Bayesian Calibration

This folder contains working for the Bayesian Calibration paper(s), and general development of the idea of Bayesian Calibration in EDGES.

Paper URL: https://www.overleaf.com/9477723471yvqbvyshqjph

The main development notebook is devel-bayesian-cal.ipynb: this is a non-production notebook where stuff gets tinkered with. Notably, it contains math etc. for the variance model. All configuration files, outputs etc. for this notebook are in the development/ folder.

Over time, I've tried a few different bits of data and tests etc. First, I tried (re-)calibrating Raul's data. His data is in raul-data/. I didn't pursue this.

The final results will be with Alan's data (in order to make it as close to the NP as possible). All data coming from Alan is in alan-data/.

With Alan's data + calibration choices, I do the following:

0. Data Investigation -- `raw_data_assumptions.ipynb`
    This is done without really requiring calibration or anything. It just looks at
    spectra over time/frequency and evaluates whether our likelihood model 
    assumptions (eg. gaussianity, independence of frequencies) are correct.
1. Pure Calibration -- `alan_calibration.ipynb`
    a) Simulated Data: the point here is to check whether the relevant likelihoods
       work properly (eg. comparing evidence for correct model versus one with
       unnecessary parameters). The MCMC runner script is `run_cal_simulation_mcmc.py`.
       Outputs in `sim_cal/`.
    b) Real Data: the point here is to figure out what number of terms are required
       to successfully characterize the calibration solutions. Overall, outputs 
       consist of multiple runs against the same data with different numbers of terms.
       MCMC runner script at `run_alan_cal_mcmc.py`.  Outputs in `alan_cal/`
2. Calibration + Field Data
    a) Simulated Data: the point here goes beyond that things just "work", to 
       exploring the impact of correlations between foregrounds and calibration, eg.
       by introducing more foreground terms in the *data* but not the model. 
       Notebooks are where??
    b) Real Data: the final point of the paper. Produce MCMCs with `run_alan_data_mcmc.py`
       and analyse with `alan-data-calibration.ipynb`

Along with the above scripts/outputs, there is a CLI utils script that helps to deal with the MCMC outputs: mcmc_utils.py that can be used to eg. print out all runs, show evidences, rename things, etc.