{
  "id": 105535,
  "title": "Learning rate and batch size?",
  "url": "/competitions/champs-scalar-coupling/discussion/105535",
  "author_name": "Yassine Alouini",
  "post_date": "2019-08-23T22:17:50.372000",
  "votes": -3,
  "comment_count": 4,
  "views": 0,
  "content": "<p>How small are your learning rates and batch sizes for GNN and NN models?\nSo far, best result with lr around 0.0001 (though starting higher and reducing on plateau) and batch size around 10. \nMaybe we will know after the end of this competition... :)</p>",
  "messages": [
    {
      "id": 606686,
      "postDate": "2019-08-23T23:01:10.530Z",
      "content": "<p>What is GNN? </p>",
      "rawMarkdown": "What is GNN? ",
      "votes": 2,
      "replies": [
        {
          "id": 606706,
          "postDate": "2019-08-23T23:49:45.023Z",
          "content": "<p>Graph Neural Network</p>",
          "rawMarkdown": "Graph Neural Network",
          "votes": 1
        }
      ]
    },
    {
      "id": 606826,
      "postDate": "2019-08-24T06:16:46.283Z",
      "content": "<p>LR 0.001 intially with reduce LR on plateau, Batch Size 256 or 512</p>",
      "rawMarkdown": "LR 0.001 intially with reduce LR on plateau, Batch Size 256 or 512"
    },
    {
      "id": 606675,
      "postDate": "2019-08-23T22:17:50.373Z",
      "content": "<p>How small are your learning rates and batch sizes for GNN and NN models?\nSo far, best result with lr around 0.0001 (though starting higher and reducing on plateau) and batch size around 10. \nMaybe we will know after the end of this competition... :)</p>",
      "rawMarkdown": "How small are your learning rates and batch sizes for GNN and NN models?\nSo far, best result with lr around 0.0001 (though starting higher and reducing on plateau) and batch size around 10. \nMaybe we will know after the end of this competition... :)",
      "votes": -3
    },
    {
      "id": 606701,
      "postDate": "2019-08-23T23:30:46.003Z",
      "rawMarkdown": "",
      "votes": -1,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 606686,
      "author_name": "Osman",
      "author_url": "",
      "post_date": "2019-08-23T23:01:10.530000",
      "content": "<p>What is GNN? </p>",
      "votes": 2,
      "replies": [
        {
          "id": 606706,
          "author_name": "JM",
          "author_url": "",
          "post_date": "2019-08-23T23:49:45.023000",
          "content": "<p>Graph Neural Network</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 606826,
      "author_name": "Artificially Intelligent",
      "author_url": "",
      "post_date": "2019-08-24T06:16:46.283000",
      "content": "<p>LR 0.001 intially with reduce LR on plateau, Batch Size 256 or 512</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 606701,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-08-23T23:30:46.003000",
      "content": "",
      "votes": -1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "606686": "What is GNN? ",
    "606826": "LR 0.001 intially with reduce LR on plateau, Batch Size 256 or 512",
    "606675": "How small are your learning rates and batch sizes for GNN and NN models?\nSo far, best result with lr around 0.0001 (though starting higher and reducing on plateau) and batch size around 10. \nMaybe we will know after the end of this competition... :)",
    "606701": ""
  }
}