{
  "id": 392425,
  "title": "Finetuning GraphNeT example model",
  "url": "/competitions/icecube-neutrinos-in-deep-ice/discussion/392425",
  "author_name": "Araik Tamazian",
  "post_date": "2023-03-05T07:50:43.330000",
  "votes": 8,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I've been finetuning GraphNeT example model by changing various parameters and training the model from scratch on batch 1 with validation on batch 51. It seems that original parameters are quite optimal - the only thing that improved my score was to reduce <code>dynedge_layer_sizes</code> in <code>DynEdge</code> to <code>[(128, 256), (336, 256), (336, 256)]</code>.</p>",
  "messages": [
    {
      "id": 2169560,
      "postDate": "2023-03-05T07:50:43.330Z",
      "content": "<p>I've been finetuning GraphNeT example model by changing various parameters and training the model from scratch on batch 1 with validation on batch 51. It seems that original parameters are quite optimal - the only thing that improved my score was to reduce <code>dynedge_layer_sizes</code> in <code>DynEdge</code> to <code>[(128, 256), (336, 256), (336, 256)]</code>.</p>",
      "rawMarkdown": "I've been finetuning GraphNeT example model by changing various parameters and training the model from scratch on batch 1 with validation on batch 51. It seems that original parameters are quite optimal - the only thing that improved my score was to reduce `dynedge_layer_sizes` in `DynEdge` to `[(128, 256), (336, 256), (336, 256)]`.",
      "votes": 8
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2169560": "I've been finetuning GraphNeT example model by changing various parameters and training the model from scratch on batch 1 with validation on batch 51. It seems that original parameters are quite optimal - the only thing that improved my score was to reduce `dynedge_layer_sizes` in `DynEdge` to `[(128, 256), (336, 256), (336, 256)]`."
  }
}