{
  "id": 185173,
  "title": "Autoencoder's features in tabular model",
  "url": "/competitions/osic-pulmonary-fibrosis-progression/discussion/185173",
  "author_name": "",
  "post_date": "2020-09-19T17:29:17.511303300Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello. I've just trained VAE on CT-scans and it actually works quite well but when I added latent features to my tabular model it kinda got destroyed.</p>\n<p>Without the latent features my tabular model gives about -6.83 score on the LB but with the latent features the score is -24.7981. </p>\n<p>Just for the sake of interest, I replaced every latent feature with random values using <code>np.random.rand()</code> and got -6.8883 on the LB.</p>\n<p>Any ideas what might be the reason? </p>",
  "messages": [
    {
      "id": "1018451",
      "postDate": "09/19/2020 17:29:17",
      "content": "<p>Hello. I've just trained VAE on CT-scans and it actually works quite well but when I added latent features to my tabular model it kinda got destroyed.</p>\n<p>Without the latent features my tabular model gives about -6.83 score on the LB but with the latent features the score is -24.7981. </p>\n<p>Just for the sake of interest, I replaced every latent feature with random values using <code>np.random.rand()</code> and got -6.8883 on the LB.</p>\n<p>Any ideas what might be the reason? </p>",
      "rawMarkdown": "Hello. I've just trained VAE on CT-scans and it actually works quite well but when I added latent features to my tabular model it kinda got destroyed.\n\nWithout the latent features my tabular model gives about -6.83 score on the LB but with the latent features the score is -24.7981. \n\nJust for the sake of interest, I replaced every latent feature with random values using `np.random.rand()` and got -6.8883 on the LB.\n\nAny ideas what might be the reason?",
      "votes": null
    },
    {
      "id": "1019518",
      "postDate": "09/20/2020 13:56:02",
      "content": "<p>-24.7981? It 's not a normal score. May be FVC or Confidence have problem.</p>",
      "rawMarkdown": "24.7981? It 's not a normal score. May be FVC or Confidence have problem.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1019518,
      "author_name": "gzl0506",
      "author_url": "",
      "post_date": "09/20/2020 13:56:02",
      "content": "<p>-24.7981? It 's not a normal score. May be FVC or Confidence have problem.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1018451": "Hello. I've just trained VAE on CT-scans and it actually works quite well but when I added latent features to my tabular model it kinda got destroyed.\n\nWithout the latent features my tabular model gives about -6.83 score on the LB but with the latent features the score is -24.7981. \n\nJust for the sake of interest, I replaced every latent feature with random values using `np.random.rand()` and got -6.8883 on the LB.\n\nAny ideas what might be the reason?",
    "1019518": "24.7981? It 's not a normal score. May be FVC or Confidence have problem."
  },
  "source": "meta"
}