{
  "id": 294637,
  "title": "Feature Engineering",
  "url": "/competitions/cs5785-fall-2021-final/discussion/294637",
  "author_name": "Elliot",
  "post_date": "2021-12-11T19:21:50.313000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I've seen some posts discussing how others are trying to get new features or other additional data to improve performance, so I figured I would share some of what I've done. </p>\n<ul>\n<li>Getting additional resnet data (trying to extract other layers from resnet after tuning it on the data we have)</li>\n<li>Generating captions to get more text data (too inconsistent to use)</li>\n<li>Using alternative vectorization methods to handle our text to see if I could get better performance than w2v (I've had mixed results here)</li>\n</ul>\n<p>If anyone else came up with some cool ways to get more data I would love hear what you've done! I think most of the methods I've tried have been too constrained by my overall approach to the problem, so it would be good to get some additional perspective. </p>",
  "messages": [
    {
      "id": 1615076,
      "postDate": "2021-12-11T19:21:50.313Z",
      "content": "<p>I've seen some posts discussing how others are trying to get new features or other additional data to improve performance, so I figured I would share some of what I've done. </p>\n<ul>\n<li>Getting additional resnet data (trying to extract other layers from resnet after tuning it on the data we have)</li>\n<li>Generating captions to get more text data (too inconsistent to use)</li>\n<li>Using alternative vectorization methods to handle our text to see if I could get better performance than w2v (I've had mixed results here)</li>\n</ul>\n<p>If anyone else came up with some cool ways to get more data I would love hear what you've done! I think most of the methods I've tried have been too constrained by my overall approach to the problem, so it would be good to get some additional perspective. </p>",
      "rawMarkdown": "I've seen some posts discussing how others are trying to get new features or other additional data to improve performance, so I figured I would share some of what I've done. \n\n- Getting additional resnet data (trying to extract other layers from resnet after tuning it on the data we have)\n- Generating captions to get more text data (too inconsistent to use)\n- Using alternative vectorization methods to handle our text to see if I could get better performance than w2v (I've had mixed results here)\n\nIf anyone else came up with some cool ways to get more data I would love hear what you've done! I think most of the methods I've tried have been too constrained by my overall approach to the problem, so it would be good to get some additional perspective. \n"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1615076": "I've seen some posts discussing how others are trying to get new features or other additional data to improve performance, so I figured I would share some of what I've done. \n\n- Getting additional resnet data (trying to extract other layers from resnet after tuning it on the data we have)\n- Generating captions to get more text data (too inconsistent to use)\n- Using alternative vectorization methods to handle our text to see if I could get better performance than w2v (I've had mixed results here)\n\nIf anyone else came up with some cool ways to get more data I would love hear what you've done! I think most of the methods I've tried have been too constrained by my overall approach to the problem, so it would be good to get some additional perspective. \n"
  }
}