{
  "id": 293433,
  "title": "AutoML resources",
  "url": "/competitions/cs5785-fall-2021-final/discussion/293433",
  "author_name": "Tyce Herrman",
  "post_date": "2021-12-05T19:09:00.832000",
  "votes": 6,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Compatriarts: Here are some resources on AutoML tools if you want to try them out. tl;dr auto-sklearn only works on Linux, so for cross-platform options, it seems like the two best bets are TPOT and Hyperopt sklearn <a href=\"https://towardsdatascience.com/automl-in-python-a-comparison-between-hyperopt-sklearn-and-tpot-8c12aaf7e829\" target=\"_blank\">https://towardsdatascience.com/automl-in-python-a-comparison-between-hyperopt-sklearn-and-tpot-8c12aaf7e829</a>. second tl;dr </p>\n<p>Hyperopt Sklearn</p>\n<p>Pros<br>\n    easy to install<br>\n    fast<br>\n    the best model reaches high performance</p>\n<p>Cons<br>\n    different best models for the same (simple) dataset<br>\n    simple and poor documentation</p>\n<p>TPOT</p>\n<p>Pros<br>\n    always the same best model for the same (simple) dataset<br>\n    well written documentation<br>\n    the best model reaches good performance</p>\n<p>Cons<br>\n    very slow<br>\n    difficult to install</p>",
  "messages": [
    {
      "id": 1607258,
      "postDate": "2021-12-05T19:09:00.833Z",
      "content": "<p>Compatriarts: Here are some resources on AutoML tools if you want to try them out. tl;dr auto-sklearn only works on Linux, so for cross-platform options, it seems like the two best bets are TPOT and Hyperopt sklearn <a href=\"https://towardsdatascience.com/automl-in-python-a-comparison-between-hyperopt-sklearn-and-tpot-8c12aaf7e829\" target=\"_blank\">https://towardsdatascience.com/automl-in-python-a-comparison-between-hyperopt-sklearn-and-tpot-8c12aaf7e829</a>. second tl;dr </p>\n<p>Hyperopt Sklearn</p>\n<p>Pros<br>\n    easy to install<br>\n    fast<br>\n    the best model reaches high performance</p>\n<p>Cons<br>\n    different best models for the same (simple) dataset<br>\n    simple and poor documentation</p>\n<p>TPOT</p>\n<p>Pros<br>\n    always the same best model for the same (simple) dataset<br>\n    well written documentation<br>\n    the best model reaches good performance</p>\n<p>Cons<br>\n    very slow<br>\n    difficult to install</p>",
      "rawMarkdown": "Compatriarts: Here are some resources on AutoML tools if you want to try them out. tl;dr auto-sklearn only works on Linux, so for cross-platform options, it seems like the two best bets are TPOT and Hyperopt sklearn https://towardsdatascience.com/automl-in-python-a-comparison-between-hyperopt-sklearn-and-tpot-8c12aaf7e829. second tl;dr \n\nHyperopt Sklearn\n\nPros\n    easy to install\n    fast\n    the best model reaches high performance\n\nCons\n    different best models for the same (simple) dataset\n    simple and poor documentation\n\nTPOT\n\nPros\n    always the same best model for the same (simple) dataset\n    well written documentation\n    the best model reaches good performance\n\nCons\n    very slow\n    difficult to install",
      "votes": 5
    },
    {
      "id": 1615769,
      "postDate": "2021-12-12T16:57:33.627Z",
      "content": "<p>I ran into a problem with getting hyperopt-sklearn working.  In case any others are having the same problem I wanted to share what I did.  This took me a fair amount of time over the last day or so to resolve.  To install the current version of hyperopt-sklearn the documentation  (<a href=\"url\" target=\"_blank\">https://github.com/hyperopt/hyperopt-sklearn</a>) suggests using the following command:</p>\n<pre><code>pip install git+https://github.com/hyperopt/hyperopt-sklearn\n</code></pre>\n<p>When I did this and then tried to build a model I received the following error:</p>\n<p>AttributeError: 'numpy.random.mtrand.RandomState' object has no attribute 'integers'</p>\n<p>After some searching, I found out that Numpy no longer supports the RandomState syntax and instead is using a \"Generator\" object for a similar purpose.</p>\n<p>Looking at the Github repository for this project there were a few comments on this error in the last two weeks.  It took me a while to get to the bottom of this but I was able to find that the current version of hyperopt that gets installed is 0.2.7.  Also, it looks like starting with version 0.2.6, they broke the way they were handling backward compatibility with Numpy.</p>\n<p>The solution that worked for me was to install version 0.2.5 of hyperopt after running the above pip command that caused this issue.  You can do this by running:</p>\n<pre><code>pip install hyperopt==0.2.5\n</code></pre>\n<p>After doing this, hyperopt-sklearn starting working for me.</p>\n<p>I hope this helps anyone who had this issue and was trying to get this to work.</p>",
      "rawMarkdown": "I ran into a problem with getting hyperopt-sklearn working.  In case any others are having the same problem I wanted to share what I did.  This took me a fair amount of time over the last day or so to resolve.  To install the current version of hyperopt-sklearn the documentation  ([https://github.com/hyperopt/hyperopt-sklearn](url)) suggests using the following command:\n\n\tpip install git+https://github.com/hyperopt/hyperopt-sklearn\n\t\nWhen I did this and then tried to build a model I received the following error:\n\nAttributeError: 'numpy.random.mtrand.RandomState' object has no attribute 'integers'\n\nAfter some searching, I found out that Numpy no longer supports the RandomState syntax and instead is using a \"Generator\" object for a similar purpose.\n\nLooking at the Github repository for this project there were a few comments on this error in the last two weeks.  It took me a while to get to the bottom of this but I was able to find that the current version of hyperopt that gets installed is 0.2.7.  Also, it looks like starting with version 0.2.6, they broke the way they were handling backward compatibility with Numpy.\n\nThe solution that worked for me was to install version 0.2.5 of hyperopt after running the above pip command that caused this issue.  You can do this by running:\n\n\tpip install hyperopt==0.2.5\n\nAfter doing this, hyperopt-sklearn starting working for me.\n\nI hope this helps anyone who had this issue and was trying to get this to work.",
      "votes": 1
    },
    {
      "id": 1609050,
      "postDate": "2021-12-06T22:02:53.250Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1615769,
      "author_name": "VinceF99",
      "author_url": "",
      "post_date": "2021-12-12T16:57:33.627000",
      "content": "<p>I ran into a problem with getting hyperopt-sklearn working.  In case any others are having the same problem I wanted to share what I did.  This took me a fair amount of time over the last day or so to resolve.  To install the current version of hyperopt-sklearn the documentation  (<a href=\"url\" target=\"_blank\">https://github.com/hyperopt/hyperopt-sklearn</a>) suggests using the following command:</p>\n<pre><code>pip install git+https://github.com/hyperopt/hyperopt-sklearn\n</code></pre>\n<p>When I did this and then tried to build a model I received the following error:</p>\n<p>AttributeError: 'numpy.random.mtrand.RandomState' object has no attribute 'integers'</p>\n<p>After some searching, I found out that Numpy no longer supports the RandomState syntax and instead is using a \"Generator\" object for a similar purpose.</p>\n<p>Looking at the Github repository for this project there were a few comments on this error in the last two weeks.  It took me a while to get to the bottom of this but I was able to find that the current version of hyperopt that gets installed is 0.2.7.  Also, it looks like starting with version 0.2.6, they broke the way they were handling backward compatibility with Numpy.</p>\n<p>The solution that worked for me was to install version 0.2.5 of hyperopt after running the above pip command that caused this issue.  You can do this by running:</p>\n<pre><code>pip install hyperopt==0.2.5\n</code></pre>\n<p>After doing this, hyperopt-sklearn starting working for me.</p>\n<p>I hope this helps anyone who had this issue and was trying to get this to work.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1609050,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-12-06T22:02:53.250000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1607258": "Compatriarts: Here are some resources on AutoML tools if you want to try them out. tl;dr auto-sklearn only works on Linux, so for cross-platform options, it seems like the two best bets are TPOT and Hyperopt sklearn https://towardsdatascience.com/automl-in-python-a-comparison-between-hyperopt-sklearn-and-tpot-8c12aaf7e829. second tl;dr \n\nHyperopt Sklearn\n\nPros\n    easy to install\n    fast\n    the best model reaches high performance\n\nCons\n    different best models for the same (simple) dataset\n    simple and poor documentation\n\nTPOT\n\nPros\n    always the same best model for the same (simple) dataset\n    well written documentation\n    the best model reaches good performance\n\nCons\n    very slow\n    difficult to install",
    "1615769": "I ran into a problem with getting hyperopt-sklearn working.  In case any others are having the same problem I wanted to share what I did.  This took me a fair amount of time over the last day or so to resolve.  To install the current version of hyperopt-sklearn the documentation  ([https://github.com/hyperopt/hyperopt-sklearn](url)) suggests using the following command:\n\n\tpip install git+https://github.com/hyperopt/hyperopt-sklearn\n\t\nWhen I did this and then tried to build a model I received the following error:\n\nAttributeError: 'numpy.random.mtrand.RandomState' object has no attribute 'integers'\n\nAfter some searching, I found out that Numpy no longer supports the RandomState syntax and instead is using a \"Generator\" object for a similar purpose.\n\nLooking at the Github repository for this project there were a few comments on this error in the last two weeks.  It took me a while to get to the bottom of this but I was able to find that the current version of hyperopt that gets installed is 0.2.7.  Also, it looks like starting with version 0.2.6, they broke the way they were handling backward compatibility with Numpy.\n\nThe solution that worked for me was to install version 0.2.5 of hyperopt after running the above pip command that caused this issue.  You can do this by running:\n\n\tpip install hyperopt==0.2.5\n\nAfter doing this, hyperopt-sklearn starting working for me.\n\nI hope this helps anyone who had this issue and was trying to get this to work.",
    "1609050": ""
  }
}