phlower.nn.MeanConstrainedSigmoid

class phlower.nn.MeanConstrainedSigmoid(output_feature_dim, reference_variable_name, weight_field_name=None, slope=2.0, learnable_slope=False, convergence_threshold=1e-05, max_iteration=100, epsilon=1e-06, **kwards)

Bases: IGenericPhlowerCoreModule[IPhlowerLayerParameters, PhlowerTensor], Module

MeanConstrainedSigmoid projects the input feature to (0, 1)-valued features with its mean equaling to the given mean. It applies sigmoid(a * h + b), with b determined by solving the equation mean sigmoid(a * h + b) = target_mean.

Parameters:

• output_feature_dim (int) – Output feature dimension.

• reference_variable_name (str) – Variable name to compute the target mean.

• weight_field_name (str | None = None) – Name of weight (e.g., volume) data in field_data.

• slope (float = 2.0) – Slope of the linear transformation before sigmoid. If learnable_slope is True, it is used as an initial value.

• learnable_slope (bool = False) – If True, lean slope.

• convergence_threshold (float = 1e-5) – Threshould for the Newton solver to obtain the target mean.

• max_iteration (int = 100) – Max iteration for the Newton solver.

• epsilon (float = 1.0e-6) – Small value to suppress numerical instability.

Examples

>>> mix = MeanConstrainedSigmoid(reference_variable_name="a")
>>> mix(data)

Methods

forward(data, *[, field_data])	forward function which overloads torch.nn.Module
from_setting(setting)	Create MeanConstrainedSigmoid from setting object
get_nn_name()	Return name of MeanConstrainedSigmoid

Attributes

T_destination
call_super_init
dump_patches
training

forward(data, *, field_data=None, **kwards)

forward function which overloads torch.nn.Module

Parameters:

• data (IPhlowerTensorCollections) – IPhlowerTensorCollections data which receives from predecessors

• field_data (ISimulationField | None) – ISimulationField | None Constant information through training or prediction

Returns:

Tensor object

Return type:

PhlowerTensor

classmethod from_setting(setting)

Create MeanConstrainedSigmoid from setting object

Parameters:

setting (MeanConstrainedSigmoidSetting) – setting object

Returns:

MeanConstrainedSigmoid

Return type:

Self

classmethod get_nn_name()

Return name of MeanConstrainedSigmoid

Returns:

name

Return type:

str