{
  "id": 24712,
  "title": "Feature Correlation Matrix / PCA Analysis on numerical Data",
  "url": "/competitions/bosch-production-line-performance/discussion/24712",
  "author_name": "",
  "post_date": "2016-10-24T20:39:04.017Z",
  "votes": 1,
  "comment_count": 2,
  "views": 869,
  "content": "",
  "messages": [
    {
      "id": "141033",
      "postDate": "10/24/2016 20:39:04",
      "content": "",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "141625",
      "postDate": "10/27/2016 15:44:13",
      "content": "<p>Correlation heat map is difficult  to read. It appears as though that highly correlated variables appear in a pattern symmetric to the diagonal. But that is not surprising ..as if a is related to b it will also show b is related to a.  But even in one half of diagonal, the correlation s appear as blocks. That is interesting. It may be reflecting time Association.\nCan you provide your understanding or interpretation of it?</p>\n\n<p>Also, how did you achieve pca ? As far as I know pca or ica will not work with NA values in data? Did you ignore Nas or replaced them with a constant?</p>\n\n<p>Good thought process!</p>",
      "rawMarkdown": "Correlation heat map is difficult  to read. It appears as though that highly correlated variables appear in a pattern symmetric to the diagonal. But that is not surprising ..as if a is related to b it will also show b is related to a.  But even in one half of diagonal, the correlation s appear as blocks. That is interesting. It may be reflecting time Association.\r\nCan you provide your understanding or interpretation of it?\r\n\r\nAlso, how did you achieve pca ? As far as I know pca or ica will not work with NA values in data? Did you ignore Nas or replaced them with a constant?\r\n\r\nGood thought process!",
      "votes": null
    },
    {
      "id": "142225",
      "postDate": "10/31/2016 21:10:22",
      "content": "<p>I replace the Na with 0's to a achieve PCA. \nI am know using xgboost <code>feature_importances_</code> to reduce the feature space.</p>",
      "rawMarkdown": "I replace the Na with 0's to a achieve PCA. \r\nI am know using xgboost `feature_importances_` to reduce the feature space.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 141625,
      "author_name": "sanjay1970",
      "author_url": "",
      "post_date": "10/27/2016 15:44:13",
      "content": "<p>Correlation heat map is difficult  to read. It appears as though that highly correlated variables appear in a pattern symmetric to the diagonal. But that is not surprising ..as if a is related to b it will also show b is related to a.  But even in one half of diagonal, the correlation s appear as blocks. That is interesting. It may be reflecting time Association.\nCan you provide your understanding or interpretation of it?</p>\n\n<p>Also, how did you achieve pca ? As far as I know pca or ica will not work with NA values in data? Did you ignore Nas or replaced them with a constant?</p>\n\n<p>Good thought process!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 142225,
      "author_name": "locojay",
      "author_url": "",
      "post_date": "10/31/2016 21:10:22",
      "content": "<p>I replace the Na with 0's to a achieve PCA. \nI am know using xgboost <code>feature_importances_</code> to reduce the feature space.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "141033": "",
    "141625": "Correlation heat map is difficult  to read. It appears as though that highly correlated variables appear in a pattern symmetric to the diagonal. But that is not surprising ..as if a is related to b it will also show b is related to a.  But even in one half of diagonal, the correlation s appear as blocks. That is interesting. It may be reflecting time Association.\r\nCan you provide your understanding or interpretation of it?\r\n\r\nAlso, how did you achieve pca ? As far as I know pca or ica will not work with NA values in data? Did you ignore Nas or replaced them with a constant?\r\n\r\nGood thought process!",
    "142225": "I replace the Na with 0's to a achieve PCA. \r\nI am know using xgboost `feature_importances_` to reduce the feature space."
  },
  "source": "meta"
}