Experimental Design

One part of SWIFT-Emulator’s i/o features is the ability to generate Latin Hypercube (LH) designs and save them to SWIFT parameter files, one file for each set of parameters. For this example you will need to download some data from http://virgodb.cosma.dur.ac.uk/swift-webstorage/IOExamples/emulator_output.zip.

We do this by combining the swiftemulator.design() with swiftemulator.io.swift(). First we have to specify what parameter we want to vary.

from swiftemulator.design import latin
from swiftemulator.io.swift import write_parameter_files
from swiftemulator import ModelSpecification

spec = ModelSpecification(
    number_of_parameters=5,
    parameter_names=[
        "EAGLEFeedback:SNII_energy_fraction_min",
        "EAGLEFeedback:SNII_energy_fraction_max",
        "EAGLEFeedback:SNII_energy_fraction_n_Z",
        "EAGLEFeedback:SNII_energy_fraction_n_0_H_p_cm3",
        "EAGLEFeedback:SNII_energy_fraction_n_n",
    ],
    parameter_printable_names=[
        "$f_{\\rm E, min}$",
        "$f_{\\rm E, max}$",
        "$n_{Z}$",
        "$\\log_{10}$ $n_{\\rm H, 0}$",
        "$n_{n}$",
    ],
    parameter_limits=[
        [0.0, 1.0],
        [1.0, 7.0],
        [-0.5, 5.0],
        [-1.0, 1.5],
        [-0.5, 5.0],
    ],
)

parameter_transforms = {"SNII_energy_fraction_n_0_H_p_cm3": lambda x: 10.0 ** x}

In this case it is important that your parameter_names are identical to the names in the SWIFT parameter file. The parameter file is a .yml file so the individual parameters should be named in that format.

In this case the fourth parameter, EAGLEFeedback:SNII_energy_fraction_n_0_H_p_cm3 is sampled in log-space. The parameter_limits are given in log-space as well in this case, but you need to define the transformation needed when going from the design space, to the value you want to put in the parameter file.

Generating the LH can then be done with swiftemulator.design.latin.create_hypercube().

number_of_simulations = 30

model_parameters = latin.create_hypercube(
    model_specification=spec,
    number_of_samples=number_of_simulations,
)

Now we can use the SWIFT i/o to write these to a set of parameter files. You will have noticed that we only need to provide the parameters that we want to vary. This is because we provide write_parameter_files with a base parameter file. This file should hold the base values for all other parameters. Example parameter files can be found on the main SWIFT repository. For this example we will use one of the parameter files from the example hypercube.

base_parameter_file = "emulator_output/input_data/1.yml"
output_path = "."

write_parameter_files(
filenames={
    key: f"{output_path} / {key}.yml"
    for key in model_parameters.model_parameters.keys()
},
model_parameters=model_parameters,
parameter_transforms=parameter_transforms,
base_parameter_file=base_parameter_file,
)

This writes 30 files to the current directory. These files can then be used to run SWIFT for each of the models.