Setting PYTHONPATH for CosmoMC under Windows

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Andrey Egorov
Posts: 4
Joined: June 08 2015
Affiliation: University of Southern California

Setting PYTHONPATH for CosmoMC under Windows

Post by Andrey Egorov » June 08 2015

Hello!
Could anyone have an answer on such a relatively trivial question: I would like to run the python version of Getdist package (part of CosmoMC) including gui.py on Windows 7 machine. I've installed Python 2.7 with necessary packages. However, can't do PYTHONPATH setting. The instruction on top of this page - http://cosmologist.info/cosmomc/readme_python.html - specifies it for Linux-like systems only. What should I do exactly in Windows command prompt? Simple attempts like "set PYTHONPATH=C:\cosmomc\python" didn't help - python gui.py replies "Configure your PYTHONPATH as described in Readme!". If I'm going to run python Getdist only, I don't have to do any cosmomc installation, right? I'm setting PYTHONPATH just to downloaded cosmomc distribution?
Thanks in advance!

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: Setting PYTHONPATH for CosmoMC under Windows

Post by Antony Lewis » June 08 2015

You can use

My Computer ‣ Properties ‣ Advanced ‣ Environment Variables

(then reopen prompt).

You just want to use GetDist but not CosmoMC you can install the new separate GetDist package

https://pypi.python.org/pypi/GetDist/

which should not require any separate python path configuration. (but also won't be able to fully handle Planck grids, etc).

Andrey Egorov
Posts: 4
Joined: June 08 2015
Affiliation: University of Southern California

Setting PYTHONPATH for CosmoMC under Windows

Post by Andrey Egorov » June 09 2015

Thanks, Antony - today I managed to run the separate Getdist successfully!
Meanwhile, did I understand correctly that I can't easily visualize 1D parameter distributions having all other parameters fixed to their best-fit values (essentially 1D likelihood slices)?

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: Setting PYTHONPATH for CosmoMC under Windows

Post by Antony Lewis » June 09 2015

Not without running a separate chain with the other parameters all fixed.

Andrey Egorov
Posts: 4
Joined: June 08 2015
Affiliation: University of Southern California

Setting PYTHONPATH for CosmoMC under Windows

Post by Andrey Egorov » June 11 2015

I see, Antony.
And finally the last question. Apologize if we bother you too much. This is more result-dependent question, but in case you had some experience with it - your opinion will be very helpful. Particularly, we ran CosmoMC for our task and got parameter chains (everything looks fine). Then we computed likestats and margestats for our parameters by Getdist. From general intuitive considerations we expect that marginalized constraining intervals (like 68,95%) for parameters should be wider (or at least not narrower) than the sliced ones from likestats. However, we see a bit opposite picture. I mean not insane opposite - but opposite. Don't you know what could be the reasons for that unless our likelihood is kind of irregular with multiple maxima with big separations?

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: Setting PYTHONPATH for CosmoMC under Windows

Post by Antony Lewis » June 11 2015

likestats are not "sliced" conditional limits: they are extra of the projections of the full-space N-D confidence contour into each dimension, and usually much wider.

Andrey Egorov
Posts: 4
Joined: June 08 2015
Affiliation: University of Southern California

Setting PYTHONPATH for CosmoMC under Windows

Post by Andrey Egorov » June 12 2015

Got it, thanks!

Post Reply