How rousenghak/monte-carlo---likelihood-free-inference-and-sums-of-log-normal-variates works
This research project estimates parameters of sums of log-normal variates using likelihood-free inference. Its main workflow takes observed data and an epsilon tolerance, runs either Reject-ABC or MCMC-ABC, and produces posterior samples and summaries. The repository also documents epsilon and prior sensitivity, posterior-predictive checks, a Gibbs comparison, and method benchmarks. The core ABC implementation and benchmark calls are sampled; some notebook and experiment details are documented in the README or filenames but were not inspected, so uncertain wiring is omitted.
Main entry points: algorithms.py Reject-ABC · benchmark2.py ABC benchmark · Reject_ABC.ipynb Reject-ABC study · MCMC_ABC_optimizedv3.ipynb MCMC-ABC study
Components 15
Data model
- Observed data
- Log-normal simulator
algorithms.py
ABC inference
- Reject-ABC
algorithms.py - MCMC-ABC
algorithms.py - Reject-ABC study
Reject_ABC.ipynb - MCMC-ABC study
MCMC_ABC_optimizedv3.ipynb - MCMC base functions
MCMC_ABCv2.py - ABC algorithms
algorithms.py
Method evaluation
- ABC benchmark
benchmark2.py - Benchmark results
benchmark_results/benchmark_results.csv - Epsilon and Gibbs study
estimating_espsilon_bias.py - Epsilon results
epsilon_comparison
Results analysis
- Posterior samples
- Epsilon plots
plots_epsilon_bias.py
Other components
- Researcher
Connections 15
- Researcher → Observed data: provides data
- Researcher → Reject-ABC study: runs study
- Researcher → MCMC-ABC study: runs study
- Reject-ABC study → Reject-ABC: documents use
- MCMC-ABC study → MCMC base functions: documents use
- Reject-ABC → Log-normal simulator: simulates proposals
- MCMC-ABC → Log-normal simulator: simulates proposals
- Observed data → Reject-ABC: compares against
- Observed data → MCMC-ABC: compares against
- Reject-ABC → Posterior samples: returns samples
- MCMC-ABC → Posterior samples: returns samples
- ABC benchmark → ABC algorithms: calls methods
- ABC benchmark → Benchmark results: writes results
- Epsilon and Gibbs study → Epsilon results: writes results
- Epsilon results → Epsilon plots: loads data