{
  "id": 4158,
  "title": "time structure",
  "url": "/competitions/cause-effect-pairs/discussion/4158",
  "author_name": "",
  "post_date": "2013-04-01T19:31:38.313Z",
  "votes": null,
  "comment_count": 6,
  "views": 2268,
  "content": "<p>Is there are time structure in the data? Is all data cross-sectional or time series? It would be great to have time t for time series data.</p>\r\n<p>How were categories encoded?</p>",
  "messages": [
    {
      "id": "21947",
      "postDate": "04/01/2013 19:31:38",
      "content": "<p>Is there are time structure in the data? Is all data cross-sectional or time series? It would be great to have time t for time series data.</p>\r\n<p>How were categories encoded?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "21948",
      "postDate": "04/01/2013 19:35:55",
      "content": "<p>There is no time structure. The data points are sampled independently.</p>\r\n<p>We plan to have another challenge for time series data.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "21950",
      "postDate": "04/01/2013 19:37:18",
      "content": "<p>Thanks.</p>\r\n<p><span>How were categories encoded?</span></p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "21952",
      "postDate": "04/01/2013 19:48:10",
      "content": "<p>For real data, we used our knowledge of the meaning of the variables. Some variables like age are &quot;exogenous&quot; i.e. they cannot be caused by something else, hence if there is a dependency with another variable, they are necessarily a cause. for semi-artificial\r\n data, we used real variables and mixed them with random functions. hence we know which ones are the &quot;causes&quot;. We also randomly permuted some real variables to create independent variables.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "23029",
      "postDate": "04/18/2013 02:32:33",
      "content": "Hi Isabelle- Can you please explain what you mean by &quot;semi-artificial data&quot;? Are all such data sets classified as Target = 0 ?",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "23036",
      "postDate": "04/18/2013 03:55:37",
      "content": "<p>Maybe age is easy to pick on, but it seems to me that it is not that easy to use judgment to determine what is exogenous. For example, if you are comparing age and health, there could be a survivor bias in the sample where only healthy people survive to\r\n an old age, and it appears that increasing age increases health (I'm not saying this is the case, just making a point). So maybe there's a difference between identifying what people will intuitively think of as a cause, vs. identifying a cause that will hold\r\n up when you run an experiment (hard to randomize age, though, eh?).</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "23096",
      "postDate": "04/18/2013 19:10:54",
      "content": "<p>There are advantages and disadvantages to using real and artificial data, this is why we are using both types.</p>\r\n<p>For real data, we rely on human judgement of what we know of the &quot;physics&quot; of the system. We may be wrong in some cases. Also, as indicated above, we may suffer from sampling bias problems. Otherwise we could also give experimental data, but if the experiments\r\n follow a plan, the distribution of the input is ussualy a give away.&nbsp;</p>\r\n<p>For artificial data, we know the &quot;truth&quot; of the causal relationship for sure, but we don not know whether it is identifiable from the data sample we give, and we may bias the results with the type of data generating model we use.</p>\r\n<p>I do not want to give details about how the artificial data were generated to avoid biasing the results. Semi-artificial means that some real data were used to generate the artificial data in some way to make the end result more realistic.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 21948,
      "author_name": "iguyon",
      "author_url": "",
      "post_date": "04/01/2013 19:35:55",
      "content": "<p>There is no time structure. The data points are sampled independently.</p>\r\n<p>We plan to have another challenge for time series data.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 21950,
      "author_name": "dirknbr",
      "author_url": "",
      "post_date": "04/01/2013 19:37:18",
      "content": "<p>Thanks.</p>\r\n<p><span>How were categories encoded?</span></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 21952,
      "author_name": "iguyon",
      "author_url": "",
      "post_date": "04/01/2013 19:48:10",
      "content": "<p>For real data, we used our knowledge of the meaning of the variables. Some variables like age are &quot;exogenous&quot; i.e. they cannot be caused by something else, hence if there is a dependency with another variable, they are necessarily a cause. for semi-artificial\r\n data, we used real variables and mixed them with random functions. hence we know which ones are the &quot;causes&quot;. We also randomly permuted some real variables to create independent variables.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 23029,
      "author_name": "feloniouspunk",
      "author_url": "",
      "post_date": "04/18/2013 02:32:33",
      "content": "Hi Isabelle- Can you please explain what you mean by &quot;semi-artificial data&quot;? Are all such data sets classified as Target = 0 ?",
      "votes": null,
      "replies": []
    },
    {
      "id": 23036,
      "author_name": "n8fo96082",
      "author_url": "",
      "post_date": "04/18/2013 03:55:37",
      "content": "<p>Maybe age is easy to pick on, but it seems to me that it is not that easy to use judgment to determine what is exogenous. For example, if you are comparing age and health, there could be a survivor bias in the sample where only healthy people survive to\r\n an old age, and it appears that increasing age increases health (I'm not saying this is the case, just making a point). So maybe there's a difference between identifying what people will intuitively think of as a cause, vs. identifying a cause that will hold\r\n up when you run an experiment (hard to randomize age, though, eh?).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 23096,
      "author_name": "iguyon",
      "author_url": "",
      "post_date": "04/18/2013 19:10:54",
      "content": "<p>There are advantages and disadvantages to using real and artificial data, this is why we are using both types.</p>\r\n<p>For real data, we rely on human judgement of what we know of the &quot;physics&quot; of the system. We may be wrong in some cases. Also, as indicated above, we may suffer from sampling bias problems. Otherwise we could also give experimental data, but if the experiments\r\n follow a plan, the distribution of the input is ussualy a give away.&nbsp;</p>\r\n<p>For artificial data, we know the &quot;truth&quot; of the causal relationship for sure, but we don not know whether it is identifiable from the data sample we give, and we may bias the results with the type of data generating model we use.</p>\r\n<p>I do not want to give details about how the artificial data were generated to avoid biasing the results. Semi-artificial means that some real data were used to generate the artificial data in some way to make the end result more realistic.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "21947": "",
    "21948": "",
    "21950": "",
    "21952": "",
    "23029": "",
    "23036": "",
    "23096": ""
  },
  "source": "meta"
}