Cosmological Inference

Cosmological Inference#

After training the emulator, the final step in the pipeline is to perform cosmological inference using the observed data. This is done using the sunbird inference module.

Inference Methods#

The sunbird.inference module provides several inference methods:

  • Likelihood-based inference: Traditional Bayesian inference using MCMC or nested sampling

  • Simulation-based inference (SBI): Neural network-based inference methods

  • Approximate Bayesian Computation (ABC): Likelihood-free inference

Running Inference#

The general workflow for inference is:

  1. Load the trained emulator for your chosen statistic(s)

  2. Prepare the observed data in the correct format

  3. Define the priors for your cosmological and galaxy-halo parameters

  4. Run the inference using the appropriate method

  5. Save and analyze the chains

Example Usage#

from sunbird.inference import Inference
from acm.observables import MyObservable  # Your observable class

# Initialize the observable with paths to model and data
obs = MyObservable(...)

# Get the observed data and emulator
data = obs.get_data()
emulator = obs.get_model()

# Set up inference
inference = Inference(
    emulator=emulator,
    data=data,
    priors=priors,
)

# Run inference
chains = inference.run()
inference.save_chain(path='chains/my_chain.pkl')

See also

For detailed examples, see the sunbird documentation and the inference scripts in the projects directory.

Chain Storage#

Inference chains are stored following the conventions described in the Data Storage section.

The chain files contain:

  • samples: The parameter samples from the chain

  • weights: Sample weights (if applicable)

  • ranges: Parameter ranges

  • names: Parameter names

  • labels: LaTeX labels for plotting

  • log_likelihood, log_prior, log_posterior: (method-dependent)

Visualization and Analysis#

After running inference, you can use getdist or other tools to visualize and analyze the chains:

import getdist
from getdist import plots

# Load chain
samples = getdist.loadMCSamples('chains/my_chain')

# Create triangle plot
g = plots.get_subplot_plotter()
g.triangle_plot([samples], filled=True)

Tip

The acm.utils module provides helper functions for chain analysis and visualization.

Contents