{
  "id": 547877,
  "title": "why don't we build a CNN model with 8 or 9 layers ",
  "url": "/competitions/czii-cryo-et-object-identification/discussion/547877",
  "author_name": "ulasdesouza",
  "post_date": "2024-11-23T21:57:39.499000",
  "votes": 0,
  "comment_count": 3,
  "views": 0,
  "content": "<p>let's use a model where we give relu to each layer and softmax to the last layer, why are we making things so complicated?</p>",
  "messages": [
    {
      "id": 3060052,
      "postDate": "2024-12-01T11:30:51.497Z",
      "content": "<p>I tried it.  Learned a lot.  Abandoned it.  You can actually get really good results for apo-ferritin, and possibly ribosomes too.  Of course, you can get around 0.4 with ribosomes using just a single convolution layer in which you load the weights from the results of linear regression.  The others…even viruses, were frustratingly difficult though.  Best score I was able to come away with was 0.221 before it became apparent it wasn't going to happen.</p>",
      "rawMarkdown": "I tried it.  Learned a lot.  Abandoned it.  You can actually get really good results for apo-ferritin, and possibly ribosomes too.  Of course, you can get around 0.4 with ribosomes using just a single convolution layer in which you load the weights from the results of linear regression.  The others...even viruses, were frustratingly difficult though.  Best score I was able to come away with was 0.221 before it became apparent it wasn't going to happen.",
      "votes": 1
    },
    {
      "id": 3055529,
      "postDate": "2024-11-25T21:05:05.063Z",
      "content": "<p>try it out</p>",
      "rawMarkdown": "try it out",
      "votes": 2
    },
    {
      "id": 3053772,
      "postDate": "2024-11-23T21:57:39.500Z",
      "content": "<p>let's use a model where we give relu to each layer and softmax to the last layer, why are we making things so complicated?</p>",
      "rawMarkdown": "let's use a model where we give relu to each layer and softmax to the last layer, why are we making things so complicated?"
    },
    {
      "id": 3056555,
      "postDate": "2024-11-27T05:17:04.987Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 3060052,
      "author_name": "David List",
      "author_url": "",
      "post_date": "2024-12-01T11:30:51.497000",
      "content": "<p>I tried it.  Learned a lot.  Abandoned it.  You can actually get really good results for apo-ferritin, and possibly ribosomes too.  Of course, you can get around 0.4 with ribosomes using just a single convolution layer in which you load the weights from the results of linear regression.  The others…even viruses, were frustratingly difficult though.  Best score I was able to come away with was 0.221 before it became apparent it wasn't going to happen.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 3055529,
      "author_name": "mkieffer",
      "author_url": "",
      "post_date": "2024-11-25T21:05:05.063000",
      "content": "<p>try it out</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 3056555,
      "author_name": "",
      "author_url": "",
      "post_date": "2024-11-27T05:17:04.987000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3060052": "I tried it.  Learned a lot.  Abandoned it.  You can actually get really good results for apo-ferritin, and possibly ribosomes too.  Of course, you can get around 0.4 with ribosomes using just a single convolution layer in which you load the weights from the results of linear regression.  The others...even viruses, were frustratingly difficult though.  Best score I was able to come away with was 0.221 before it became apparent it wasn't going to happen.",
    "3055529": "try it out",
    "3053772": "let's use a model where we give relu to each layer and softmax to the last layer, why are we making things so complicated?",
    "3056555": ""
  }
}