LINFA options
Option |
Type |
Description |
---|---|---|
|
|
Name of the experiment |
|
|
Type of normalizing flow ( |
|
|
Number of normalizing flow layers (default |
|
|
Number of neurons in MADE hidden layer (default |
|
|
Number of hidden layers in MADE (default |
|
|
Activation function for MADE network
used by MAF (default:
'relu' ) |
|
|
Input order for MADE mask creation
(default:
'sequential' or 'random' ) |
|
|
Adds batchnorm layer after each MAF or
RealNVP layer (default
True ) |
|
|
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 dimensionality (default |
|
|
Number of samples from the basic distribution
generated at each iteration (default
100 ) |
|
|
Number of additional true model evaluations at
each surrogate model update (default
2 ) |
|
|
Total number of NF iterations (default |
Option |
Type |
Description |
---|---|---|
|
|
Type of SGD optimizer (default |
|
|
Learning rate (default |
|
|
Learning rate decay (default |
|
|
Type of learning rate scheduler
(
'StepLR' or 'ExponentialLR' ) |
|
|
Number of steps before learning rate
reduction for the step scheduler
|
|
|
Number of iterations between successive
loss printouts (default
10 ) |
Option |
Type |
Description |
---|---|---|
|
|
Name of output folder where
results files are written
|
|
|
Name of the log file which stores the iteration number,
annealing temperature and value of the loss function at each iteration
|
|
|
Seed for random number generator |
Option |
Type |
Description |
---|---|---|
|
|
Batch size used to generate results
after
save_interval iterations |
|
|
Number of NF iteration between successive
updates of the surrogate model (default
1000 ) |
|
|
Maximum allowable number of true model evaluations |
|
|
Number of pre-training iterations
for surrogate model (default
40000 ) |
|
|
Number of iterations for the surrogate model update (default |
|
|
Folder where the surrogate model is stored (default |
|
|
Start by pre-training a new surrogate
and ignore existing surrogates (default
True ) |
|
|
Save interval for surrogate model (None for no save, default |
Option |
Type |
Description |
---|---|---|
|
|
Flag to activate the annealing scheduler.
If this is
False , the target posteriordistribution is left unchanged during
the iterations
|
|
|
Type of annealing scheduler
(default
'AdaAnn' or 'fixed' ) |
|
|
KL tolerance. It is kept constant during inference and used
in the numerator of equation (2).
|
|
|
Initial inverse temperature. |
|
|
Number of batch samples during annealing. |
|
|
Number of batch samples at \(t=1\). |
|
|
Number of initial parameter updates at \(t_0\). |
|
|
Number of parameter updates after each temperature update.
During such updates the temperature is kept fixed.
|
|
|
Number of parameter updates at \(t=1\) |
|
|
Number of Monte Carlo samples used to evaluate
the denominator in equation (2)
|
Option |
Type |
Description |
---|---|---|
|
|
Do not use GPU acceleration |