{
  "id": 291609,
  "title": "Does anyone tried RL model?",
  "url": "/competitions/g-research-crypto-forecasting/discussion/291609",
  "author_name": "",
  "post_date": "2021-11-30T07:27:45.999012300Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I wonder if anyone use rl method in this competition.</p>",
  "messages": [
    {
      "id": "1600175",
      "postDate": "11/30/2021 07:27:46",
      "content": "<p>I wonder if anyone use rl method in this competition.</p>",
      "rawMarkdown": "I wonder if anyone use rl method in this competition.",
      "votes": null
    },
    {
      "id": "1600236",
      "postDate": "11/30/2021 08:22:44",
      "content": "<p>I can't formulate an argument as to why it would work in our case. It's not like we have a stack of cash to manage over time, with feedback and timing… I would be happy to be wrong and someone manage to get it working better than standard ML. For the moment we have a starter <a href=\"https://www.kaggle.com/yamqwe/g-research-reinforcement-learning-starter\" target=\"_blank\">https://www.kaggle.com/yamqwe/g-research-reinforcement-learning-starter</a> but I wouldn't consider it a good starter (no feature engineering, loss not corresponding to the metric, no trust in convergence, no submission…). </p>",
      "rawMarkdown": "I can't formulate an argument as to why it would work in our case. It's not like we have a stack of cash to manage over time, with feedback and timing... I would be happy to be wrong and someone manage to get it working better than standard ML. For the moment we have a starter https://www.kaggle.com/yamqwe/g-research-reinforcement-learning-starter but I wouldn't consider it a good starter (no feature engineering, loss not corresponding to the metric, no trust in convergence, no submission...).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1600236,
      "author_name": "lucasmorin",
      "author_url": "",
      "post_date": "11/30/2021 08:22:44",
      "content": "<p>I can't formulate an argument as to why it would work in our case. It's not like we have a stack of cash to manage over time, with feedback and timing… I would be happy to be wrong and someone manage to get it working better than standard ML. For the moment we have a starter <a href=\"https://www.kaggle.com/yamqwe/g-research-reinforcement-learning-starter\" target=\"_blank\">https://www.kaggle.com/yamqwe/g-research-reinforcement-learning-starter</a> but I wouldn't consider it a good starter (no feature engineering, loss not corresponding to the metric, no trust in convergence, no submission…). </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1600175": "I wonder if anyone use rl method in this competition.",
    "1600236": "I can't formulate an argument as to why it would work in our case. It's not like we have a stack of cash to manage over time, with feedback and timing... I would be happy to be wrong and someone manage to get it working better than standard ML. For the moment we have a starter https://www.kaggle.com/yamqwe/g-research-reinforcement-learning-starter but I wouldn't consider it a good starter (no feature engineering, loss not corresponding to the metric, no trust in convergence, no submission...)."
  },
  "source": "meta"
}