def simulator(params):
N = params.shape[0]
data = torch.zeros(params.shape[0], 19)
for i in range(N):
data[i, :] = torch.as_tensor(
func.whiten(simulator_wrapper(params[i, :]), whiteness_params))
return data
v_true, Iinj = model.simulator(model.log_theta_true, seed_data, True)
summary_stats_obs = model.calculate_summary_statistics(v_true)
# set up model simulator
def simulator_wrapper(params):
# return tensor
return model.calculate_summary_statistics(model.simulator(params, None))
# run pilot to calc mean and std of summary stats
whiteness_params = func.pilot_run(model, simulator_wrapper, summary_stats_obs)
summary_stats_obs_w = func.whiten(summary_stats_obs, whiteness_params)
w_sim_wrapper = lambda param: torch.as_tensor(func.whiten(simulator_wrapper(param), whiteness_params))
# run inference using SNPE
from sbi.inference import SNPE_C, prepare_for_sbi
print(model.prior)
simulator, prior = prepare_for_sbi(w_sim_wrapper, model.prior)
def build_custom_post_net(batch_theta, batch_x):
flow_lik, flow_post = func.set_up_networks(model.prior.base_dist.low,
model.prior.base_dist.high,
v_true, Iinj = model.simulator(model.log_theta_true, seed_data, True)
summary_stats_obs = model.calculate_summary_statistics(v_true)
# set up model simulator
def simulator_wrapper(params):
# return tensor
return model.calculate_summary_statistics(model.simulator(params, None))
# run pilot to calc mean and std of summary stats
whiteness_params = func.pilot_run(model, simulator_wrapper, summary_stats_obs)
summary_stats_obs_w = func.whiten(summary_stats_obs, whiteness_params)
w_sim_wrapper = lambda param: torch.as_tensor(
func.whiten(simulator_wrapper(param), whiteness_params))
# run inference using SNL
from sbi.inference import SNLE_A, prepare_for_sbi
simulator, prior = prepare_for_sbi(w_sim_wrapper, model.prior)
print("---")
print(model.prior.base_dist.low)
print(model.prior.base_dist.high)
v_true, Iinj = model.simulator(model.log_theta_true, seed_data, True)
summary_stats_obs = model.calculate_summary_statistics(v_true)
# set up model simulator
def simulator_wrapper(params):
# return tensor
return model.calculate_summary_statistics(model.simulator(params, None))
# run pilot to calc mean and std of summary stats
whiteness_params = func.pilot_run(model, simulator_wrapper, summary_stats_obs)
summary_stats_obs_w = func.whiten(summary_stats_obs, whiteness_params)
# m_prior, std_prior = func.load_summary_stats_mean_and_std()
# w_sim_wrapper = lambda param: torch.as_tensor(func.whiten(simulator_wrapper(param), whiteness_params))
def simulator(params):
N = params.shape[0]
data = torch.zeros(params.shape[0], 19)
for i in range(N):
data[i, :] = torch.as_tensor(
func.whiten(simulator_wrapper(params[i, :]), whiteness_params))
return data