Modifying Grids

Resampling Grids

If you want to resample a grid after instantiation, you can apply the interp_spectra method:

import numpy as np
from unyt import angstrom

from synthesizer import Grid

grid = Grid("test_grid.hdf5", grid_dir="../../../tests/test_grid")

# Define a new set of wavelengths
new_lams = np.logspace(2, 5, 10000) * angstrom

print("The old grid had dimensions:", grid.spectra["incident"].shape)

# Get the grid interpolated onto the new wavelength array
grid.interp_spectra(new_lam=new_lams)

print("The interpolated grid has dimensions:", grid.spectra["incident"].shape)

The old grid had dimensions: (51, 13, 9244)
The interpolated grid has dimensions: (51, 13, 10000)

Note that this will overwrite the spectra and wavelengths read from the file in place. To get back to the original arrays, a separate Grid can be instantiated without the modified wavelength array.

Collapsing Grids

While most of the models within Synthesizer are capable of handling higher dimensionality grids (i.e. grids with more dimensions than age and metallicity), other workflows might require some method to reduce the dimensionality.

This functionality is provided via the collapse() method, which collapses the grid over a specified axis. There are three ways to actually collapse the grid, specified by the method keyword argument:

• marginalize over the entire axis. This is useful if you don’t know anything about this parameter, and just want to adopt the average over it. You can specify the function used to marginalize with the keyword argument marginalize_function; the default is np.average.

• Pick the value nearest to a specified value. If you know the value of the parameter you want to use, you can collapse the grid by picking the value closest to your specified value. For this, you need to specify the value keyword argument.

• interpolate to a specified value. Similar to nearest, but with a linear interpolation to your specified value. This is useful in workflows where you can’t adopt a discrete value, but be warned that interpolating over a coarse grid can give unrealistic results. You can apply a transformation to the axis before interpolating, e.g. to interpolate in log-space rather than linear space, with the keyword argument pre_interp_function.

For example, here we collapse the grid over the metallicity axis:

print("The old grid had dimensions:", grid.spectra["incident"].shape)
print("and axes:", ", ".join(grid.axes))

# Collapse the grid to a single metallicity value
grid.collapse("metallicities", value=0.03, method="nearest")

print("The collapsed grid has dimensions:", grid.spectra["incident"].shape)
print("and axes:", ", ".join(grid.axes))

The old grid had dimensions: (51, 13, 10000)
and axes: ages, metallicities
The collapsed grid has dimensions: (51, 10000)
and axes: ages

Note that collapse() will overwrite the Grid in place. You can restore the grid to its original dimensionality by re-loading from the HDF5 file:

# Re-load the original grid
grid = Grid("test_grid")

Converting a Grid into an Sed

Any of the spectra arrays stored within a Grid can be returned as Sed objects (see the Sed docs). This enables all of the analysis methods provide on an Sed to be used on the whole spectra grid. To do this we simply call get_sed with the spectra type we want to extract, and then use any of the included methods.

# Get the sed object
sed = grid.get_sed(spectra_type="incident")

# Measure the balmer break for all spectra in the grid (limiting the output)
sed.measure_balmer_break()[5:10, 5:10]

array([[0.98984907, 1.0073854 , 1.00280181, 0.99874568, 0.94231088],
       [1.01585662, 1.01514631, 1.00152695, 0.97659766, 0.97649268],
       [1.04604272, 1.03708356, 1.02677301, 1.00043255, 0.98165109],
       [1.12498086, 1.07987898, 1.05375405, 1.01931841, 1.01376902],
       [1.09541215, 1.06731203, 1.05205108, 1.04534501, 1.03791808]])

Working with flattened grids

Sometimes it’s useful to work with flattened (i.e. one dimensional) versions of a grid of spectra, photometry etc. To facilitate this the get_flattened_axes_values method on grid can be used to get the flattened axes values.

# Get the flattened version of the axes
flattened_axes_values = grid.get_flattened_axes_values()