LINFA options

Table 2 General parameters

Option	Type	Description
name	str	Name of the experiment
flow_type	str	Type of normalizing flow ('maf' or 'realnvp')
n_blocks	int	Number of normalizing flow layers (default 5)
hidden_size	int	Number of neurons in MADE hidden layer (default 100)
n_hidden	int	Number of hidden layers in MADE (default 1)
activation_fn	str	Activation function for MADE network used by MAF (default: 'relu')
input_order	str	Input order for MADE mask creation (default: 'sequential' or 'random')
batch_norm_order	bool	Adds batchnorm layer after each MAF or RealNVP layer (default True)
save_interval	int	How often to save results from the normalizing flow iterations. Saved results include posterior samples, loss profile, samples from the posterior predictive distribution with observations and marginal statistics
input_size	int	Input dimensionality (default 2)
batch_size	int	Number of samples from the basic distribution generated at each iteration (default 100)
true_data_num	int	Number of additional true model evaluations at each surrogate model update (default 2)
n_iter	int	Total number of NF iterations (default 25001)

Table 3 Optimizer and learning rate parameters

Option	Type	Description
optimizer	string	Type of SGD optimizer (default 'Adam')
lr	float	Learning rate (default 0.003)
lr_decay	float	Learning rate decay (default 0.9999)
lr_scheduler	string	Type of learning rate scheduler ('StepLR' or 'ExponentialLR')
lr_step	int	Number of steps before learning rate reduction for the step scheduler
log_interval	int	Number of iterations between successive loss printouts (default 10)

Table 4 Output parameters

Option	Type	Description
output_dir	string	Name of output folder where results files are written
log_file	string	Name of the log file which stores the iteration number, annealing temperature and value of the loss function at each iteration
seed	int	Seed for random number generator

Table 5 Surrogate model parameters (NoFAS)

Option	Type	Description
n_sample	int	Batch size used to generate results after save_interval iterations
calibrate_interval	int	Number of NF iteration between successive updates of the surrogate model (default 1000)
budget	int	Maximum allowable number of true model evaluations
surr_pre_it	int	Number of pre-training iterations for surrogate model (default 40000)
surr_upd_it	int	Number of iterations for the surrogate model update (default 6000)
surr_folder	str	Folder where the surrogate model is stored (default './')
use_new_surr	bool	Start by pre-training a new surrogate and ignore existing surrogates (default True)
store_surr_interval	int	Save interval for surrogate model (None for no save, default None)

Table 6 Parameters for the adaptive annealing scheduler (AdaAnn)

Option	Type	Description
annealing	bool	Flag to activate the annealing scheduler. If this is False, the target posterior distribution is left unchanged during the iterations
scheduler	string	Type of annealing scheduler (default 'AdaAnn' or 'fixed')
tol	float	KL tolerance. It is kept constant during inference and used in the numerator of equation (2).
t0	float	Initial inverse temperature.
N	int	Number of batch samples during annealing.
N_1	int	Number of batch samples at \(t=1\).
T_0	int	Number of initial parameter updates at \(t_0\).
T	int	Number of parameter updates after each temperature update. During such updates the temperature is kept fixed.
T_1	int	Number of parameter updates at \(t=1\)
M	int	Number of Monte Carlo samples used to evaluate the denominator in equation (2)

Table 7 Device parameters

Option	Type	Description
no_cuda	bool	Do not use GPU acceleration