0.19.4

What's Changed

New Features 🎉

• Add Zarr save/load support to ModelIO (zarr>=3 shim) by @PabloRoque in #2512
• Load models without fit data by @ColtAllen in #2538

Documentation 📖

• Adding Consideration Sets Model to Mixed Logit by @NathanielF in #2442
• fix grammar in causal approach notebook by @npandolfi in #2528
• fix(docs): convert Resources section from raw HTML to markdown heading by @erete22 in #2537
• Add Maxdiff methods to Consumer Choice by @NathanielF in #2514
• migrating time_slice_cv plots to its own PlotSuite (MMMCVPlotSuite) by @isofer in #2530

Maintenance 🔧

• fix: cap jax below 0.10.0 due to breaking changes in new release by @isofer in #2500
• feat(mmm): chainable mu effects by @williambdean in #2503
• Adds DecompositionPlots to the pymc_marketing.mmm.plotting namespace by @isofer in #2488
• adds a SensitivityPlots namespace class to mmm.plotting by @isofer in #2494
• fix: restore original_scale Deterministics on MMM save/load round-trip by @PabloRoque in #2508
• Restore from_netcdf with deprecation warning by @cetagostini-wise in #2529
• Unpin JAX by @juanitorduz in #2534

New Contributors

• @npandolfi made their first contribution in #2528
• @erete22 made their first contribution in #2537

Full Changelog: 0.19.3...0.19.4

0.19.3

What's Changed

New Features 🎉

• Update serialization process (#2379) by @isofer in #2391
• Support fixed/user-defined scaling factors for stable model refreshes by @drbenvincent in #2479

Bugfixes 🐛

• Fix Serialization LogNormal from YAML by @juanitorduz in #2476
• Fix: Latex rendering in new Adstock guide by @timbo112711 in #2472

Documentation 📖

• Porting causal identification notebook to MMM multidimensional [V2] by @juanitorduz in #2475
• Use arviz-darkgrid style globally for all plot directives via plot_pre_code by @williambdean in #2492

Maintenance 🔧

• Move mmm-modeling skill out of dot-directory for crawler discovery by @lfiaschi in #2489
• Initial PR for MMMPlotSuite by @isofer in #2437
• Add DiagnosticsPlots class with 6 time-series diagnostic methods by @isofer in #2459
• fix: register xr.DataArray deserializer in SpecialPrior to fix save/load round-trip by @PabloRoque in #2496
• Bump version to 0.19.3 by @juanitorduz in #2498

Full Changelog: 0.19.2...0.19.3

0.19.2

What's Changed

Bugfixes 🐛

• Fix MaskedPrior with non-scalar Prior parameters by @juanitorduz in #2464
• Fix: IndexError in MultiDimensionalBudgetOptimizerWrapper by @juanitorduz in #2452

Documentation 📖

• Adstock documentation - Guide to Adstock Transformations in PyMC-Marketing by @timbo112711 in #2449

Maintenance 🔧

• Bump version to 0.19.2 by @juanitorduz in #2465

Full Changelog: 0.19.1...0.19.2

0.19.1

What's Changed

Bugfixes 🐛

• Make MMM workflow idempotent by @ricardoV94 in #2413
• Restore metric coordinate order after to_xarray by @cetagostini in #2445

Documentation 📖

• Get data of case study from csv by @juanitorduz in #2447

Maintenance 🔧

• Bump version to 0.19.1 by @juanitorduz in #2446

Full Changelog: 0.19.0...0.19.1

0.19.0

Warning

We have migrated all our documentation to the more complete and modern class multidimensional.MMM. See the migration guide: https://www.pymc-marketing.io/en/latest/notebooks/mmm/mmm_migration_guide.html
We will remove the old mmm.MMM class in 0.20.0. Please migrate your code! Most changes are internal optimizations. The APIs are very similar!

