Neural Importance Sampling

How to run code

Server:

integration_experiment.py -c some_config.conf

Example config file:

train
{
  function = Gaussian
  epochs = 200
  batch_size = 100
  learning_rate = 0.0001
  num_hidden_dims = 128
  num_coupling_layers = 3
  num_hidden_layers = 5
  network_type = 'unet' # ['mlp', 'unet']
  num_blob_bins = 0
  num_piecewise_bins = 5
  loss = MSE
  num_context_features = 6
  hybrid_sampling = False
  coupling_name="piecewiseLinear",
}
logging
{
  plot_dir_name = "./Plots/SaveOutput"
  save_plots = False
  save_plt_interval = 10
  plot_dimension = 2
  tensorboard
  {
    use_tensorboard = False
    # wandb_project = "NIS"
  }
}

Client:

client.py

Tests

For testing use the pytest module:

pip install pytest

To test the application, run the command in the terminal inside the project folder:

...\NIS>pytest