Submitted by xuefeimi on Wed, 08/11/2010 - 07:44
Posted in
Dear Anup and Kyle,
Thank you two for the advises of the PCA model.
Now it is summer break in Germany, and I have some time to play with Red-R and Python.
As we discussed in the UseR conference, I would like to implement R package mvtnorm, which calculates the multivariate normal and t distribution and is one of the "most important" R-packages, into Red-R.
The user interface could be quite simple, because most of the user will integrate it into their own scheme.
step1
read corr matrix -> set integral boundry and set method (Brez or Miwa)--> call --> return result.
I am going to use the R execute to test it today. i am coauthor of mvtnorm package, so any bugs from R side can be easily traced.
step2
there are also some other two packages which do different methods.
The authors have discussed via email, but reached no plan to integrate all the methods.
maybe Red-R can achieve this.
So i will tell you want i will get
Best Regards
Xuefei Mi
Hi Xuefei,
Glad you could get it working.
One of our major problems is that rpy is only compiled for some of the R distributions. We are currently using the latest rpy version which works with R2.9.1
We just have to wait for rpy to make a new version or try to compile ourselves. Building rpy by ourselves has run into several problems with compilation when I tried it. Anup is working now on it and we want help from the rpy community. We want to stick with rpy because rpy2 uses a different conversion from R to python which would require a major rewrite of the software just to get us back to where we are now with rpy.
Thanks
Kyle
hi, it is week end guys.
how are you?
this week i have some free time. so i learn more about red-r and make my first widget.
it is the rnorm() function in the mvtnorm package.
here is some problems and questions.
1
i follow the instruction here https://red-r.org/base-widget-maker
in here i have problem
"When you are done click “Launch Widget”. The widget will be compiled and loaded into Red-R. You could use it right away if you wanted. Your widgets file is located in the Red-R\libraries\blank\widgets directory in wherever you installed Red-R (usually Program Files). "
my solution:
" the widgets in libraries\blank\widgets directory have to be copied into libraries\base\widgets directory. And 'launch Widget' have to be pressed again to create the widgets"
Anup told me the trick before, so i remember that.
2
problem https://red-r.org/base-widget-maker
"We haven’t automated this procedure yet so just zip all the files together and e-mail them to or and we can get them onto the website for you. "
solution,
so i have to mail e.g. all the (.py) files or ?
3
problem:
i have created my own rnorm.rrs, which only have 2 wights, but after i save it, 7 wights come out. i have just make a screen shoot and put all the stuffs in a .zip file.
maybe you could look at it.
4
https://red-r.org/development/anatomy-of-a-widget
it is a nice explain for what incide the .py file.
i would prefer to change all the 't' function in the page into cor, then it is matched with the cor.py in other pages. also add line number is a good idea, because the code is little bitte long, some times i missed them.
ptw, have you think about publication? i think the help documents can be a very good publication in R journal or J. of Stat.Software. and by doing so, someone will check you to make perfect documents.
Best Regards
Xuefei Mi
hi, it is week end guys.
how are you?
this week i have some free time. so i learn more about red-r and make my first widget.
it is the rnorm() function in the mvtnorm package.
here is some problems and questions.
1
i follow the instruction here https://red-r.org/base-widget-maker
in here i have problem
"When you are done click “Launch Widget”. The widget will be compiled and loaded into Red-R. You could use it right away if you wanted. Your widgets file is located in the Red-R\libraries\blank\widgets directory in wherever you installed Red-R (usually Program Files). "
my solution:
" the widgets in libraries\blank\widgets directory have to be copied into libraries\base\widgets directory. And 'launch Widget' have to be pressed again to create the widgets"
Anup told me the trick before, so i remember that.
2
problem https://red-r.org/base-widget-maker
"We haven’t automated this procedure yet so just zip all the files together and e-mail them to or and we can get them onto the website for you. "
solution,
so i have to mail e.g. all the (.py) files or ?
3
problem:
i have created my own rnorm.rrs, which only have 2 wights, but after i save it, 7 wights come out. i have just make a screen shoot and put all the stuffs in a .zip file.
maybe you could look at it.
4
https://red-r.org/development/anatomy-of-a-widget
it is a nice explain for what incide the .py file.
i would prefer to change all the 't' function in the page into cor, then it is matched with the cor.py in other pages. also add line number is a good idea, because the code is little bitte long, some times i missed them.
ptw, have you think about publication? i think the help documents can be a very good publication in R journal or J. of Stat.Software. and by doing so, someone will check you to make perfect documents.
Best Regards
Xuefei Mi
Hi Xuefei,
Red-R does require python 2.6.5 or greater. The installer looks for python 2.6 but doesn't check the subversion number. I'll see if I can fix the installer to look for that and tell the user to install the new one.
The problem with R might stem from the way that we handle R in Red-R. I actually just move R2.9 into the Red-R folder and make no R changes to any environment variables. We did this so that installing R2.9 would not conflict if there were later versions of R on the machine.
Anup is working on compiling rpy for later R versions but we were having some problems with compilation last I heard.
Were you able to get Red-R running?
Best
Kyle
Hi Xuefei,
If you have any trouble with the mvtnorm package just let us know. I'm not sure if the call and result should be in the same widget or should just use a summary like widget, but it's your decision.
I'm not sure I understand step 2 and what the problems with integration might be so if you could explain that more I might be better able to address the problems.
When making your widgets you can start with the RVariable signal class which will work as long as you do everything right. Moving to the other signal classes will help the user to make choices and prevent them from doing wrong things (sending a list to a widget that expects a matrix for example).
Good luck and keep us posted.
Kyle
Dear Xuefei,
That sounds great. Take a look at the development documentation at https://red-r.org/development.
Please let us know if there is anything confusing in documentation as you start.
Thanks,
Anup
Business Intelligence (BI) R - package is good one to be added. But I think this is big project. But most of the components is already in RED-R - data widget
Rattle sort of leading this area. Gnome/Gtk :-). I love Qt i.e. PyQt.
http://rattle.togaware.com/
Have a look at these screenshots
http://rattle.togaware.com/rattle-screenshots.html
I love RED-R in Meego tablet later on.
http://decisionstats.wordpress.com/2010/05/29/data-mining-through-the-android/
In the future releases, not now.
Geo Business Intelligence (GeoBI) would be next!
http://www.innovativegis.com/basis/MapAnalysis/Topic28/Topic28.htm
http://www.innovativegis.com/basis/present/GeoTec08/GeoBusiness08.ppt
Some of the tools are already in R packages
http://cran.r-project.org/web/views/Spatial.html
What do you think about these things?
Thanks.
Noli
A prototype pls package has been uploaded into the package repository. Using the package manager you should be able to get it and test it out.
This is a prototype package. For help you should refer to to the R pls package documentation in CRAN. The widgets attempt to emulate the pls package functions with some restrictions.
Any comments on this package would be appreciated to increase usability. These can be made either here or on the pls package forum page. This is a prototype package so there are likely to be many bugs though when I tested it using example data from the pls package I could recreate all of the expected analysis.
Thanks for the posts,
Kyle
I'd love to have the plsr-package available in Red-R, as this is a standard tool in chemometrics.
http://cran.r-project.org/web/packages/pls/index.html
Thanks for the great effort with Red-R. So far it looks great.
Cheers
Oliver
c-DNA microarray.
Good idea. I'll look at the data types and see how we can integrate. Bioconductor likely has packages that we can interface with. Some of the analysis widgets that we have already could also use matrix like data such as differential expression and significance criteria.
I'll look into it and either append to the bioinformatics package or make a cDNA package.
c-DNA Microarray data analysis still uncovered by many analysis software ,i think that you people initiate in that area.b'cos its really need of people.
Thanks
If anyone has ideas for R packages (or even non-R packages) that they would like to see made part of Red-R, please leave a posting here. The more descriptive the better. Tell us how you want the package to work, what it will do, how it could work with the pipeline, whatever you can think of.
There are thousands of R packages and we need help prioritizing.
Thanks
kyle
"Were you able to get Red-R running?"
yes, it works well in my desktop. i am learning to make wights now.
in the labtop it runs, but the RevolutionR crushed.
i will reinstall windows later, so not put much time on what happened on the notebook.
"just move R2.9 into the Red-R folder and make no R changes to any environment "
when i install red-r in my desktop, it ask for Rgraphapp.dll and i copy it from \R\bin to c:\windows\system32 to solve this problem. but i think it is the problem of default working path.
i would suggest you stays to R 2.9.2, which is stable and well spreaded. also some commercial softwere such as Asreml-R is based on this version.
wait half or one year until R2.11.x gets stable and they inplement it.
bests
hi, thanks for the suggests.
for the step two,
there are only two algorithm be supported by mvtnorm (Bretz' quashi-monte-carlo and Miwa's moving avarage), Dunnett method and Auto Regression method are included in other two packages in Cran. And the authors of these three packages haven't reach a agreement to merge the packages.
so it would be a good thing, if we could make an UI and the user can call these 3 packages in a same widget.
just like the correlation/covariance widget, on the up left corner, you can choose Bretz, Miwa, AR, Dunnett etc. i am trying to change the code of cor.py to get it now.
>>>>>>>>>>>>>>>>>>>>>>>>>>
for Anup's hit, i am looking at the developments now.
and i have problems again.
Problem 1
it seems that i have some confiliction with python2.6.
i had python installed long time ago (python2.6.2). when i install red-r, it ask me to install python again, so i install everything in the python folder. now the python ide ask for python2.6.2 msi, which i find in python.org.
i would suggest you give some warning message, if the user had a older version of python already installed. the new version must be installed in seperated folder as the older one.
the way to use 2.6.5 rewrite 2.6.2 not work. so i deinstall 2.6.2 and reinstall 2.6.5.
now i can follow the instruction from https://red-r.org/development.
i think red-r requires python >2.6.4 right?
problems2
i gusee there might be some confilictions caused by install and reinstall.
e.g. if you still remember last time that i can not run some widgets in my labtop.
i think that is a confilication for red-r with multiple version installed of R.
now i also doubt if it is a problem with the Revolution-R. i dont know if you have try it.
after i fixed the version confiliction with R. the Revolution-R crushed.
i think we dont have to fix these confilication.
for me, i just say, reinstall windows, people have to do it every three month anyway.
Bests
xuefei