What's Changed

Major Changes 🛠

• Default to unpooled transformation priors in Multidimensional MMM by @ricardoV94 in #2293
• CLV API Deprecation by @ColtAllen in #2347

New Features 🎉

• Some fixes TimeSlice Cross Validation and Extend Case Study by @juanitorduz in #2393
• Validate YAML with Pydantic by @juanitorduz in #2418

Bugfixes 🐛

• Raise explicitly when trying to optimize or measure incrementality of discrete channel data by @ricardoV94 in #2337
• bug : mlflow uses the name param on versions 3.0.0 and greater by @daniel-saunders-phil in #2339
• Respect channel ordering when applying budget bounds by @cetagostini in #2350
• Implement MultiDimensionalMMM with pymc.dims by @ricardoV94 in #2204
• Fix: Handle DateIndex in CSA plot_data to resolve crash (fixes #1864) by @shivamlalakiya in #2302
• remove superfluous use of arbitrary_types_allowed in annotations by @nataziel in #2403
• BUG: HSGPKwargs.cov_func does not support custom Cov by @isofer in #2398
• fix nb: GAM NOTEBOOK BROKEN by @juanitorduz in #2416
• fix: work around RuntimeError by @williambdean in #2421
• Add compute_mean_contributions_over_time to MMM multidim by @juanitorduz in #2423
• Fix: Channel coordinates are alphabetically sorted in multidimensional MMM, causing ROAS calculation errors by @juanitorduz in #2422
• Hot Fix: Remove Constrain Variables Sample Posterior Predictive by @juanitorduz in #2440

Documentation 📖

• Revert "cleanup (#2285)" by @ricardoV94 in #2324
• Incrementally, ROAs, CAC, etc... by @isofer in #2290
• Docs: Add CSA module to README header #1821 by @shivamlalakiya in #2301
• docs: Add missing modules by @williambdean in #2343
• Incrementality - the rest of the story by @isofer in #2300
• update docstring for MMM by @isofer in #2358
• Group the custom components notebooks by @daniel-saunders-phil in #2362
• MMM Data Generator by @juanitorduz in #2292
• Demo notebook for cost per unit by @isofer in #2363
• Clarify budget distribution over period by @ricardoV94 in #2394
• Mmm dims migration guide by @ricardoV94 in #2392
• Fix wrong constraints docstrings by @ricardoV94 in #2395
• Custom MMM Model: ROAS Parametrization by @juanitorduz in #2361
• Expose the new incrementality module in some key examples by @juanitorduz in #2410
• Update case study title for clarity by @juanitorduz in #2412
• ROAS parametrizatio notebook fix by @juanitorduz in #2414
• Remove CLV lifetimes Dependency from Notebooks and Unit Tests by @ColtAllen in #2428
• Replace potential by likelihood in calibration by target by @cetagostini in #2407

Maintenance 🔧

• Upper Bound For ArviZ 1.0.0 by @juanitorduz in #2326
• Fix/redo #2290 (incrementally) without utility rename by @isofer in #2333
• Allow all_tests to pass when test jobs are skipped by @isofer in #2334
• Update README.md by @juanitorduz in #2332
• FancierLinearRegression by @williambdean in #2315
• Unblock merges for non-code PRs by always reporting all_tests status by @isofer in #2344
• Cast channel_data to float at construction time by @isofer in #2341
• Add Claude and Cursor AI skill definitions by @cetagostini in #2349
• MMM SKILLs by @juanitorduz in #2354
• skills: clarify lift test calibration with time-varying media by @drbenvincent in #2360
• Support cost_per_unit for channels with non-monetary data (impressions, clicks, etc.) - part 1 by @isofer in #2320
• feat: support sample_prior_predictive kwargs to sample_curve by @williambdean in #2365
• Extends the cost-per-unit support to plotting by @isofer in #2364
• Fix optimization with time varying media by @ricardoV94 in #2396
• Validate YAML structure and warn on unknown keys by @cetagostini in #2409
• [HOT-FIX] NDArray.record() now requires an in_warmup keyword by @juanitorduz in #2429
• Fix: MediaConfigList should be parametrized explicitly by core_dim/media_dim by @juanitorduz in #2426
• make mmm optimization tests deterministic by @juanitorduz in #2431
• Bump version to 0.19.0 by @juanitorduz in #2441

New Contributors

• @shivamlalakiya made their first contribution in #2301

Full Changelog: 0.18.2...0.19.0

0.18.2

What's Changed

Documentation 📖

• mmm utils cleanup by @juanitorduz in #2285

Maintenance 🔧

• Fix compatibility with latest pytensor by @ricardoV94 in #2316

Full Changelog: 0.18.1...0.18.2

0.18.1

What's Changed

Major

• Pin Upper Bound PyMC-Extras because of Prior class breaking changes by @isofer in #2311

Documentation 📖

• Create a notebook to display plot_interactive abilities by @isofer in #2272
• add watermark to interactive notebook by @juanitorduz in #2278
• MMM Custom Model: Splines as Varying Intercept by @juanitorduz in #2277
• fix generate_gallery and add plot interactive image to gallery by @isofer in #2279
• notebook typo by @juanitorduz in #2280
• Add Geo-Level Lift Test Calibration Notebook by @drbenvincent in #2248

Maintenance 🔧

• chore: remove mlflow module experimental warning by @williambdean in #2281
• Update requirements.txt by @stdp95 in #2263
• chore: migrate to pytest-split by @williambdean in #2284
• fix #2299 (Sensitivity Analysis: Integer Truncation when channel_data is integer and sweep_values are fractional) by @isofer in #2307
• fix cv with rng by @juanitorduz in #2308
• Pin Upper Bound PyMC-Extras because of Prior class breaking changes by @isofer in #2311

Full Changelog: 0.18.0...0.18.1

0.18.0

Important

We have migrated all our documentation to the more complete and modern class multidimensional.MMM. See the migration guide: https://www.pymc-marketing.io/en/latest/notebooks/mmm/mmm_migration_guide.html
We will remove the old mmm.MMM class in 0.20.0. Please migrate your code! Most changes are internal optimizations. The APIs are very similar!

What's Changed

Major Changes 🛠

• Run whole suite with C and Numba linkers by @ricardoV94 in #2135

New Features 🎉

• Time-Slice CV Class by @timbo112711 in #2037
• feat: VariableFactory support in HSGP components by @williambdean in #2164
• Pydantic MuEffect by @PabloRoque in #1922
• Add Interactive Plotly Plotting for MMM (Phase 1) by @isofer in #2235
• Add narwhals support to Fivetran data adaptors by @williambdean in #2224

Bugfixes 🐛

• Fix ShiftedBetaGeoModel broadcasting bug in predictive methods by @ColtAllen in #2121
• Fix: Use **kwargs instead of kwargs parameter in MNLogit methods by @Eshaan-byte in #2131
• Add check to to_datetime by @tibisabau in #2238
• Fix Plot Contributions Over Time by @juanitorduz in #2250

Documentation 📖

• Fix typos in clv_quickstart by @JoshTripleWhale in #2120
• [MMM DOCS- Case Study ]Add custom optimizer section. by @juanitorduz in #2125
• Fix plot_curve documentation examples not rendering by @Eshaan-byte in #2130
• Add Tutorial Notebook for ShiftedBetaGeoModel by @ColtAllen in #2122
• remove dims from MediaTransform call docstring by @williambdean in #2144
• work on better SEO properties by @OriolAbril in #2119
• Harmonize HSGP arguments in MMM Multidimensional & Test using ROAS calibration notebook by @juanitorduz in #2172
• Fix call to .sample_curves in multi-dim notebook by @daniel-saunders-phil in #2192
• Fix call to .sample_curves in all multi-dim notebooks by @isofer in #2196
• Adding Discrete choice mixed logit Model by @NathanielF in #2150
• plot_channel_contribution_grid to MMM Multidimensional + Documentation Notebook by @juanitorduz in #2177
• [Doc]: Advanced Notebook: Chronos by @juanitorduz in #2203
• Port TVP Intercept Notebook to MMM Multidimensional API by @timbo112711 in #2207
• Port MMM End-to-End Case Study to MMM Multidimensional API by @juanitorduz in #2191
• Port MMM evaluation notebook to new MMM class by @cetagostini in #2217
• Clarify hierarchical prior specification by @ricardoV94 in #2219
• Bug Fixes for IData Wrapper & Summary by @isofer in #2233
• Migrating: MMM with time-varying media baseline by @cetagostini in #2218
• Migrate MMM Example Notebook by @juanitorduz in #2212
• default sampler - mmm example notebook by @juanitorduz in #2245
• Port README to new MMM class. by @juanitorduz in #2244
• Port MMM Quickstart Notebook to use Multidimensional MMM by @ricardoV94 in #2232
• MMM Migration guide by @juanitorduz in #2254

Maintenance 🔧

• Add Cursor command and rules documentation by @cetagostini in #2132
• Unpin JAX version by @juanitorduz in #2143
• expose priors property to Transformations by @williambdean in #2146
• update headers 2026 by @juanitorduz in #2162
• chore: consolidate component test by @williambdean in #2159
• Docs: clean up Streamlit MMM explainer README by @AD1007 in #2170
• Isofer/idata codification by @isofer in #2176
• Add sample_saturation_curve() Method to MMM by @isofer in #2195
• Add sample_adstock_curve() Method to MMM by @github-actions[bot] in #2198
• Implement MMM Summary functionality by @isofer in #2206
• Small fixes to multidimensional MMM by @ricardoV94 in #2234
• Remove Deprecated classes ContContract, ContNonContract by @naveen28204280 in #2231
• Support MLflow 2.18.0+ API updates by @tibisabau in #2237
• MMM FutureWarning by @juanitorduz in #2256
• Fix SpecialPrior serialization and deterministic model IDs by @PabloRoque in #2258
• Nullify Schema When filter_dims (or Other Functions) Drop Dimensions by @isofer in #2262
• Pin some docs and tests dependencies because of lifetimes by @juanitorduz in #2269
• Interactive Plotly Plotting for MMM - phase 2 by @isofer in #2259

New Contributors

• @JoshTripleWhale made their first contribution in #2120
• @Eshaan-byte made their first contribution in #2130
• @AD1007 made their first contribution in #2170
• @isofer made their first contribution in #2176
• @naveen28204280 made their first contribution in #2231
• @tibisabau made their first contribution in #2237

Full Changelog: 0.17.1...0.18.0

0.17.1

What's Changed

Major Changes 🛠

• Refactor and enhance SensitivityAnalysis for MMM by @cetagostini in #1899

Deprecations 🚨

• Remove Prior classes and functions from pymc_marketing.prior module by @Copilot in #2060

New Features 🎉

• Add binomial adstock by @gdeleva in #2046
• Expose number of points in linspace for SaturationTransformation by @harshasiddartha in #2050
• Add covariate support to ShiftedBetaGeoModel by @ColtAllen in #2080

Bugfixes 🐛

• fix: serialization for arbitrary HSGP instance to time_varying_intercept/media in MMM by @jamespooley in #2016
• Fix ShiftedBetaGeoModel Cohort Bug by @ColtAllen in #2115
• [DOCS] Improve MMM Case Study Notebook by @juanitorduz in #2116

Documentation 📖

• Different corrections in ES files by @cetagostini in #2035
• Add Causal Reasoning and Discovery example to gallery by @cetagostini in #2033
• Update documentation to include weekly Fourier transformation by @juanitorduz in #2045
• minor improvements funel nb by @juanitorduz in #2069
• Correct spelling of 'FiveTran' to 'Fivetran' by @cluhmann in #2084
• Clearer explanation of how lift tests are implemented behind the scenes in existing lift test notebook by @drbenvincent in #2061
• Update Docstrings for ShiftedBetaGeoModel by @ColtAllen in #2091
• Introduction Version of Funnel Analysis Notebok by @juanitorduz in #2070
• add the non-centered laplace to the multi-dim notebook by @daniel-saunders-phil in #2100
• Preparing plots and notebook for old API deprecation by @cetagostini in #2055

Maintenance 🔧

• Enforce original scale variable before CPT calibration by @cetagostini in #2036
• use StrEnum over Enum by @williambdean in #2088
• Add special Laplace prior by @daniel-saunders-phil in #2093
• fix: store off Laplace Prior as Deterministic by @daniel-saunders-phil in #2096
• import bass function from module instead of submodule by @williambdean in #2097
• chore: update UML action by @williambdean in #2112

New Contributors

• @harshasiddartha made their first contribution in #2050
• @jamespooley made their first contribution in #2016
• @clsandoval made their first contribution in #2059

Full Changelog: 0.17.0...0.17.1

0.17.0

What's Changed

Major Changes 🛠

• Add CLV Cohort ShiftedBetaGeoModel by @ColtAllen in #2005

New Features 🎉

• Add BuildModelFromDAG class for causal components by @cetagostini in #1950
• Introduce experimental TBFPC causal discovery algorithms by @cetagostini in #1947
• Add fivetran pandas accessor by @williambdean in #1924
• Add Special priors by @daniel-saunders-phil in #1939
• Add ModelSamplerEstimator for PyMC Marketing model benchmarking by @cetagostini in #1943
• Adds Optional dims arg for subsetting by dimensions when plotting a multidimensional model by @timbo112711 in #1955
• Add calibration support to YAML MMM builder by @cetagostini in #1979
• Add expected_residual_lifetime predictions to ShiftedBetaGeoModel by @ColtAllen in #2010
• Add expected_retention_elasticity to ShiftedBetaGeoModel by @ColtAllen in #2028
• Add GPU support in BudgetOptimizer by @PabloRoque in #1937

Bugfixes 🐛

• BUG [Issue with fourier] by @cetagostini-wise in #1936
• Fix Mypy - HotFix by @juanitorduz in #1963
• Fix np.testing.assert_array_almost_equal argument names by @cetagostini in #1949
• Hotfix - Fix dimension assertion for adding multidimensional event effects to multidimensional MMM by @TeemuSailynoja in #1969
• Hotfix: use target_column in create_fit_data and build_from_idata by @PabloRoque in #1918
• HOT FIX: Update image for sBG Model in gallery by @juanitorduz in #2008
• Solving issue with wrapper detection in optimizer by @cetagostini in #2021
• Fixing fixed_beta bug adding new test for nested_logit by @NathanielF in #2015

Documentation 📖

• Notebook: How to model upper funnel with causal reasoning and PyMC by @cetagostini in #1971
• Notebook: Add multi-objective optimization notebook and update optimizer by @cetagostini in #1973
• correct typo in multidimensional nb by @daniel-saunders-phil in #1919
• Add approximate_fit method for variational inference by @cetagostini in #1923
• Add lognormalprior to the multidimensional notebook by @daniel-saunders-phil in #1961
• Adding calibration by potential based on CPT (Additional calibration style) by @cetagostini in #1959
• Setting infraestructure for Translations (ES site) by @OriolAbril in #1906
• Add MMM Quickstart Guide by @juanitorduz in #1972
• Update notebook risk allocation by @cetagostini in #1982
• Split notebooks test run mmm vs others by @juanitorduz in #1986
• Adding additional translations to the spanish site by @cetagostini in #1984
• Translate to spanish certain notebooks! by @cetagostini in #1995
• Notebook: PyMC-Marketing GAM capabilities by @cetagostini in #1985
• Translate guide folder to spanish by @cetagostini in #1997
• Adding spanish translations to API docs by @cetagostini in #1999
• Finishing all translations to API in Castellano by @cetagostini in #2000
• Notebooks translations to ES by @cetagostini in #2001
• Small error on render notebook tips/warnings/notes by @cetagostini in #2003
• MMM Class Claude Sonnet 4.5 Refactor by @juanitorduz in #1974
• Solving issue in #2006 doing the optimization more obvious not close to the limit. by @cetagostini in #2007
• Updated How We Compare Section by @lfiaschi in #2012
• Notebook: Causal discovery with PyMC-Marketing by @cetagostini in #1994
• Improve documentation links by @juanitorduz in #2025

Maintenance 🔧

• Adjusting calibration by costs to work with sample posterior predictive by @cetagostini in #1988
• Fix ModelBuilder docstring formatting by @williambdean in #1926
• Keep zeroes in add_noise_to_channel_allocation by @PabloRoque in #1916
• Update gamma_control prior dimensions in MMM model by @cetagostini in #1920
• Add model merging utilities and tests for multi-model optimization by @cetagostini in #1944
• Rename LogNormalPositiveParam to LogNormalPrior by @cetagostini in #1948
• Correcting optimizer docstring by @cetagostini in #1962
• Add MaskedPrior class for masked variable creation by @cetagostini in #1945
• Clarify mean_tightness_score docstring and formula by @cetagostini in #1980
• don't use paid defaults channel for dev env by @OriolAbril in #1987
• Translations in the API documents by @cetagostini in #1998
• Last translations by @cetagostini in #2002
• Translations for the new recent merge PRs by @cetagostini in #2020
• Translations by @cetagostini in #2029
• Bump version to 0.17.0 by @juanitorduz in #2030
• Update Spanish translations for CLV and MMM API docs by @cetagostini in #2031

Full Changelog: 0.16.0...0.17.0