Updating MMM Budget allocation examples, functionalities and dependencies (#1849)

* initial commit

* Update multidimensional MMM example and utils

Refactored the multidimensional MMM budget allocation notebook to improve date handling, add xarray usage, and update sample response distribution parameters. Added a new multidimensional_model.nc file, updated utility functions, and adjusted related tests for compatibility with new data structures and outputs. Also modified .gitignore to allow tracking of .nc files.

* deprecation warning

* Solving tests

* Pushing test with error.

* Solving test discrepancies between pytensor and sample posterior predictive

* Adding new tests

* Add multidimensional_model.nc with Git LFS tracking

* Solving issues with tests

* Changes

* remove repetitive front matter.

* We don't need to explain about the model builder in this one. People will get that they can load a pre-trained notebook and other parts of the docs can explain how it works. Goal is to get to optimziation as quickly as possible.

* remove discussion of other saturation functions. It doesn't add much to the optimization conversation and other parts of the docs should compare the saturation functions. We're assuming the reader already has read the intro notebook so they at least now that saturation functions have diminishing returns, etc.

* minor typos.

* Adding missing tests suites

---------

Co-authored-by: daniel-saunders-phil <[email protected]>

Author: cetagostini

.gitattributes

@@ -0,0 +1 @@
+docs/source/notebooks/mmm/multidimensional_model.nc filter=lfs diff=lfs merge=lfs -text

data/config_files/multi_dimensional_example_model.yml

@@ -0,0 +1,124 @@
+model:
+  class: pymc_marketing.mmm.multidimensional.MMM
+  kwargs:
+    date_column: "date"
+    target_column: "y"
+    channel_columns: ["x1", "x2"]
+    control_columns: ["event_1", "event_2"]
+    dims: ["geo"]
+    yearly_seasonality: 2
+    scaling:
+      class: pymc_marketing.mmm.scaling.Scaling
+      kwargs:
+        channel:
+          method: "max"
+          dims: []
+        target:
+          method: "max"
+          dims: []
+
+    # --- media transformations ---------------------------------------
+    adstock:
+      class: pymc_marketing.mmm.GeometricAdstock
+      kwargs:
+        priors:
+          alpha:
+            distribution: "Beta"
+            alpha: 2
+            beta: 3
+            dims: "channel"
+        l_max: 8
+
+    saturation:
+      class: pymc_marketing.mmm.LogisticSaturation
+      kwargs:
+        priors:
+          beta:
+            distribution: "Gamma"
+            mu: [0.35, 0.35]
+            sigma: [0.1, 0.1]
+            dims: "channel"
+          lam:
+            distribution: "Gamma"
+            mu: 3
+            sigma: 2
+            dims: "channel"
+
+
+    # --- model (hierarchical) priors ---------------------------------
+    model_config:
+      intercept:
+        distribution: Normal
+        mu: 0.5
+        sigma: 0.5
+        dims: geo
+
+      gamma_control:
+        distribution: Normal
+        mu: 0
+        sigma: 0.5
+        dims: control
+
+      gamma_fourier:
+        distribution: Laplace
+        mu: 0
+        b:
+          distribution: HalfNormal
+          sigma: 0.2
+        dims: [geo, fourier_mode]
+
+      likelihood:
+        distribution: TruncatedNormal
+        lower: 0
+        sigma:
+          distribution: HalfNormal
+          sigma:
+            distribution: HalfNormal
+            sigma: 1.5
+        dims: [date, geo]
+
+# ----------------------------------------------------------------------
+# Effects with complex priors
+effects:
+  - class: pymc_marketing.mmm.additive_effect.LinearTrendEffect
+    kwargs:
+      trend:
+        class: pymc_marketing.mmm.LinearTrend
+        kwargs:
+          n_changepoints: 5
+          include_intercept: false
+          dims: ["geo"]  # Keep as array format
+          priors:
+            delta:
+              distribution: "Laplace"
+              mu: 0
+              b:
+                distribution: "HalfNormal"
+                sigma: 0.2
+              dims: ["changepoint", "geo"]
+      prefix: "trend"
+
+# ----------------------------------------------------------------------
+# (optional) sampler options you plan to forward to pm.sample():
+    sampler_config:
+      tune: 1000
+      draws: 200
+      chains: 8
+      random_seed: 42
+      target_accept: 0.90
+      nuts_sampler: "nutpie"
+
+# ----------------------------------------------------------------------
+# (optional) idata from a previous sample
+idata_path: "multidimensional_model.nc"
+
+original_scale_vars:
+  - channel_contribution
+  - intercept_contribution
+  - y
+
+# ----------------------------------------------------------------------
+# (optional) Data paths
+# data:
+#   X_path: "data/X.csv"
+#   y_path: "data/y.csv"