{
  "id": 98418,
  "title": "In regression take care of extrapolation using GP or KRR",
  "url": "/competitions/champs-scalar-coupling/discussion/98418",
  "author_name": "cyberia",
  "post_date": "2019-07-03T16:25:53.756000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>In terms of xkcd graphic humor\n<img src=\"https://imgs.xkcd.com/comics/extrapolating.png\" alt=\"\">\n<img src=\"https://imgs.xkcd.com/comics/sustainable.png\" alt=\"\"></p>\n\n<p><strong>Extrapolation</strong> - Extrapolation is a way to make guesses about the future or about some hypothetical situation based on data that you already know. You’re basically taking your “best guess”.</p>\n\n<p>One solution I came across is Gaussian Processes. \n<img src=\"https://scikit-learn.org/stable/_images/sphx_glr_plot_compare_gpr_krr_001.png\" alt=\"\"></p>\n\n<p>Great talk on GP at NeurIPS 2017 <a href=\"https://www.youtube.com/watch?v=NHTGY8VCinY\">https://www.youtube.com/watch?v=NHTGY8VCinY</a></p>\n\n<p>Rather than fitting one curve or an ensemble of multiple curves, GP is probabilistic in nature and more realistic in terms of real world applications.</p>\n\n<p>Detailed notebook on nbviewer <a href=\"https://nbviewer.jupyter.org/github/krasserm/bayesian-machine-learning/blob/master/gaussian_processes.ipynb?flush_cache=true\">https://nbviewer.jupyter.org/github/krasserm/bayesian-machine-learning/blob/master/gaussian_processes.ipynb?flush_cache=true</a></p>\n\n<p>Best book dedicated to GP <a href=\"http://www.gaussianprocess.org/gpml/\">http://www.gaussianprocess.org/gpml/</a> (Do a google search you may find a free pdf)</p>",
  "messages": [
    {
      "id": 567524,
      "postDate": "2019-07-03T16:25:53.757Z",
      "content": "<p>In terms of xkcd graphic humor\n<img src=\"https://imgs.xkcd.com/comics/extrapolating.png\" alt=\"\">\n<img src=\"https://imgs.xkcd.com/comics/sustainable.png\" alt=\"\"></p>\n\n<p><strong>Extrapolation</strong> - Extrapolation is a way to make guesses about the future or about some hypothetical situation based on data that you already know. You’re basically taking your “best guess”.</p>\n\n<p>One solution I came across is Gaussian Processes. \n<img src=\"https://scikit-learn.org/stable/_images/sphx_glr_plot_compare_gpr_krr_001.png\" alt=\"\"></p>\n\n<p>Great talk on GP at NeurIPS 2017 <a href=\"https://www.youtube.com/watch?v=NHTGY8VCinY\">https://www.youtube.com/watch?v=NHTGY8VCinY</a></p>\n\n<p>Rather than fitting one curve or an ensemble of multiple curves, GP is probabilistic in nature and more realistic in terms of real world applications.</p>\n\n<p>Detailed notebook on nbviewer <a href=\"https://nbviewer.jupyter.org/github/krasserm/bayesian-machine-learning/blob/master/gaussian_processes.ipynb?flush_cache=true\">https://nbviewer.jupyter.org/github/krasserm/bayesian-machine-learning/blob/master/gaussian_processes.ipynb?flush_cache=true</a></p>\n\n<p>Best book dedicated to GP <a href=\"http://www.gaussianprocess.org/gpml/\">http://www.gaussianprocess.org/gpml/</a> (Do a google search you may find a free pdf)</p>",
      "rawMarkdown": "In terms of xkcd graphic humor\n![](https://imgs.xkcd.com/comics/extrapolating.png)\n![](https://imgs.xkcd.com/comics/sustainable.png)\n\n**Extrapolation** - Extrapolation is a way to make guesses about the future or about some hypothetical situation based on data that you already know. You’re basically taking your “best guess”.\n\nOne solution I came across is Gaussian Processes. \n![](https://scikit-learn.org/stable/_images/sphx_glr_plot_compare_gpr_krr_001.png)\n\nGreat talk on GP at NeurIPS 2017 https://www.youtube.com/watch?v=NHTGY8VCinY\n\nRather than fitting one curve or an ensemble of multiple curves, GP is probabilistic in nature and more realistic in terms of real world applications.\n\nDetailed notebook on nbviewer https://nbviewer.jupyter.org/github/krasserm/bayesian-machine-learning/blob/master/gaussian_processes.ipynb?flush_cache=true\n\nBest book dedicated to GP http://www.gaussianprocess.org/gpml/ (Do a google search you may find a free pdf)\n\n",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "567524": "In terms of xkcd graphic humor\n![](https://imgs.xkcd.com/comics/extrapolating.png)\n![](https://imgs.xkcd.com/comics/sustainable.png)\n\n**Extrapolation** - Extrapolation is a way to make guesses about the future or about some hypothetical situation based on data that you already know. You’re basically taking your “best guess”.\n\nOne solution I came across is Gaussian Processes. \n![](https://scikit-learn.org/stable/_images/sphx_glr_plot_compare_gpr_krr_001.png)\n\nGreat talk on GP at NeurIPS 2017 https://www.youtube.com/watch?v=NHTGY8VCinY\n\nRather than fitting one curve or an ensemble of multiple curves, GP is probabilistic in nature and more realistic in terms of real world applications.\n\nDetailed notebook on nbviewer https://nbviewer.jupyter.org/github/krasserm/bayesian-machine-learning/blob/master/gaussian_processes.ipynb?flush_cache=true\n\nBest book dedicated to GP http://www.gaussianprocess.org/gpml/ (Do a google search you may find a free pdf)\n\n"
  }
}