tag:blogger.com,1999:blog-1909863961516003719.post7530972976834381376..comments2024-01-01T05:50:02.704-08:00Comments on joepy: Statistical Power in StatsmodelsUnknownnoreply@blogger.comBlogger4125tag:blogger.com,1999:blog-1909863961516003719.post-17121754248454716622015-10-07T09:25:21.730-07:002015-10-07T09:25:21.730-07:00Hi Jarad and Zoey,
Writing comprehensive documenta...Hi Jarad and Zoey,<br />Writing comprehensive documentation is a lot of work, and I usually stay with a topic just long enough to figure out the statistical background and the computational issues.<br /><br />statsmodels is very open to contributions, preferably as a pull request on github. There is also the possibility to contribute notebooks to or link notebooks from the wiki https://github.com/statsmodels/statsmodels/wiki/Examples . In some popular areas users are writing examples, notebooks and blog posts, but they usually don't end up in statsmodels because they are not submitted for inclusion.<br /><br />(I have neglected this blog for years now, because I'm not patient enough or don't want to allocate enough time to write something "comprehensible".)<br /><br />On the other hand, the documentation has improved considerably over the years, many developers, contributors and users provide documentation and notebook. However, in some areas they are not easy to find, and in some areas code development moves faster than writing documentation. <br /><br />As example, I just wrote an overview of which models can be used depending on the type (continuous, binary, count, ..) of the response variable https://github.com/statsmodels/statsmodels/issues/2642 This should be extended to several pages with examples or link to examples, but I doubt I will take time for this right now, but will write some notebooks for specific topics in the next months.<br /><br />Another recommendation for specific questions is to ask on stackoverflow. The response time is in most cases very good. Helping there takes a limited amount of time, and points out some priorities when "improve statsmodels" is a task that is too large and daunting.Josef Perktoldhttps://www.blogger.com/profile/04810164380416139793noreply@blogger.comtag:blogger.com,1999:blog-1909863961516003719.post-70762559739300419232015-10-07T08:36:10.613-07:002015-10-07T08:36:10.613-07:00I came across this blog because I'm really str...I came across this blog because I'm really struggling finding examples of how to use statsmodels, which functions to use, how to even just get the functions to work. To take statsmodels to the next level, and gain a wider audience, there really SHOULD be a more urgent effort to at least provide EXAMPLES of each function so a user doesn't have to screw around with figuring it out on their own with little help from the documentation.<br /><br />Think about it. If you make the best product but people don't know how to use it and get frustrated, many will just give up and find alternatives. Maybe that's the way statsmodels wants it. If they didn't want it that way, they'd provide clear examples like pandas does. That's why pandas is popular - because people can LEARN how to use it QUICKLY with EXAMPLES.<br /><br />If you have time constraints, think outside the box and let people submit examples to you so you just review and post them. Have users curate the content. Lots of alternative options rather than taking it on all by yourself.Jarad and Zoeyhttps://www.blogger.com/profile/02117492063155215443noreply@blogger.comtag:blogger.com,1999:blog-1909863961516003719.post-77438753623558400402013-03-24T14:29:45.255-07:002013-03-24T14:29:45.255-07:00Thomas, I think your statistics in python introduc...Thomas, I think your statistics in python introduction looks very nice and is very needed to make python more popular in this area.<br /><br />The documentation on the power calculations will go into statsmodels in some form.<br />My main reason for not adding much in terms of introductory information in statsmodels is the time constraint (and personal priorities), I already have a hard time keeping the minimal documentation and docstrings in a usable form.<br /><br />The requirements to improve the statsmodels documentation are not really high. There are many areas that need improvements, and in most cases it is not necessary to understand all the details. And, I would rather review some pull requests for improvements than coming up with the explanations and examples by myself.Josef Perktoldhttps://www.blogger.com/profile/04810164380416139793noreply@blogger.comtag:blogger.com,1999:blog-1909863961516003719.post-63993349796148300022013-03-23T06:25:39.085-07:002013-03-23T06:25:39.085-07:00Joe, I think you are doing a hell of an amount of ...Joe, I think you are doing a hell of an amount of great work. But I think it is really difficult for others to build on it, given the level of documentation in statsmodels. Should not what you have written here go into the docu of statsmodels? I think statsmodels would/could be used a LOT more if the documentation were improved. As you may have seen, I have started some stats-intro (http://work.thaslwanter.at/Stats/html/StatsFH.html), and will in the future try to incorporate introductory documentation for statsmodels. My statistical knowledge is narrow enough that I am hesitant to "mess" around with the stats-models documentation myself.upper mountainhttps://www.blogger.com/profile/13527728270085464454noreply@blogger.com