{
  "id": 12378,
  "title": "Good next step for manual feature extraction with random forest model",
  "url": "/competitions/datasciencebowl/discussion/12378",
  "author_name": "",
  "post_date": "2015-02-01T23:09:36.440Z",
  "votes": 1,
  "comment_count": 1,
  "views": 2053,
  "content": "<p>I've been bitten by the bug after signing up for a coursera class in January. &nbsp;Context is that I'm new to Python and R with a ME background, learning by doing in this competition. &nbsp;I'm using Python (started with the tutorial) to extract image features, so far have come up with 18 features. &nbsp;I output the feature set to a csv and then import and fit a random forest model in R. &nbsp;Right now I've submitted with a score of ~2.0.</p>\n<p>Any recommendations short of build a CNN where I should focus? Would be happy to post code. &nbsp;Just learning, seems a CNN may prevail, but I'm not there yet.</p>\n<p>Thanks,<br>Scott</p>",
  "messages": [
    {
      "id": "63422",
      "postDate": "02/01/2015 23:09:36",
      "content": "<p>I've been bitten by the bug after signing up for a coursera class in January. &nbsp;Context is that I'm new to Python and R with a ME background, learning by doing in this competition. &nbsp;I'm using Python (started with the tutorial) to extract image features, so far have come up with 18 features. &nbsp;I output the feature set to a csv and then import and fit a random forest model in R. &nbsp;Right now I've submitted with a score of ~2.0.</p>\n<p>Any recommendations short of build a CNN where I should focus? Would be happy to post code. &nbsp;Just learning, seems a CNN may prevail, but I'm not there yet.</p>\n<p>Thanks,<br>Scott</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "63433",
      "postDate": "02/02/2015 12:59:15",
      "content": "<p>You could have a look at skimage.feature for some ideas. You can look at the images texture with greycomatrix and greycoprops.</p>\n<p>Or look at the OpenCV library (Python), there is lots of useful tools there for feature engineering. Things like thresholding, contour fitting, contour moments, ellipse fitting.</p>\n<p>Hope that helps.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 63433,
      "author_name": "npetitclerc",
      "author_url": "",
      "post_date": "02/02/2015 12:59:15",
      "content": "<p>You could have a look at skimage.feature for some ideas. You can look at the images texture with greycomatrix and greycoprops.</p>\n<p>Or look at the OpenCV library (Python), there is lots of useful tools there for feature engineering. Things like thresholding, contour fitting, contour moments, ellipse fitting.</p>\n<p>Hope that helps.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "63422": "",
    "63433": ""
  },
  "source": "meta"
}