Model Results API Reference

Model results analysis and visualization for SciStanPy.

This submodule provides comprehensive tools for analyzing and visualizing results from fitted SciStanPy models, supporting both maximum likelihood estimation (MLE) and Markov Chain Monte Carlo (MCMC) inference methods. It offers specialized classes and utilities for processing, diagnosing, and presenting model outputs in formats suitable for scientific analysis.

The submodule is organized around two main result types, each with related analysis capabilities:

  1. scistanpy.model.results.mle.MLE, which holds results from calls to scistanpy.model.model.Model.mle() and provides tools for MLE-based parameter estimation analysis.

  2. scistanpy.model.results.mle.MLEInferenceRes, which holds bootstrapped samples of observed data and parameters for uncertainty quantification and inference using MLE results.

  3. scistanpy.model.results.hmc.SampleResults, which holds results from calls to scistanpy.model.model.Model.mcmc() and provides tools for MCMC-based sampling analysis.

Some notable analysis features available to both results classes include:

  • Interactive Analysis: Widget-based interfaces for exploratory analysis

  • Scalable Processing: Efficient handling of large datasets and complex models

  • Comprehensive Diagnostics: Automated detection and reporting of model issues

  • Integration with Scientific Python: Interoperability with ArviZ, xarray, and Dask

Results classes are designed to integrate with SciStanPy modeling workflows and external analysis tools. For example, the inference_obj attribute shared by both results classes provides an ArviZ InferenceData object representing the results, allowing for further analysis using ArviZ’s rich set of diagnostics and plotting functions.

Users will not typically instantiate result classes directly. Instead, they are returned by model fitting methods such as scistanpy.model.model.Model.mle() and scistanpy.model.model.Model.mcmc(), such as below:

>>> import scistanpy as ssp
>>>
>>> # MLE analysis workflow
>>> mle_result = model.mle(data=observed_data)
>>> mle_analysis = mle_result.get_inference_obj() # Bootstrap observed data
>>> dashboard = mle_analysis.run_ppc()  # Interactive analysis
>>>
>>> # MCMC analysis workflow
>>> mcmc_results = model.mcmc(data=observed_data, chains=4) # Sample from posterior
>>> sample_failures, var_failures = mcmc_results.diagnose()
>>> analyzer = mcmc_results.plot_variable_failure_quantile_traces()

This module is further broken down into the following submodules, each holding results and analysis tools for specific inference methods: