharing their methods and working together to improve. I'm looking forward to seeing the final results!",
    "1782449": "very impressive progression!",
    "1782976": "😂  congratulation solo silver medal"
  },
  "source": "meta"
}