Prior Predictive

Interactive prior predictive check functionality for SciStanPy models.

This module provides interactive widgets and visualizations for conducting prior predictive checks on SciStanPy models. Prior predictive checks allow users to examine the behavior of their models before fitting to data by sampling from prior distributions and visualizing the resulting predictions.

The module centers around the PriorPredictiveCheck class, which creates an interactive dashboard with sliders for model hyperparameters and dropdown menus for selecting visualization options. This enables rapid exploration of how different prior specifications affect model behavior.

The interface automatically adapts to the structure of the provided model, exposing only relevant parameters and visualization options based on the data dimensions and available coordinates.

Note

Instances of the PriorPredictiveCheck class will not typically be accessed directly by users. Instead, they are accessed via the Model.prior_predictive() method.

class scistanpy.plotting.prior_predictive.PriorPredictiveCheck(model: ssp_model.Model, copy_model: bool = False)[source]

Bases: object

Interactive dashboard for conducting prior predictive checks.

This class creates a comprehensive interface for exploring model behavior through prior predictive sampling. It automatically generates appropriate widgets based on the model structure and provides multiple visualization modes for examining parameter distributions and relationships.

Parameters:

  • model (ssp_model.Model) – SciStanPy model to analyze

  • copy_model (bool) – Whether to create a copy of the model to avoid modifying the original. Defaults to False, meaning updates to model parameters in the interactive plot will be reflected in the original model object.

Variables:

  • model – The model being analyzed (copy or reference)

  • float_sliders – Dictionary of parameter adjustment sliders

  • target_dropdown – Widget for selecting which parameter to visualize

  • group_dim_dropdown – Widget for selecting grouping dimension

  • independent_var_dropdown – Widget for selecting independent variable

  • plot_type_dropdown – Widget for selecting visualization type

  • draw_seed_entry – Widget for setting random seed

  • draw_entry – Widget for setting number of experiments

  • update_model_button – Button to update model and redraw data

  • update_plot_button – Button to update plot without redrawing data

  • fig – HoloViews pane containing the current plot

display() → Row[source]

Create and return the complete interactive dashboard layout.

This method assembles all widgets and the plot display into a comprehensive dashboard layout suitable for display in Jupyter notebooks, Panel applications, or web interfaces.

Returns:

Panel layout containing all interface elements

Return type:

pn.Row

The layout consists of:

  • Left panel: Model hyperparameter sliders and viewing options

  • Right panel: Interactive plot display that updates based on selections

Example:
>>> check = PriorPredictiveCheck(model)
>>> dashboard = check.display() # For Jupyter notebook
>>> dashboard.servable()  # For web deployment
get_ecdf_kwargs() → dict[source]

Generate keyword arguments for empirical CDF plots.

This method constructs the parameter dictionary needed for creating ECDF visualizations using hvplot, including cumulative probability calculations and appropriate hover interactions.

Returns:

Dictionary of hvplot parameters for ECDF plotting

Return type:

dict

The returned dictionary includes:

  • ‘kind’: Set to ‘line’ for step-like ECDF appearance

  • ‘x’: Target parameter values

  • ‘y’: ‘Cumulative Probability’ (computed during data processing)

  • ‘by’: Independent label for grouping multiple ECDFs

  • ‘datashade’: Disabled (False) for ECDF plots to maintain clarity

  • ‘hover’: Set to ‘hline’ for horizontal hover lines

get_kde_kwargs() → dict[source]

Generate keyword arguments for kernel density estimate plots.

This method constructs the parameter dictionary needed for creating KDE visualizations using hvplot, including appropriate grouping and styling options.

Returns:

Dictionary of hvplot parameters for KDE plotting

Return type:

dict

The returned dictionary includes:

  • ‘kind’: Set to ‘kde’ for kernel density estimation

  • ‘x’: None for KDE plots.

  • ‘y’: Target parameter name for the y-axis

  • ‘by’: Independent label for grouping (if applicable)

  • ‘datashade’: Disabled (False) for KDE plots to maintain clarity

get_relationship_kwargs() → dict[source]

Generate keyword arguments for relationship plots.

This method constructs the parameter dictionary for creating relationship visualizations that show how parameters vary with respect to independent variables, using datashading for performance.

Returns:

Dictionary of hvplot parameters for relationship plotting

Return type:

dict

The returned dictionary includes:

  • kind: Set to ‘line’ for continuous relationships

  • x: Independent variable name

  • y: Target parameter name

  • datashade: Enabled (True) for performance with large datasets

  • dynamic: Disabled (False), resulting in all data being embedded in output

  • aggregator: Set to ‘count’ for density-based coloring

  • cmap: ‘inferno’ colormap

get_violin_kwargs() → dict[source]

Generate arguments for violin plot creation.

This method constructs the arguments needed for creating violin plots using HoloViews, handling complex grouping scenarios and determining appropriate categorization based on data structure.

Returns:

Dictionary containing plot arguments and dimensions

Return type:

dict

The method handles multiple grouping scenarios:

  • Single grouping by dimension index

  • Grouping by both dimension and independent variable

  • Automatic determination of primary vs. secondary grouping

The returned dictionary includes:

  • ‘args’: Tuple of (groups…, values) for HoloViews Violin constructor

  • ‘kdims’: List of key dimension names

  • ‘vdims’: Value dimension name (target parameter)

set_group_dim_options(event: Event) → None[source]

Update grouping dimension options based on selected target parameter.

This method configures the group dimension dropdown with appropriate options based on the dimensionality of the currently selected target parameter. It ensures that grouping options are only available for multi-dimensional parameters.

Parameters:

event (Event) – Panel event containing the new target parameter selection

The method updates both the dropdown options and the descriptive text showing dimension sizes to help users understand the data structure. If the previously selected grouping dimension is no longer valid, it automatically resets to a sensible default.

set_independent_var_options(event: Event) → None[source]

Update independent variable options based on grouping dimension.

This method configures the independent variable dropdown with coordinates and data variables that are compatible with the selected grouping dimension. Only variables that vary along the grouping dimension are included as options.

Parameters:

event (Event) – Panel event containing the new grouping dimension selection

The method scans both coordinates and data variables in the xarray Dataset to find suitable independent variables, ensuring compatibility with the selected visualization approach.

set_plot_type_options(event: Event) → None[source]

Update available plot types based on current dimension selections.

This method determines which visualization types are appropriate given the current selections for target parameter, grouping dimension, and independent variable. It enables more sophisticated plot types as more structure is specified.

Parameters:

event (Event) – Panel event (used for callback compatibility is not used)

Plot Type Logic:

  • ECDF and KDE: Always available for any parameter

  • Violin: Available when grouping dimension is selected

  • Relationship: Available when both grouping and independent variable are selected

The method automatically selects the most sophisticated available plot type as the default when options change.