[Qgis-user] Principal Component Analysis ?

Fri Jun 23 13:13:53 PDT 2017

Hi,

thanks but nope:

CANOCO comes from the 1970s and is widely outdated by now,
and not achieving (much).
Just a few botanists (=not trained statisticians or coders) still use it,
if at all.
It usually asks the wrong questions and
uses pretty old methods that are widely improved by now,
e.g. machine learning. It all sits there instead, inference from
predictions. Many publications on that matter.
Sorry (any programmer knows it that machine learning performs 'very high').
Like with LMs, GLMs and AIC, p-values or Bayesian stuff, CANOCO does not
generalize nor predict so well.

R is free, a hodgepodge of sorts, and thus, the cutting edge stuff does NOT
all sit in R (high quality swords have a price, sorry).
It's reality. Google etc are not programmed in R or with R for good reasons.
Perhaps a certain blend is helpful, agreed, but hardly anything else.
R packages also tend to suffer from poor conceptual reviews, e.g. what for
and why ?

Again,
please be aware about such things.
Just because it's statistics, coded by your good buddy, freeware, R (or
QGIS), it's not all
automatically good; often the opposite is the case.

It's a fact and the crux you deal with here.
A so-called hijacking by some coders and their mindsets; we saw it many
times before,
and it's self-serving.
CANOCO is a great example for that.

Kind regards
   Falk

Falk Huettmann PhD, Associate Professor
Uni of Alaska Fairbanks UAF

On Fri, Jun 23, 2017 at 11:56 AM, Nicolas Cadieux <
nicolas.cadieux at archeotec.ca> wrote:

> Hi,
>
> I am not at what needs to be analyzed in this case but yes, canonical
> analysis needs to be well understood before being used.
>
> You can go to Pierre Legendre and Louis Legendre, Numerical Ecology, for
> help.  P Legendre also has a book on Spatial analysis in R that can be
> found.
>
> Cheers
> Nicolas
>
> P.S.
>
> As far as statistical software, R is unbeatable for cutting edge sciences.
> It is scary because it's a statistical language before being a statistical
> software.  But I agree, some statistics are implement in GIS software
> without any thought.
>
> Le 23 juin 2017 à 15:11, Falk Huettmann <fhuettmann at alaska.edu> a écrit :
>
> Hi there,
>
> in my view and if I may comment here:
> while R is scary (as stated below),  so are probably many of the wrapped R
> packages really,
> PCAs are even more scary, and their underlying mindsets,
> and the real horror starts when such things get
> implemented into 'homebrews' and such tools, or
> into QGIS and just as point and klick.
>
> As a rule, I would highly suggest to stay clear of such efforts and
> concepts.
> It's no good science, and not much defendable. Unless well thought out,
> it harms the product and reputation you want to create.
> (for a bad example see Krigging in some major commercial GIS...
>  We have seen such things many places).
>
> There should first be a good debate about statistics, if pursued as tools
> in QGIS etc.
> Arguably, unless it's predictive (for inference etc) it has no much value
> these days anymore.
>
> Agreed,
> it's a tragedy of our time that such old things still happen, all the time,
> and in the sciences.
> But it ain't no good, really, and is no good progress. My word for it.
>
> (I also agree that it is good to keep avenues open for such
> work, as an option and for people who know what they are doing and why;
> sure.
> But not more than that. Many bad examples exist all over for decades)
>
> Feel free to follow up as needed; kind regards
>    Falk Huettmann
>      Uni of Alaska Fairbanks
>
>
> On Fri, Jun 23, 2017 at 10:57 AM, Nicolas Cadieux <
> nicolas.cadieux at archeotec.ca> wrote:
>
>> Hi,
>> You can use R for PCA.
>> Use the R Commander package to add a GUI.  R can be scary at first.
>> Nicolas
>>
>> Le 23 juin 2017 à 11:03, image [via OSGeo.org] <[hidden email]
>> <http:///user/SendEmail.jtp?type=node&node=5325361&i=0>> a écrit :
>>
>> Good afternoon,
>>
>> i'm working on windows with several opensource tools (qgis, otb,
>> grass...),
>>
>> I generated several OTB texture indices. Now, i want to evaluate the
>> information 's redundancy thanks to a Principal Component Analysis.
>>
>> => Is it possible to do that with some opensource tools (otb? qgis?
>> grass?) Moreover, I would like set a mask AND set a spectral subset (in
>> order to ignore some indices bands which seems "strange" and not relevant
>> to introduce into the process).
>>
>> Could you throw light for me?
>>
>> In advance, thank you very much for your help.
>>
>> Kind regards.
>>
>> ------------------------------
>> If you reply to this email, your message will be added to the discussion
>> below:
>> http://osgeo-org.1560.x6.nabble.com/Principal-Component-
>> Analysis-tp5325338.html
>> To start a new topic under QGIS - User, email [hidden email]
>> <http:///user/SendEmail.jtp?type=node&node=5325361&i=1>
>> To unsubscribe from QGIS - User, click here.
>> NAML
>> <http://osgeo-org.1560.x6.nabble.com/template/NamlServlet.jtp?macro=macro_viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces.BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web.template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3Aemail.naml>
>>
>>
>> ------------------------------
>> View this message in context: Re: Principal Component Analysis ?
>> <http://osgeo-org.1560.x6.nabble.com/Principal-Component-Analysis-tp5325338p5325361.html>
>> Sent from the QGIS - User mailing list archive
>> <http://osgeo-org.1560.x6.nabble.com/QGIS-User-f4125267.html> at
>> Nabble.com.
>>
>> _______________________________________________
>> Qgis-user mailing list
>> Qgis-user at lists.osgeo.org
>> List info: https://lists.osgeo.org/mailman/listinfo/qgis-user
>> Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-user
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/qgis-user/attachments/20170623/58197d80/attachment.html>