{
  "id": 287380,
  "title": "Does your model pays attention to the 'Volume'  feature ? ",
  "url": "/competitions/g-research-crypto-forecasting/discussion/287380",
  "author_name": "",
  "post_date": "2021-11-13T15:24:27.616965Z",
  "votes": 11,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi ,<br>\nThis is to ask that if the models really pay attention to the 'Volume' feature i did a small test to a neural network over here using my own pretty robust method <a href=\"https://www.kaggle.com/swaralipibose/feature-importance-with-gradients\" target=\"_blank\">https://www.kaggle.com/swaralipibose/feature-importance-with-gradients</a> <br>\nFrom the plots we see the Volume feature does not matter at all maybe very very less .  I want to ask if your model is also doing the same ? Thanks .</p>\n<p>In fact most of the features which are performing good are heavily correlated to each other <a href=\"https://www.kaggle.com/swaralipibose/comparing-train-files-with-eda-on-crypto\" target=\"_blank\">https://www.kaggle.com/swaralipibose/comparing-train-files-with-eda-on-crypto</a> </p>\n<p>here is the img :-</p>\n<p><a href=\"https://ibb.co/LdTSbzM\" target=\"_blank\">https://ibb.co/LdTSbzM</a></p>",
  "messages": [
    {
      "id": "1581276",
      "postDate": "11/13/2021 15:24:27",
      "content": "<p>Hi ,<br>\nThis is to ask that if the models really pay attention to the 'Volume' feature i did a small test to a neural network over here using my own pretty robust method <a href=\"https://www.kaggle.com/swaralipibose/feature-importance-with-gradients\" target=\"_blank\">https://www.kaggle.com/swaralipibose/feature-importance-with-gradients</a> <br>\nFrom the plots we see the Volume feature does not matter at all maybe very very less .  I want to ask if your model is also doing the same ? Thanks .</p>\n<p>In fact most of the features which are performing good are heavily correlated to each other <a href=\"https://www.kaggle.com/swaralipibose/comparing-train-files-with-eda-on-crypto\" target=\"_blank\">https://www.kaggle.com/swaralipibose/comparing-train-files-with-eda-on-crypto</a> </p>\n<p>here is the img :-</p>\n<p><a href=\"https://ibb.co/LdTSbzM\" target=\"_blank\">https://ibb.co/LdTSbzM</a></p>",
      "rawMarkdown": "Hi ,\nThis is to ask that if the models really pay attention to the 'Volume' feature i did a small test to a neural network over here using my own pretty robust method https://www.kaggle.com/swaralipibose/feature-importance-with-gradients \nFrom the plots we see the Volume feature does not matter at all maybe very very less .  I want to ask if your model is also doing the same ? Thanks .\n\nIn fact most of the features which are performing good are heavily correlated to each other https://www.kaggle.com/swaralipibose/comparing-train-files-with-eda-on-crypto \n\nhere is the img :-\n \nhttps://ibb.co/LdTSbzM",
      "votes": null
    },
    {
      "id": "1581997",
      "postDate": "11/14/2021 12:29:54",
      "content": "<p>O, L H, C prices are naturally highly correlated (and hihgly dependent on assets) and are not so meaningfull themselves. I tried different strategies, standardizing them, encoding them in more meaningfull features (log returns, estimated volatilies) etc.  After this work, volumes appears quite high in my importance graphs. More to come very soon.  </p>",
      "rawMarkdown": "O, L H, C prices are naturally highly correlated (and hihgly dependent on assets) and are not so meaningfull themselves. I tried different strategies, standardizing them, encoding them in more meaningfull features (log returns, estimated volatilies) etc.  After this work, volumes appears quite high in my importance graphs. More to come very soon.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1581997,
      "author_name": "lucasmorin",
      "author_url": "",
      "post_date": "11/14/2021 12:29:54",
      "content": "<p>O, L H, C prices are naturally highly correlated (and hihgly dependent on assets) and are not so meaningfull themselves. I tried different strategies, standardizing them, encoding them in more meaningfull features (log returns, estimated volatilies) etc.  After this work, volumes appears quite high in my importance graphs. More to come very soon.  </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1581276": "Hi ,\nThis is to ask that if the models really pay attention to the 'Volume' feature i did a small test to a neural network over here using my own pretty robust method https://www.kaggle.com/swaralipibose/feature-importance-with-gradients \nFrom the plots we see the Volume feature does not matter at all maybe very very less .  I want to ask if your model is also doing the same ? Thanks .\n\nIn fact most of the features which are performing good are heavily correlated to each other https://www.kaggle.com/swaralipibose/comparing-train-files-with-eda-on-crypto \n\nhere is the img :-\n \nhttps://ibb.co/LdTSbzM",
    "1581997": "O, L H, C prices are naturally highly correlated (and hihgly dependent on assets) and are not so meaningfull themselves. I tried different strategies, standardizing them, encoding them in more meaningfull features (log returns, estimated volatilies) etc.  After this work, volumes appears quite high in my importance graphs. More to come very soon."
  },
  "source": "meta"
}