SyntheticSEDΒΆ

To start with a 1D image group from the Hyperion output hyperion.rtout your Python script example.py should contain:

import numpy as np

from hyperion.model import ModelOutput
from hyperion.util.constants import pc
from fluxcompensator.sed import *

# read in from Hyperion
m = ModelOutput('hyperion.rtout')
array = m.get_sed(group=0, inclination=0, distance=300*pc,
                  units='ergs/cm^2/s')

Now your SED (from image group=0 starting from 0, which contains a SED in this example) is scaled to 300 pc in units of 'ergs/cm^2/s'. For further details see ModelOutput.

To start with the FluxCompensator class SyntheticSED, you simlpy write:

# initial FluxCompensator array
FC_object = SyntheticSED(input_array=array, unit_out='ergs/cm^2/s', name='test_sed')

The output unit of FluxCompensator unit_out can be defined. Possible are all units like in get_sed. unit_out='ergs/cm^2/s' is the default. The name of the FluxCompensator run name can be defined with a str. All outputs (e.g. plots) will start with this name.