Problem reading MILES library #29
Comments
|
This is odd. For some reason sps_setup is looking for a MILES library at Z=0. (whereas it should be something like Z=0.0008). So it seems like the zlegend values are being truncated. Can you let me know the output of more $SPS_HOME/SPECTRA/MILES/zlegend.dat
ls $SPS_HOME/SPECTRA/MILES/*bin |
|
lluis:~ lluisgalbany$ more $SPS_HOME/SPECTRA/MILES/zlegend.dat |
|
While I am able to run the executable 'autosps.exe' in the fsps/src folder (so fsps in fortran runs well), the problem seems to be in the python fsps.StellarPopulation function. |
|
hmm. well, I'm not sure why this file exists Can you try adding the following line to WRITE(*, *) z, nzinit, zstype, zlegendinit(z)and then from the command line and reinstall python-fsps BTW, how did you install python-fsps? pip or a development version? |
|
It is bizarre that what are your fsps compile flags? |
|
'imiles_z0..spectra.bin' that was just me trying to check if it was a truncation problem... flags for (fortran) fsps were F90FLAGS = -O3 -march=native -fPIC -cpp and this si what I got wfter reinstalling all: In [2]: import fsps In [3]: sp = fsps.StellarPopulation(compute_vega_mags=False, zcontinuous=1, sfh=0, logzsol=0.0, dust_type=2, dust2=0.2) |
|
and what is the output when you run autosps.exe? |
|
it seems to do fine! ./autosps.exe ---> Using IMF 0 Specify SFH [0-2, def:0] ---> Computing an SSP enter metallicity [1-22; def:20]: ---> Using metallicity 20 corresponding to log(Z/Zsol)= 0.00 Include default dust model? [yes/no, def:no] ---> tau1= 0.00, tau2= 0.00 Enter filename [def: "CSP.out"] ---> Output filename: CSP.out |
|
Very strange. python-fsps does not touch anything in sps_setup, and indeed uses the compiled fsps/src/sps_setup.o module. All I can think is something has gone very strange in the f2py wrapping and linking process. What version of numpy are you using? |
|
I run OSX 10.6.4, Homebrew GCC 6.3.0, numpy 1.12.0 and do not have this problem. The only major difference from your setup besides numpy is that I don't use the march=native f90 compile flag. Maybe try without that? |
|
I have numpy 1.11, but I am trying to update to 1.13. |
|
ok. and I meant OSX 10.12.4, so similar to your OS. If you till have the problem after upgrading numpy please post the log of the python-fsps install, i.e. the output of python setup.py install. |
|
It works. I did something more than just updating numpy. I started a new python3.6 environment and installed everything again. It could be due to some stupid conflict in my previous python installation. thank you very much for your help! |
|
glad it's working! |
|
Wait! I just created a python2.7 environment AND I HAVE AGAIN THE SAME PROBLEM! Do you use 2.7 or 3.6? |
|
I use 2.7.13 So weird! |
|
Oh... at least I can still use fsps with python 3... just to follow up your previous comment, this is the output of python (2.7) setup.py install: |
|
you have a number of entries of the form
This indeed looks like a problem with the linking. I'm not sure where it's coming from. I do note that multiple compilers were found during installation, though I think the right one does get used eventually.
You also have the following entry in the install log that I don't see in mine, which might be related?
|
|
also this near the very beginning seems wrong - why does the python installer think you are on OSX 10.6?
perhaps related to #15. |
|
with
all warnings went off. Although it still builds in '-10.6-x86_64' There is also at the beginning this warning:
|
|
I don't think filter_keys.txt is used anymore. Anyway I don't have it, though my install doesn't complain that it's missing. Did you try to install python-fsps using pip at some point? |
|
yes (to the pip installation), and I have the same problem. |
|
and did you make sure to pip uninstall? |
|
sure. |
|
what about that warning to f2py_size? |
|
I get the same warning and can't reproduce the error you get. I'm kind of at a loss. Can you do Also just to confirm, you are still getting the same error after setting the deployment target (but no linker warnings)? |
|
FWIW here's the travis build script (which passed with all tested numpy and python versions on the last commit in July) |
|
this is what I get in my python 2.7 environment (the one not working):
and this is what I get in the default python3.6:
|
|
Yes! still getting the error but with no warnings after the deployment target. |
|
Any progress on this issue? I'm running into the same problem with python 2.7, Mac 10.13.1, gfortran 7.2.0, and numpy 1.13.3. Thanks. |
|
sorry no progress yet @johnnygreco. Thanks for the additional datapoint. |
|
Thanks @johnnygreco! Did you try any versions in between or did you go straight to 6.4.0 from 7.2.0? |
|
I went straight from 7.2.0 to 6.4.0. |
|
I do have 7.1.0. I will try downgrading to 6.4.0. |
|
this is all useful info, thanks. |
|
OS: EL7 I'm seeing a similar issue trying to use the python-fsps module: Having done a bit of digging around and trying to execute the Line 47 of file spec_bin.f90: |
|
Hi there, I am not sure if I should start a seperate thread here but I am having a very similar issue as described where the only difference is the metallicity with which the issue seems to exist. SPS_SETUP ERROR: miles spectral library cannot be opened Z=0.0008 I have uninstalled and reinstalled FSPS on my computer as suggestd above however I am unsure if this is an issue with gcc or an update to FSPS that I may have missed. Thank you very much for your help as always! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I am using a Mac 10.12.6, gcc 7.1.0 (but also tried with v6.3.0), and I am finding a problem when running the first instruction in prospector tutorial:
python prospector.py --param_file=demo_params.py --objid=0 --outfile=demo_obj0
gives the following error:
SPS_SETUP ERROR: miles spectral library cannot be opened Z=0.
It seems that something is being truncated.
The text was updated successfully, but these errors were encountered: