Updates examples for new major releases of pymc/pytensor/arviz

[jessegrabowski]

Arviz v1 is going to break everything for sure, but there might be small details that will stop working due to pymc v6/pytensor v3. This issue is a tracker for checking that everything works post-release.

Introductory

• pymc_overview
• glm_linear
• api_quickstart

Library Fundamentals

• dimensionality
• pymc_pytensor
• data_container

How To

• posterior_predictive
• model_comparison
• lkj
• missing_data_imputation
• blackbox_external_likelihood_numpy
• copula_estimation
• debugging
• hypothesis_testing
• marginalizing_models
• model_builder
• profiling
• spline
• updating_priors
• wrapping_jax_function

Generalized Linear Models

• glm_binomial_regression
• glm_discrete_choice_models
• glm_hierarchical_binomial_model
• glm_missing_values_in_covariates
• glm_model_selection
• glm_negative_binomial_regression
• glm_ordinal_features
• glm_ordinal_regression
• glm_out_of_sample_predictions
• glm_poisson_regression
• glm_robust_with_outlier_detection
• glm_robust
• glm_rolling_regression
• glm_truncated_censored_regression
• multilevel_modeling

Case Studies

• best
• cfa_sem
• gev
• bayesian_sem_workflow
• bayesian_workflow
• binning
• factor_analysis
• hierarchical_partial_pooling
• item_response_nba
• occupancy
• probabilistic_matrix_factorization
• putting_workflow
• reinforcement_learning
• reliability_and_calibrated_prediction
• rugby_analytics
• ssm_hurricane_tracking

Causal Inference

• glm_simpsons_paradox
• bayesian_ab_testing_introduction
• bayesian_nonparametric_causal
• interventional_what_if_do_operator
• difference_in_differences
• excess_deaths
• interrupted_time_series
• interventional_distribution
• mediation_analysis
• moderation_analysis
• regression_discontinuity

Gaussian Processes

• gp_births
• gp_circular
• gp_heteroskedastic
• gp_kron
• gp_latent
• gp_marginal
• gp_maunaloa
• gp_maunaloa2
• gp_means_and_covs
• gp_sparse_approx
• gp_tprocess
• gp_smoothing
• hsgp_advanced
• hsgp_basic
• mogp_coregion_hadamard
• gaussian_process
• log_gaussian_cox_process

Time Series

• ar
• air_passengers_prophet_bayesian_workflow
• euler_maruyama_sdes
• forecasting_structural_timeseries
• mv_gaussian_random_walk_demo
• time_series_generative_graph
• bayesian_var_model
• longitudinal_models
• stochastic_volatility

Spatial Analysis

• conditional_autoregressive_priors
• malaria_prevalence
• nyc_bym

Diagnostics and Model Criticism

• bayes_factor
• diagnosing_biased_inference_with_divergences
• model_averaging
• sampler_stats

Bayesian Additive Regression Trees

• bart_categorical_hawks
• bart_heteroscedasticity
• bart_introduction
• bart_quantile_regression

Mixture Models

• dependent_density_regression
• dirichlet_mixture_of_multinomials
• dp_mix
• gaussian_mixture_model
• marginalized_gaussian_mixture_model

Survival Analysis

• bayes_param_survival
• censored_data
• frailty_models
• survival_analysis
• weibull_aft

ODE Models

• ode_api_introduction
• ode_api_shapes_and_benchmarking
• ode_lotka_volterra_multiple_ways
• ode_with_manual_gradients

MCMC

• demetropolisz_efficiency_comparison
• demetropolisz_tune_drop_fraction
• smc_abc_lotka_volterra_example
• smc2_gaussians
• fast_sampling_jax_numba
• lasso_block_update
• sampling_compound_step
• sampling_conjugate_step

Variational Inference

• glm_hierarchical_advi_minibatch
• bayesian_neural_network_advi
• empirical_approx_overview
• pathfinder
• variational_api_quickstart

Statistical Rethinking Lectures

• sr_02_garden_of_forking_data
• sr_03_geocentric_models
• sr_04_categories_curves
• sr_05_elemental_confounds
• sr_06_good_bad_controls
• sr_07_fitting_over_under
• sr_08_mcmc
• sr_09_modeling_events
• sr_10_counts_hidden_confounds
• sr_11_ordered_categories
• sr_12_multilevel_models
• sr_13_multilevel_adventures
• sr_14_correlated_features
• sr_15_social_networks
• sr_16_gaussian_processes
• sr_17_measurement_misclassification
• sr_18_missing_data
• sr_19_glm_madness
• sr_20_horoscopes

[snakefood3232]

Noted that Arviz v1 and other updates could break existing examples, specifically within the pymc/pytensor/arviz ecosystem. I'd handle this by testing each example systematically against the new versions, updating code and dependencies where necessary for compatibility. Can have a PR ready in 3-5 days to address these issues. I've previously handled similar updates in machine learning libraries, ensuring smooth transitions to major releases.

[jessegrabowski]

If you want to help, please open one PR for one notebook.

[aloctavodia]

Regarding arviz, there are a few examples already using the new arviz. Those are using import arviz.preview as az, we should switch those imports to import arviz as az. I am working on this

[YichengYang-Ethan]

I'd like to take variational_api_quickstart. Tried running it on pymc 6.0.1 / pytensor 3.0.3 / arviz 1.1.0 and hit a few things:

• approx.sample() needs to be inside with model: now (cells 9, 21, 50, 53)
• az.plot_posterior is gone, used az.plot_dist instead (cell 21)
• pm.callbacks moved to pm.variational.callbacks (cells 28, 40, 84)
• total_size=data.shape (tuple) crashes get_scaling in cell 100, switched to data.shape[0]
  Will open a draft PR once I've done a clean Restart & Run All.

[YichengYang-Ethan]

Opened draft PR #882.