Results and Plotting

By default, the ACID method returns a Result object which contains many methods to analyse the results and simple tools to return the profile and errors.

import ACID_code as acid
spec_file = fits.open('sample_spec_1.fits')
wavelength = spec_file[0].data
spectrum = spec_file[1].data
error = spec_file[2].data
sn = spec_file[3].data
linelist = 'example_linelist.txt'
acid = acid.Acid(velocities, linelist)

# We will use the above result object for the rest of this page
result = acid.ACID(wavelength, spectrum, error, sn, max_steps=5000)

Using the Results Class

The final profiles can be obtained by indexing the class with the first axis corresponding to the frame number, and the second axis corresponding to either the profile values (0), the errors (1), or the covariance matrix (2).

# Access per-frame profiles, for eg frame 0:
profiles = results[0,0]
errors = results[0,1]
cov_matrix = results[0,2]

# Access the combined profile
combined_profile = result[0]
errors = result[1]
cov_matrix = result[2]

Saving the Result

The Result class returned by ACID contains a method to save the entire Result object to a .pkl file for later use. This can be done using the Result.save() method.

# Save the result to a .pkl file
result.save('example_result.pkl', store_sampler=True)
...
# Later, load the result back:
result = acid.Result.load('example_result.pkl')

The save method does not actually store the instance of the class as previously we had issues pickling class instances that stored samplers. This method instead saves just the Data instance stored within and reinitialises the Data and Result instance on load. The Data instance itself stores the sampler which is not pickled directly on save, but instead has its internal dictionary pickled and then reinitialised on load. This is done to avoid issues with pickling issues trying to load instances.

If store_sampler=True (default is True), the sampler is saved with the Data instance, otherwise it is discarded. We recommended leaving this on True when running the deterministic profile model, otherwise you can disable it for the larger chains sampling the entire profile. If you try to open the dictionary yourself with pickle.load() and without using the class method, you will run into errors. If the sampler is not stored, some of the methods when loading the result will not work (eg. plotting walkers).

Plotting

The Result class contains a number of plotting methods to visualise the results of ACID. These include:

:py:function:`ACID_code.Result.plot_profiles`: Plots the final LSD profiles returned by ACID. Can plot multiple profiles if multiple spectra were input.
:py:function:`ACID_code.Result.plot_walkers`: Plots the MCMC walkers for the continuum fit parameters.
:py:function:`ACID_code.Result.plot_corner`: Plots a corner plot of the posterior distributions of the continuum fit parameters.
:py:function:`ACID_code.Result.plot_forward_model`: Plots the forward model fit to the data.
:py:function:`ACID_code.Result.plot_autocorrelation`: Plots the autocorrelation of the MCMC chains for the continuum fit parameters.
:py:function:`ACID_code.Result.plot_acf`: Plots the autocorrelation function for each parameter, averaged across walkers. This is less useful than the above, but kept as it was a part of the emcee example.

These plotting functions have a number of keyword arguments to tailor the plots to your needs. See the API (linked above) for more information on these.

All of the plots which plot parameters (eg. walkers, corner, autocorrelation) will plot the parameters for the continuum parameters. Where deterministic_profile=False and the sampler also fitted the profile, the parameters of the first, last and max profile point are shown to save space.

import ACID_code as acid
result = acid.Result.load('example_result.pkl')
result.plot_profiles()
result.plot_walkers()
result.plot_corner()
result.plot_forward_model()
result.plot_autocorrelation()
result.plot_acf()

These functions can be called directly from the Result object returned by ACID, or from a Result object loaded from a saved .pkl file.