GPD?

[ivicajan]

Nice tool I'm working with extreme values for sea level and would like to compare with my genpareto.fit... It would bi cool if you could add GPD as well, then more/less you have everything.

thanks! Ivica

[kikocorreoso]

Hi,

Yes, it would be cool. Dev has staged and I would like to retake it but I don't know when it will be possible.

If you find the time to make a PR it would be great 😄

[ivicajan]

Hi mate,I have it working with scipy and genpareto to deal with POT.But it is not that good as your gev... Recently just read  a few papers and installed R ismev gev which is quite powerful but unfortunately not python. CheersIvica P.S. I  can send you my example for case Fremantle gauge station sea level usung POT and gpd (genpareto scipy) -------- Original message --------From: kikocorreoso notifications@github.com Date: 23/9/17 5:23 pm (GMT+08:00) To: kikocorreoso/scikit-extremes scikit-extremes@noreply.github.com Cc: Ivica Janekovic ivica.jan@gmail.com, Author author@noreply.github.com Subject: Re: [kikocorreoso/scikit-extremes] GPD? (#1) Hi, Yes, it would be cool. Dev has staged and I would like to retake it but I don't know when it will be possible. If you find the time to make a PR it would be great 😄

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

{"api_version":"1.0","publisher":{"api_key":"05dde50f1d1a384dd78767c55493e4bb","name":"GitHub"},"entity":{"external_key":"github/kikocorreoso/scikit-extremes","title":"kikocorreoso/scikit-extremes","subtitle":"GitHub repository","main_image_url":"https://cloud.githubusercontent.com/assets/143418/17495839/a5054eac-5d88-11e6-95fc-7290892c7bb5.png","avatar_image_url":"https://cloud.githubusercontent.com/assets/143418/15842166/7c72db34-2c0b-11e6-9aed-b52498112777.png","action":{"name":"Open in GitHub","url":"https://github.com/kikocorreoso/scikit-extremes"}},"updates":{"snippets":[{"icon":"PERSON","message":"@kikocorreoso in #1: Hi,\r\n\r\nYes, it would be cool. Dev has staged and I would like to retake it but I don't know when it will be possible.\r\n\r\nIf you find the time to make a PR it would be great 😄 "}],"action":{"name":"View Issue","url":"https://github.com/kikocorreoso/scikit-extremes/issues/1#issuecomment-331622345"}}}

[kikocorreoso]

I've used extRemes in the past. It is robust and powerful. Check it out. But as you said. it is not python.

My idea is to provide an easy API to do the most basic univariate extremes but I need to find time to include GPD.

[zizzipupp]

Hi, I am not able to install this package. I have tried pip install scikit-extremes but it says that it "Could not find a version that satisfies the requirement scikit-extremes - No matching distribution found for scikit-extremes".

What do you suggest?

[kikocorreoso]

@zizzipupp please, comment here #2

[harshagrao]

Hi, when is the GPD expected to arrive? Currently using extRemes in R but want to switch to python, looking for GPD equivalents :)

[kikocorreoso]

@harshagrao There is not expected date.

Maybe this summer I will find some time to work on this but there is no guarantee. If you find some time you can send a PR to include GPD in the library.

[kartikeya13]

@kikocorreoso Hello

Will your GPD have all those plots which you had in gev(QQ plot, pp plot, return plot, density)?

[kartikeya13]

@ivicajan @harshagrao Hello, Did you guys get gpd for pot python ?

[kartikeya13]

@ivicajan Can you please help me in knowing how can i fit GPD?

[kikocorreoso]

This PR https://github.com/scipy/scipy/pull/9513 could help in the implementation of GPD but it is delayed to review for Scipy v1.4,...

I will have a look to try to implement the relevant funcs in the skextreme's utils module in the meantime if I find some bandwidth.

[cosstas]

@kikocorreoso Hello, scipy.signal.find_peaks is a good starting point for peak detection for POT analysis. It has parameters like 'distance' that can already perform some basic 'declustering'. PR scipy/scipy#9513 I suppose just adds more elaborate 'declustering' methods. POT/GPD fitting can be done without it.

[kikocorreoso]

Hi @cosstas

Thanks for the link.

I've analysed that func from scipy but it is not suitable to do declustering "correctly" (e.g., "correctly" in EVA as it is defined in Coles (2001)).

In my message above I've linked new funcs that could be included in scipy in the future. I will check them to see if them can be included directly in skextremes and when they are included in scipy refactor to use them from scipy.

[cosstas]

@kikocorreoso I was just trying to streamline your bandwidth allocation. :) Any built-in peak detection/declustering functionality would be welcome of course, but I think its importance is secondary to the actual Analysis implementation. In my opinion, POT detection, although necessary, should be independent of the POT analysis because it greatly depends on the estimated quantity. For example POT in sea-wave height timeseries is more complicated than rainfall or temperature.

[kikocorreoso]

@cosstas for me would be paramount to provide functionality to choose the correct data, a declustering function, 'correct' threshold selection, mean residual life plot,... I think all this should be included in the package. Until I do not have this i prefer to delay the inclusion of a GPD fit function because for me it is not independent the POT detection from the GPD fit. The POT detection drives the final GPD fit and I prefer to provide tools to help the user on that 😉

In the meantime, of course, all PRs are 🆗 👌 👍