https://github.com/sri-csl/dnre

Code accompanying paper: Direct Amortized Likelihood Ratio Estimation
https://github.com/sri-csl/dnre

Last synced: about 2 months ago
JSON representation

Code accompanying paper: Direct Amortized Likelihood Ratio Estimation

• Host: GitHub
• URL: https://github.com/sri-csl/dnre
• Owner: SRI-CSL
• License: mit
• Created: 2023-11-16T15:37:33.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2023-11-21T22:23:17.000Z (about 1 year ago)
• Last Synced: 2023-11-21T23:28:38.727Z (about 1 year ago)
• Language: Jupyter Notebook
• Size: 1.62 MB
• Stars: 1
• Watchers: 18
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Code accompanying paper: Direct Amortized Likelihood Ratio Estimation

To run the code in this repository, first run:

```

pip install .

```

This repository contains:

* `./notebooks/quadcopter_example.ipynb`: This is an example notebook that runs all three likelihood ratio estimators (DNRE, BNRE, NRE) to both train and sample quadcopter designs.

* `./src/dnre/benchmark.py`: This code runs the benchmark using the following command:

    ```

    python src/dnre/benchmark.py --model_type dnre --task two_moons --save_dir ./benchmark_results/two_moons/dnre --device 0

    ```

    Where we have selected DNRE as the approach to perform grid search over in the above command. The code includes comments for all additional options.

    

* `./src/dnre/benchmark_evaluate.py`: This code evaluates the above best result from the grid search using the following command:

    ```

    python src/dnre/benchmark_evaluate.py --model_type dnre --task two_moons --path_dir ./benchmark_results/two_moons/dnre --device 0 --metric coverage

    ```

    Where we point to the same directory as above. This will evaluate the expected coverage, but there are also options for C2ST and the log posterior.

* `./data/data_dict_4490`: Contains the data from the quadcopter experiment.

## Acknowledgements

This project was supported by DARPA under the Symbiotic

Design for Cyber-Physical Systems (SDCPS) with contract

FA8750-20-C-0002.

The views, opinions and/or findings expressed

are those of the author and should not be interpreted as

representing the official views or policies of the Department

of Defense or the U.S. Government.