kmzzhang / nbi

Package for Neural Posterior Estimation and Importance Sampling focused on Astronomical Applications
Other
34 stars 4 forks source link

First code example from README does not execute #18

Open stefanv opened 1 year ago

stefanv commented 1 year ago
import nbi

# hyperparameters
featurizer = {
    "type": "resnet-gru",
    "dim_in": 1,
    "max_hidden": 64
}

flow = {
    "n_dims": 1,        # parameter space dimension
    "flow_hidden": 32,  # generally no larger than max_hidden
    "num_blocks": 10    # depends on complexity of posterior shape
}

engine = nbi.NBI(
    flow,
    featurizer,
    simulator,
    noise,
    priors,
    device='cpu'        # 'cuda', 'cuda:0', 'mps' for M1/M2 Mac GPU
)
engine.fit(
    n_sims=1000,
    n_rounds=1,
    n_epochs=100
)
y_pred, weights = engine.predict(x_obs, x_err, n_samples=2000)