{
  "id": 12591,
  "title": "Feature engineering and file processing starter code in R",
  "url": "/competitions/diabetic-retinopathy-detection/discussion/12591",
  "author_name": "Scrooge",
  "post_date": "2015-02-23T17:01:42.250000",
  "votes": 21,
  "comment_count": 0,
  "views": 2673,
  "content": "<p>This is fairly simple code written to read the two libraries of images (train and test). I have identified a hook in the <strong>ImageStats</strong> function so you can insert your own clever logic to extract features, enhance edges, adjust contrast, resize and subsample images, etc. to prepare for Random Forests or CNN.</p>\n<p><br>It took about 21 hours to process the libraries even without transformations. Change the sample size on line 70 to speed your own experiments and check <strong>system.time</strong> before processing the entire library.</p>\n<p>It is interesting to see how the images fall into 32 classes depending on image resolution. Some classes have a mixture of inverted and upright images and I found at least one example where the left and right images are different resolutions.</p>\n<p>Cheers, Alastair</p>",
  "messages": [
    {
      "id": 64770,
      "postDate": "2015-02-23T17:01:42.250Z",
      "content": "<p>This is fairly simple code written to read the two libraries of images (train and test). I have identified a hook in the <strong>ImageStats</strong> function so you can insert your own clever logic to extract features, enhance edges, adjust contrast, resize and subsample images, etc. to prepare for Random Forests or CNN.</p>\n<p><br>It took about 21 hours to process the libraries even without transformations. Change the sample size on line 70 to speed your own experiments and check <strong>system.time</strong> before processing the entire library.</p>\n<p>It is interesting to see how the images fall into 32 classes depending on image resolution. Some classes have a mixture of inverted and upright images and I found at least one example where the left and right images are different resolutions.</p>\n<p>Cheers, Alastair</p>",
      "votes": 21
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "64770": ""
  }
}