{
  "id": 411255,
  "title": "👀 Embeddings for Catboost",
  "url": "/competitions/predict-student-performance-from-game-play/discussion/411255",
  "author_name": "",
  "post_date": "2023-05-18T11:53:17.706858Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi✋👀<br>\nAs I write there <a href=\"https://www.kaggle.com/code/ivanisaev/catboost-with-coordinates-features-patches/\" target=\"_blank\">https://www.kaggle.com/code/ivanisaev/catboost-with-coordinates-features-patches/</a> I have an idea to add embedding features t to Catboost model. I still believe that addition of features with sequences of 'event+coordinates' steps could improve model's quality. 🤔<br>\nMy hypothesis is that if features with patches statistics didn't significantly i improved quality then possibly embedding feature witch encodes the information about 'event+coordinates' sequences could do that.🙄<br>\nCutboost supports embedding features <a href=\"https://catboost.ai/en/docs//features/embeddings-features\" target=\"_blank\">https://catboost.ai/en/docs//features/embeddings-features</a> . But all examples with embeddings I found in official Catboost repo are about texts and not about seq2seq prediction. And yes tests is also could treated as seq2seq prediction but ideally I want to find example like \"churn prediction with embedding features with Catboost\".🙂<br>\nSo the most closer thing I found is pythorch-lifestream <a href=\"https://github.com/dllllb/pytorch-lifestream#demo-notebooks\" target=\"_blank\">https://github.com/dllllb/pytorch-lifestream#demo-notebooks</a><br>\nI even reached the developers to understand how to fix deprecated dependencies in code. Now trying to run this engine and will share the results.😉<br>\nAnd could you please share what other alternatives of use cases and libs for embedding features with Catboost for predictive analytics (not just texts) did you see/try? <br>\nThank you!</p>",
  "messages": [
    {
      "id": "2264392",
      "postDate": "05/18/2023 11:53:17",
      "content": "<p>Hi✋👀<br>\nAs I write there <a href=\"https://www.kaggle.com/code/ivanisaev/catboost-with-coordinates-features-patches/\" target=\"_blank\">https://www.kaggle.com/code/ivanisaev/catboost-with-coordinates-features-patches/</a> I have an idea to add embedding features t to Catboost model. I still believe that addition of features with sequences of 'event+coordinates' steps could improve model's quality. 🤔<br>\nMy hypothesis is that if features with patches statistics didn't significantly i improved quality then possibly embedding feature witch encodes the information about 'event+coordinates' sequences could do that.🙄<br>\nCutboost supports embedding features <a href=\"https://catboost.ai/en/docs//features/embeddings-features\" target=\"_blank\">https://catboost.ai/en/docs//features/embeddings-features</a> . But all examples with embeddings I found in official Catboost repo are about texts and not about seq2seq prediction. And yes tests is also could treated as seq2seq prediction but ideally I want to find example like \"churn prediction with embedding features with Catboost\".🙂<br>\nSo the most closer thing I found is pythorch-lifestream <a href=\"https://github.com/dllllb/pytorch-lifestream#demo-notebooks\" target=\"_blank\">https://github.com/dllllb/pytorch-lifestream#demo-notebooks</a><br>\nI even reached the developers to understand how to fix deprecated dependencies in code. Now trying to run this engine and will share the results.😉<br>\nAnd could you please share what other alternatives of use cases and libs for embedding features with Catboost for predictive analytics (not just texts) did you see/try? <br>\nThank you!</p>",
      "rawMarkdown": "Hi✋👀\nAs I write there https://www.kaggle.com/code/ivanisaev/catboost-with-coordinates-features-patches/ I have an idea to add embedding features t to Catboost model. I still believe that addition of features with sequences of 'event+coordinates' steps could improve model's quality. 🤔\nMy hypothesis is that if features with patches statistics didn't significantly i improved quality then possibly embedding feature witch encodes the information about 'event+coordinates' sequences could do that.🙄\nCutboost supports embedding features https://catboost.ai/en/docs//features/embeddings-features . But all examples with embeddings I found in official Catboost repo are about texts and not about seq2seq prediction. And yes tests is also could treated as seq2seq prediction but ideally I want to find example like \"churn prediction with embedding features with Catboost\".🙂\nSo the most closer thing I found is pythorch-lifestream https://github.com/dllllb/pytorch-lifestream#demo-notebooks\nI even reached the developers to understand how to fix deprecated dependencies in code. Now trying to run this engine and will share the results.😉\nAnd could you please share what other alternatives of use cases and libs for embedding features with Catboost for predictive analytics (not just texts) did you see/try? \nThank you!",
      "votes": null
    },
    {
      "id": "2280952",
      "postDate": "05/30/2023 13:28:04",
      "content": "<p>I implemented this approach💪🙂 Results are there <a href=\"https://www.kaggle.com/discussions/general/413712#2280949\" target=\"_blank\">https://www.kaggle.com/discussions/general/413712#2280949</a></p>",
      "rawMarkdown": "I implemented this approach💪🙂 Results are there https://www.kaggle.com/discussions/general/413712#2280949",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2280952,
      "author_name": "ivanisaev",
      "author_url": "",
      "post_date": "05/30/2023 13:28:04",
      "content": "<p>I implemented this approach💪🙂 Results are there <a href=\"https://www.kaggle.com/discussions/general/413712#2280949\" target=\"_blank\">https://www.kaggle.com/discussions/general/413712#2280949</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2264392": "Hi✋👀\nAs I write there https://www.kaggle.com/code/ivanisaev/catboost-with-coordinates-features-patches/ I have an idea to add embedding features t to Catboost model. I still believe that addition of features with sequences of 'event+coordinates' steps could improve model's quality. 🤔\nMy hypothesis is that if features with patches statistics didn't significantly i improved quality then possibly embedding feature witch encodes the information about 'event+coordinates' sequences could do that.🙄\nCutboost supports embedding features https://catboost.ai/en/docs//features/embeddings-features . But all examples with embeddings I found in official Catboost repo are about texts and not about seq2seq prediction. And yes tests is also could treated as seq2seq prediction but ideally I want to find example like \"churn prediction with embedding features with Catboost\".🙂\nSo the most closer thing I found is pythorch-lifestream https://github.com/dllllb/pytorch-lifestream#demo-notebooks\nI even reached the developers to understand how to fix deprecated dependencies in code. Now trying to run this engine and will share the results.😉\nAnd could you please share what other alternatives of use cases and libs for embedding features with Catboost for predictive analytics (not just texts) did you see/try? \nThank you!",
    "2280952": "I implemented this approach💪🙂 Results are there https://www.kaggle.com/discussions/general/413712#2280949"
  },
  "source": "meta"
}