Using Neural Posterior Estimator to retrieve the orbital parameters of the
- orbitize
- lampe
- zuko
- h5py
The script generate.py
creates training, validation and test datasets in HDF5 format. These datasets are used to train the neural posterior estimator.
Here's how to run the script:
python generate.py --size [size] --name [name]
The script mcmc.py
performs Markov Chain Monte Carlo (MCMC) to estimate the orbital parameters. Takes more than 24 hours
python mcmc.py
python generate.py --size [size]
Should be the same size as was generated
The corner plot obtained with the two methods. MCMC took around 27 hours to run and the Neural posterior
estimation model about 1h on a GTX 1080 Ti for a trainingset of size
We can also visualise the confidence interval produced by the two models with the real observations