Filters & Hooks Integration¶

This page describes how the Reflectometry plugin integrates with VIPR’s filter and hook system.

Filter System¶

Reflectometry filters operate on vipr.plugins.inference.dataset.DataSet and are mainly used in preprocessing.

Active filter namespaces¶

Namespace	Purpose
`INFERENCE_PREPROCESS_PRE_FILTER`	Data transforms before prediction (cleaning/interpolation/scaling)
`INFERENCE_PREDICTION_POST_FILTER`	Post-prediction filtering (for example physical constraints)

Preprocess filters (`INFERENCE_PREPROCESS_PRE_FILTER`)¶

Class	Method	Weight	Default in discovery config
`vipr_reflectometry.reflectorch.preprocess.error_bar_filter.ErrorBarFilter`	`preprocess_error_bar_filter`	-10	disabled
`vipr_reflectometry.reflectorch.preprocess.interpolation_filter.InterpolationFilter`	`preprocess_interpolate`	0	enabled
`vipr_reflectometry.flow_models.preprocess.flow_preprocessor.FlowPreprocessor`	`_preprocess_flow`	0	disabled

Execution order is weight-based (lower first), so error-bar cleaning runs before interpolation.

Prediction post-filter (`INFERENCE_PREDICTION_POST_FILTER`)¶

Class	Method	Weight	Purpose
`vipr_reflectometry.flow_models.predict.physical_constraint_filter.PhysicalConstraintFilter`	`filter_unphysical_samples`	-5	Remove unphysical flow samples (for example negative roughness/thickness)

Hook System¶

Reflectometry uses both decorator-discovered hooks and directly registered hooks.

Environment setup hook¶

Registered in vipr_reflectometry.load(app):

app.hook.register('INFERENCE_BEFORE_START_HOOK', setup_reflectometry_env_defaults)

This sets REFLECTOMETRY_ROOT_DIR if missing.

Postprocess visualization hooks¶

Decorator-discovered hooks (enabled via YAML):

Namespace	Class	Method
`INFERENCE_POSTPROCESS_PRE_PRE_FILTER_HOOK`	`vipr_reflectometry.flow_models.postprocess.basic_corner_plot.BasicCornerPlot`	`_create_basic_corner_plot`
`INFERENCE_POSTPROCESS_PRE_PRE_FILTER_HOOK`	`vipr_reflectometry.flow_models.postprocess.marginal_distributions.MarginalDistributions`	`_create_marginal_distributions`
`INFERENCE_POSTPROCESS_PRE_PRE_FILTER_HOOK`	`vipr_reflectometry.flow_models.postprocess.cluster.clustering.hook.ClusterHook`	`_cluster_posterior_samples`
`INFERENCE_POSTPROCESS_PRE_PRE_FILTER_HOOK`	`vipr_reflectometry.flow_models.postprocess.cluster.model_selection.hook.ClusterDiagnosticsHook`	`_cluster_model_selection`

Directly registered collector hooks (active once plugin is loaded):

Namespace	Callback
`INFERENCE_POSTPROCESS_PRE_PRE_FILTER_HOOK`	`ReflectorchDataCollector.collect_prediction_results`
`INFERENCE_POSTPROCESS_PRE_PRE_FILTER_HOOK`	`PanpeDataCollector.collect_panpe_results` (PANPE optional)

YAML Configuration¶

Example: Reflectorch preprocess chain¶

vipr:
  inference:
    filters:
      INFERENCE_PREPROCESS_PRE_FILTER:
      - class: vipr_reflectometry.reflectorch.preprocess.error_bar_filter.ErrorBarFilter
        enabled: true
        method: preprocess_error_bar_filter
        weight: -10
        parameters:
          filter_threshold: 0.3
          filter_consecutive: 3
          filter_remove_singles: true
          filter_remove_consecutives: true
          filter_q_start_trunc: 0.1

      - class: vipr_reflectometry.reflectorch.preprocess.interpolation_filter.InterpolationFilter
        enabled: true
        method: preprocess_interpolate
        weight: 0

Example: Flow postprocess hooks¶

vipr:
  inference:
    hooks:
      INFERENCE_POSTPROCESS_PRE_PRE_FILTER_HOOK:
      - class: vipr_reflectometry.flow_models.postprocess.basic_corner_plot.BasicCornerPlot
        enabled: true
        method: _create_basic_corner_plot
        weight: 0

      - class: vipr_reflectometry.flow_models.postprocess.cluster.clustering.hook.ClusterHook
        enabled: true
        method: _cluster_posterior_samples
        weight: 1
        parameters:
          method: gmm
          n_components: 5
          n_init: 10

Execution Flow (Simplified)¶

LoadDataStep
  -> LoadModelStep
  -> PreprocessStep
       -> INFERENCE_PREPROCESS_PRE_FILTER (weight order)
  -> PredictionStep
       -> INFERENCE_PREDICTION_POST_FILTER (optional)
  -> PostprocessStep
       -> INFERENCE_POSTPROCESS_PRE_PRE_FILTER_HOOK

There is no separate normalize step in the current inference workflow; normalization is part of preprocessing filters.