Sampling

NNFE is a data-free method: no ground-truth solution data is required. Sampler generates control-variable point sets from a prescribed distribution and provides mini-batch draws for the training loop.

Sampler

nnfe.sampling.Sampler

Manages the control-variable point sets used for training and testing.

On construction the full training set (X) and test set (Y) are generated and stored as NumPy arrays. During training, mini-batches are drawn from X without replacement via :meth:draw_batch.

Parameters:

Name	Type	Description	Default
config		:class:~nnfe.nnfe_config.SamplerConfig instance.	required
rng_key		Integer seed (currently unused; JAX-based batching is seeded at draw time).	0

Attributes:

Name	Type	Description
config		:class:~nnfe.nnfe_config.SamplerConfig used to build this sampler.
training_sampler		Bound method that generates training samples.
testing_sampler		Bound method that generates test samples.
X		Full training point set, shape (N_train, n_ctrl_vars).
Y		Full test point set, shape (N_test, n_ctrl_vars).

uniform(mins=[0.0], maxes=[1.0], samples=[5], between_training=False)

Generate a uniform Cartesian grid of control-variable samples.

For each dimension, samples equally-spaced points are generated between mins and maxes using :func:numpy.linspace, then the full tensor-product grid is assembled via :func:numpy.meshgrid.

Parameters:

Name	Type	Description	Default
mins		Lower bounds for each dimension.	[0.0]
maxes		Upper bounds for each dimension.	[1.0]
samples		Number of points per dimension.	[5]
between_training		If True, shift the grid so points fall between training points (half-step offset), reducing the sample count by one per dimension.	False

Returns:

Type	Description
	NumPy array of shape (prod(samples), n_dims) containing all
	grid points.

Raises:

Type	Description
AssertionError	If mins, maxes, and samples have different lengths.

draw_batch(rng_key, batch_size)

Randomly sample a mini-batch from the training set without replacement.

Parameters:

Name	Type	Description	Default
rng_key		JAX PRNG key. A new key is split off internally so the caller's key is advanced.	required
batch_size		Number of samples to draw.	required

Returns:

Type	Description
	Tuple of (updated_rng_key, batch) where batch has shape
	(batch_size, n_ctrl_vars).

safe_eval(expr)

Return expr as a numeric value, rejecting arbitrary strings.

Parameters:

Name	Type	Description	Default
expr		A float or int value.	required

Returns:

Type	Description
	The input unchanged.

Raises:

Type	Description
ValueError	If expr is not a float or int.