sedop is a Monte-Carlo minimization code designed to optimally construct spectral energy distributions (SEDs) for sources of ultraviolet and X-ray radiation employed in numerical simulations of reionization and radiative feedback. See Mirocha et al. 2012 for details on the algorithm.
To clone a copy and install:
git clone https://github.com/mirochaj/sedop.git cd sedop python setup.py install
Currently, sedop depends on h5py, mpi4py, numpy, matplotlib, and scipy. mpi4py is not necessary if you only run in serial, but I would recommend building it as multi-frequency, multi-species optimizations can be rather expensive.
A quick example can be run in an interactive Python session. The following code snippet will run 5 independent Monte Carlo calculations, each with 10,000 steps, to find the optimal monochromatic representation of a 10^5 K blackbody spectrum:
import sedop opt = sedop.OptimizeSED(source_type=1, source_temperature=1e5, num_bins=1) results = opt.run(prefix='test_bb', nsteps=1e4, ntrials=5, clobber=0)
The results dictionary contains the end-points of each Monte Carlo run in energy (Ef) and normalization (Ff), as well as the starting points (Ei and Fi, respectively). The full trajectory of each random walk is saved in the Nsteps and Fsteps elements. A file called 'test_bb.hdf5' will also be saved in your current working directory containing the same contents as the results dictionary.
To do some simple data inspection, one could do, e.g.,
import matplotlib.pyplot as pl
colors = 'k', 'b', 'c', 'm'
for i in range(5): # loop over trials
for j in range(opt.pf['num_bins']): # loop over energy bins
# Plot trajectory of Monte Carlo run
pl.plot(results['Esteps'][j,i,:], color=colors[j], alpha=0.1,
label='bin {}'.format(j+1) if i == 0 else None)
More sophisticated analysis routines live in the sedop.Analysis.Analyze class, though likely still need to be patched up to account for changes in numpy, h5py, and Python itself that have occurred since sedop was originally written back in 2011-2012.
For more examples, checkout the documentation. For now, you'll have to build it locally, i.e., navigate to the doc folder and type make html. Then, open _build/html/index.html.