vipr_reflectometry.flow_models.predict package

Submodules

vipr_reflectometry.flow_models.predict.flow_predictor module

class vipr_reflectometry.flow_models.predict.flow_predictor.FlowPredictor(**kw: Any)

Bases: PredictorHandler

Predictor handler for Flow Network models (CINN, NSF, MAF) using inverse sampling.

Performs probabilistic parameter prediction by sampling from the flow model’s latent space conditioned on experimental curves.

Supports all flow-based architectures: - CINNWithConvEmb - NSFWithConvEmb - MAFNetworkWithConvEmb - And other flow models with a sample() method

class Meta

Bases: object

label = 'flow_predictor'
class vipr_reflectometry.flow_models.predict.flow_predictor.FlowPredictorParams(*, num_samples: int = 100, seed: int | None = None, device: str = 'cpu', q_resolution: float | None = None, number_top_to_bottom: bool = True)

Bases: BaseModel

Parameters for the flow predictor.

device: str
model_config: ClassVar[ConfigDict] = {'extra': 'forbid'}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

num_samples: int
number_top_to_bottom: bool
q_resolution: float | None
seed: int | None

vipr_reflectometry.flow_models.predict.physical_constraint_filter module

Physical constraint filter for flow model predictions.

This filter removes unphysical parameter samples (e.g., negative roughness or thickness values) that can be generated by flow-based models.

class vipr_reflectometry.flow_models.predict.physical_constraint_filter.PhysicalConstraintFilter(app: VIPR)

Bases: object

Filter class for enforcing physical constraints on flow model samples.

This class provides filtering methods to remove unphysical parameter samples generated by flow-based models.

filter_unphysical_samples(prediction_data, **kwargs)

Filter out unphysical parameter samples from flow model predictions.

Flow models (CINN, NSF, MAF) can generate samples with negative values for physical parameters like roughness or thickness. This filter removes such unphysical samples to ensure only valid parameter combinations are passed to postprocessing steps.

The filter checks for: - Roughness < 0 - Thickness < 0 - Sigma < 0

Note: - SLD values are intentionally NOT filtered by sign because negative SLD

can be physically valid in neutron reflectometry.

Weight=-5 ensures this runs early, before other postprocessing filters.

Parameters:

  • prediction_data – Dict containing: - ‘parameter_samples’: numpy array of shape (num_samples, num_params) - ‘param_names’: List of parameter names - ‘all_spectra_results’: (optional) List of per-spectrum results

  • **kwargs – Additional filter parameters

Returns:

  • ‘num_filtered’: Number of unphysical samples removed

Return type:

prediction_data with only physical samples, augmented with

Example

If 2000 samples were generated and 150 had negative roughness: - Returns 1850 valid samples - Logs warning about filtered samples - Sets num_filtered=150

Module contents

Filters for flow model predictions.

class vipr_reflectometry.flow_models.predict.PhysicalConstraintFilter(app: VIPR)

Bases: object

Filter class for enforcing physical constraints on flow model samples.

This class provides filtering methods to remove unphysical parameter samples generated by flow-based models.

filter_unphysical_samples(prediction_data, **kwargs)

Filter out unphysical parameter samples from flow model predictions.

Flow models (CINN, NSF, MAF) can generate samples with negative values for physical parameters like roughness or thickness. This filter removes such unphysical samples to ensure only valid parameter combinations are passed to postprocessing steps.

The filter checks for: - Roughness < 0 - Thickness < 0 - Sigma < 0

Note: - SLD values are intentionally NOT filtered by sign because negative SLD

can be physically valid in neutron reflectometry.

Weight=-5 ensures this runs early, before other postprocessing filters.

Parameters:

  • prediction_data – Dict containing: - ‘parameter_samples’: numpy array of shape (num_samples, num_params) - ‘param_names’: List of parameter names - ‘all_spectra_results’: (optional) List of per-spectrum results

  • **kwargs – Additional filter parameters

Returns:

  • ‘num_filtered’: Number of unphysical samples removed

Return type:

prediction_data with only physical samples, augmented with

Example

If 2000 samples were generated and 150 had negative roughness: - Returns 1850 valid samples - Logs warning about filtered samples - Sets num_filtered=150