Enhance plot_pdp and fix plot_scatter_submodels (#218)

* Add YML env files

* Expand scatter_submodels to categorical likelihood

* Add softmax option to plot_pdp

* Remove comments

* Use func for softmax

* handle func upstream

* move func upstream

* ensure p_d is an array

---------

Co-authored-by: aloctavodia <aloctavodia@gmail.com>

Author: AlexAndorra

env-dev.yml

@@ -0,0 +1,23 @@
+name: pymc-bart-dev
+channels:
+  - conda-forge
+  - defaults
+dependencies:
+  - pymc>=5.16.2,<=5.19.1
+  - arviz>=0.18.0
+  - numba
+  - matplotlib
+  - numpy
+  - pytensor
+  # Development dependencies
+  - pytest>=4.4.0
+  - pytest-cov>=2.6.1
+  - click==8.0.4
+  - pylint==2.17.4
+  - pre-commit
+  - black
+  - isort
+  - flake8
+  - pip
+  - pip:
+    - -e .

env.yml

@@ -0,0 +1,14 @@
+name: pymc-bart
+channels:
+  - conda-forge
+  - defaults
+dependencies:
+  - pymc>=5.16.2,<=5.19.1
+  - arviz>=0.18.0
+  - numba
+  - matplotlib
+  - numpy
+  - pytensor
+  - pip
+  - pip:
+    - pymc-bart

pymc_bart/utils.py

@@ -254,13 +254,13 @@ def identity(x):
             )
 
         new_x = fake_X[:, var]
-        p_d = np.array(y_pred)
+        p_d = func(np.array(y_pred))
 
         for s_i in range(shape):
             if centered:
-                p_di = func(p_d[:, :, s_i]) - func(p_d[:, :, s_i][:, 0][:, None])
+                p_di = p_d[:, :, s_i] - p_d[:, :, s_i][:, 0][:, None]
             else:
-                p_di = func(p_d[:, :, s_i])
+                p_di = p_d[:, :, s_i]
             if var in var_discrete:
                 axes[count].plot(new_x, p_di.mean(0), "o", color=color_mean)
                 axes[count].plot(new_x, p_di.T, ".", color=color, alpha=alpha)
@@ -393,14 +393,17 @@ def identity(x):
     for var in range(len(var_idx)):
         excluded = indices[:]
         excluded.remove(var)
-        p_d = _sample_posterior(
-            all_trees, X=fake_X, rng=rng, size=samples, excluded=excluded, shape=shape
+        p_d = func(
+            _sample_posterior(
+                all_trees, X=fake_X, rng=rng, size=samples, excluded=excluded, shape=shape
+            )
         )
+
         with warnings.catch_warnings():
             warnings.filterwarnings("ignore", message="hdi currently interprets 2d data")
             new_x = fake_X[:, var]
             for s_i in range(shape):
-                p_di = func(p_d[:, :, s_i])
+                p_di = p_d[:, :, s_i]
                 null_pd.append(p_di.mean())
                 if var in var_discrete:
                     _, idx_uni = np.unique(new_x, return_index=True)
@@ -1125,8 +1128,11 @@ def plot_scatter_submodels(
     plot_kwargs : dict
         Additional keyword arguments for the plot. Defaults to None.
         Valid keys are:
-        - color_ref: matplotlib valid color for the 45 degree line
+        - marker_scatter: matplotlib valid marker for the scatter plot
         - color_scatter: matplotlib valid color for the scatter plot
+        - alpha_scatter: matplotlib valid alpha for the scatter plot
+        - color_ref: matplotlib valid color for the 45 degree line
+        - ls_ref: matplotlib valid linestyle for the reference line
     axes : axes
         Matplotlib axes.
 
@@ -1140,41 +1146,69 @@ def plot_scatter_submodels(
         submodels = np.sort(submodels)
 
     indices = vi_results["indices"][submodels]
-    preds = vi_results["preds"][submodels]
+    preds_sub = vi_results["preds"][submodels]
     preds_all = vi_results["preds_all"]
 
+    if labels is None:
+        labels = vi_results["labels"][submodels]
+
+    # handle categorical regression case:
+    n_cats = None
+    if preds_all.ndim > 2:
+        n_cats = preds_all.shape[-1]
+        indices = np.tile(indices, n_cats)
+
     if ax is None:
         _, ax = _get_axes(grid, len(indices), True, True, figsize)
 
     if plot_kwargs is None:
         plot_kwargs = {}
 
-    if labels is None:
-        labels = vi_results["labels"][submodels]
-
     if func is not None:
-        preds = func(preds)
+        preds_sub = func(preds_sub)
         preds_all = func(preds_all)
 
-    min_ = min(np.min(preds), np.min(preds_all))
-    max_ = max(np.max(preds), np.max(preds_all))
-
-    for pred, x_label, axi in zip(preds, labels, ax.ravel()):
-        axi.plot(
-            pred,
-            preds_all,
-            marker=plot_kwargs.get("marker_scatter", "."),
-            ls="",
-            color=plot_kwargs.get("color_scatter", "C0"),
-            alpha=plot_kwargs.get("alpha_scatter", 0.1),
-        )
-        axi.set_xlabel(x_label)
-        axi.axline(
-            [min_, min_],
-            [max_, max_],
-            color=plot_kwargs.get("color_ref", "0.5"),
-            ls=plot_kwargs.get("ls_ref", "--"),
-        )
+    min_ = min(np.min(preds_sub), np.min(preds_all))
+    max_ = max(np.max(preds_sub), np.max(preds_all))
+
+    # handle categorical regression case:
+    if n_cats is not None:
+        i = 0
+        for cat in range(n_cats):
+            for pred_sub, x_label in zip(preds_sub, labels):
+                ax[i].plot(
+                    pred_sub[..., cat],
+                    preds_all[..., cat],
+                    marker=plot_kwargs.get("marker_scatter", "."),
+                    ls="",
+                    color=plot_kwargs.get("color_scatter", f"C{cat}"),
+                    alpha=plot_kwargs.get("alpha_scatter", 0.1),
+                )
+                ax[i].set(xlabel=x_label, ylabel="ref model", title=f"Category {cat}")
+                ax[i].axline(
+                    [min_, min_],
+                    [max_, max_],
+                    color=plot_kwargs.get("color_ref", "0.5"),
+                    ls=plot_kwargs.get("ls_ref", "--"),
+                )
+                i += 1
+    else:
+        for pred_sub, x_label, axi in zip(preds_sub, labels, ax.ravel()):
+            axi.plot(
+                pred_sub,
+                preds_all,
+                marker=plot_kwargs.get("marker_scatter", "."),
+                ls="",
+                color=plot_kwargs.get("color_scatter", "C0"),
+                alpha=plot_kwargs.get("alpha_scatter", 0.1),
+            )
+            axi.set(xlabel=x_label, ylabel="ref model")
+            axi.axline(
+                [min_, min_],
+                [max_, max_],
+                color=plot_kwargs.get("color_ref", "0.5"),
+                ls=plot_kwargs.get("ls_ref", "--"),
+            )
     return ax