{
  "id": 421977,
  "title": "Decoding the embeddings",
  "url": "/competitions/cafa-5-protein-function-prediction/discussion/421977",
  "author_name": "",
  "post_date": "2023-07-07T16:33:14.846208300Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Is there any way to turn the embeddings back into sequences? I have a reason for doing this. But if there is a straightforward way to \"decode\" the embeddings, I would be grateful to hear it.</p>",
  "messages": [
    {
      "id": "2334429",
      "postDate": "07/07/2023 16:33:14",
      "content": "<p>Is there any way to turn the embeddings back into sequences? I have a reason for doing this. But if there is a straightforward way to \"decode\" the embeddings, I would be grateful to hear it.</p>",
      "rawMarkdown": "Is there any way to turn the embeddings back into sequences? I have a reason for doing this. But if there is a straightforward way to \"decode\" the embeddings, I would be grateful to hear it.",
      "votes": null
    },
    {
      "id": "2334561",
      "postDate": "07/07/2023 19:05:54",
      "content": "<p>The embeddings are 2-d vectors, they were trained until reach one position in 2d space which is closed to others with close properties, as if that were not enough, there is the seed. Each seed create different random starting vectors, which makes more complicated to return to the previous step.<br>\nThere are \"embeddings\", which are coded vectors, with a dimension(size), which was not trained. This could be returned into sequences easily.</p>",
      "rawMarkdown": "The embeddings are 2-d vectors, they were trained until reach one position in 2d space which is closed to others with close properties, as if that were not enough, there is the seed. Each seed create different random starting vectors, which makes more complicated to return to the previous step.\nThere are \"embeddings\", which are coded vectors, with a dimension(size), which was not trained. This could be returned into sequences easily.",
      "votes": null
    },
    {
      "id": "2334883",
      "postDate": "07/08/2023 05:27:21",
      "content": "<p>I get how and why they are encoded. I'm just looking for a straightforward way to decode them. Anyone know of any code examples?</p>",
      "rawMarkdown": "I get how and why they are encoded. I'm just looking for a straightforward way to decode them. Anyone know of any code examples?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2334561,
      "author_name": "victorfernandezalbor",
      "author_url": "",
      "post_date": "07/07/2023 19:05:54",
      "content": "<p>The embeddings are 2-d vectors, they were trained until reach one position in 2d space which is closed to others with close properties, as if that were not enough, there is the seed. Each seed create different random starting vectors, which makes more complicated to return to the previous step.<br>\nThere are \"embeddings\", which are coded vectors, with a dimension(size), which was not trained. This could be returned into sequences easily.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2334883,
          "author_name": "newtoncampbell",
          "author_url": "",
          "post_date": "07/08/2023 05:27:21",
          "content": "<p>I get how and why they are encoded. I'm just looking for a straightforward way to decode them. Anyone know of any code examples?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2334429": "Is there any way to turn the embeddings back into sequences? I have a reason for doing this. But if there is a straightforward way to \"decode\" the embeddings, I would be grateful to hear it.",
    "2334561": "The embeddings are 2-d vectors, they were trained until reach one position in 2d space which is closed to others with close properties, as if that were not enough, there is the seed. Each seed create different random starting vectors, which makes more complicated to return to the previous step.\nThere are \"embeddings\", which are coded vectors, with a dimension(size), which was not trained. This could be returned into sequences easily.",
    "2334883": "I get how and why they are encoded. I'm just looking for a straightforward way to decode them. Anyone know of any code examples?"
  },
  "source": "meta"
}