<div dir="ltr"><div>Regarding NCL's Extreme Value Statistics function suite:<br><br> <a href="http://www.ncl.ucar.edu/Applications/extreme_value.shtml" target="_blank">http://www.ncl.ucar.edu/<wbr>Applications/extreme_value.<wbr>shtml</a><br><br></div><div>contains the following ... at the very top ...<br></div><br>"NCL has a small number of basic extreme value (EV) and recurrence statistical functions.
However, NCL is not <b>R</b> or <b>S+</b> or <b>Matlab</b> or <b>IDL</b> or <b>Excel</b>
or <b>Python's SciPy</b>. These tools contain many more EV related functions.
That said: <b><span style="color:rgb(0,0,255)">None of these tools are NCL either!</span>"</b><div><div><br>[1] <br> NCL function '<b><span style="color:rgb(0,0,255)">extval_mlegev</span></b>' is an interface to the fortran subroutine:<br><br>C-----------------------------<wbr>------------------------------<wbr>------<br>c Algorithm AS 215<br>c Maximum-Likelihood Estimation of the Parameters of the Generalized Extreme-Value Distribution<br>c J. R. M. Hosking<br>c Journal of the Royal Statistical Society. Series C (Applied Statistics) <br>c Vol. 34, No. 3 (1985), pp. 301-310<br>c Stable URL: <a href="http://www.jstor.org/stable/2347483" target="_blank">http://www.jstor.org/stable/<wbr>2347483</a><br>C-----------------------------<wbr>------------------------------<wbr>------<br><br></div><div>Why the differences between EASYFIT and NCL? I don't know specifically. Usually, there is a first guess at the initial value of the parameters. Then, there is an iteration method (here, steepest-ascent) before some convergence constraints are met.<br><br></div><div>So, perhaps, different convergence criteria? different number of maximum iterations? ...<br><br></div><div>My speculation is that the resulting distribution differences are 'small.' <br><br></div><div>---------<br></div><div>Also, Example 1 (same flood data) at:<br> <a href="http://www.ncl.ucar.edu/Document/Functions/Built-in/extval_mlegev.shtml" target="_blank">http://www.ncl.ucar.edu/<wbr>Document/Functions/Built-in/<wbr>extval_mlegev.shtml</a><br><br></div><div>compares NCL's <span style="color:rgb(0,0,255)"><b>mlegev </b><span style="color:rgb(0,0,0)">with the output from R<br><br></span></span></div><div> NCL<br></div><div><pre> (0) 487.1327 location
(1) 308.7767 scale
(2) 0.0936752 shape<br><br></pre><pre> R<br> [1] 487.07202252 <br> [2] 308.76501616 <br> [3] -0.09363131 <=== sign difference is due to the way <br> the function is defined.</pre><pre><br></pre><pre>The sign difference for the shape is due to the way the function is defined.<br></pre><pre>Example 1 shows the R library used. Maybe there are others?<br></pre><pre><br></pre><pre>This comparison looks pretty good to me!<br><br>===========================================================<br>[2]<br></pre><pre>Currently, NCL's Extreme Value library does not have a function <br>for estimating the parameters of the Pareto distribution. Only, two<br>max-liklihood parameter distribution functions are provided:<br></pre><pre><br><a href="http://www.ncl.ucar.edu/Document/Functions/Ext_val/extval_mlegam.shtml">http://www.ncl.ucar.edu/Document/Functions/Ext_val/extval_mlegam.shtml</a><br><a href="http://www.ncl.ucar.edu/Document/Functions/Ext_val/extval_mlegev.shtml">http://www.ncl.ucar.edu/Document/Functions/Ext_val/extval_mlegev.shtml</a><br><br>=====================================================<br>[3]<br></pre><pre>There is no percentile function (Inverse CDF) in NCL.<br><br>=====================================================<br></pre><pre>Really, supporting an entire Extreme Value library from within <br>NCL is 'kinda' beyond NCL's scope/mission. We have to pick-and-choose.<br><br></pre><pre>Good luck<br></pre><pre>D<br></pre></div></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Mon, Oct 16, 2017 at 8:42 AM, IN <span dir="ltr"><<a href="mailto:qphoton@gmail.com" target="_blank">qphoton@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div><div><div><div>Dear All,<br></div> I am using NCL for Extreme value statistics. I want to fit<br></div> generalized extreme value (GEV) distribution and generalized pareto<br></div>to my data.<br><br></div>I started with example 6 on this page.<br><div><div><br><a href="https://www.ncl.ucar.edu/Applications/extreme_value.shtml" target="_blank">https://www.ncl.ucar.edu/<wbr>Applications/extreme_value.<wbr>shtml</a><br><br>and compared the results with EASYFIT from <a href="http://www.mathwave.com" target="_blank">www.mathwave.com</a> using the river_flow data from the example.<br></div><div><br><div>I used extval_mlegev and got these values for center, scale and shape.</div><div></div><div></div><div> <br></div><div>487.1327 , 308.7766 and 0.09367528</div><div><br></div><div>However from Easyfit, I am getting these values (see attached pictures)<br></div><div><br></div><div>486.88, 322.72 and -0.10936</div><div><br></div><div>So my questions are :</div><div>Why there is difference ? Both NCL and EASYFIT use Maximum-Likelihood</div><div> Estimation for calculation of these paramters. Also I need to calculate these <br></div><div>paramters for generalized pareto distribution. How can I do that with NCL?</div><div><br></div><div>In addition, I have option in EASYFIT to get 95 percentile ( or any other percetile)</div><div>from Inverse CDF (see last two pictures), Can I do similar statistic with NCL?</div><div><br></div><div>Thanks in advance</div><div>Best Regards</div><div>Imran Nadeem<br></div><div>BOKU, Vienna.<br></div><div><br></div><div><br></div><div><br></div><div><br></div><div><br><br></div></div></div></div>
<br>______________________________<wbr>_________________<br>
ncl-talk mailing list<br>
<a href="mailto:ncl-talk@ucar.edu">ncl-talk@ucar.edu</a><br>
List instructions, subscriber options, unsubscribe:<br>
<a href="http://mailman.ucar.edu/mailman/listinfo/ncl-talk" rel="noreferrer" target="_blank">http://mailman.ucar.edu/<wbr>mailman/listinfo/ncl-talk</a><br>
<br></blockquote></div><br></div>