fix pair_plot for mixed discrete and continuous variables

CHANGELOG.md

@@ -18,6 +18,7 @@
 * Have `from_pystan` store attrs as strings to allow netCDF storage ([1417](https://github.com/arviz-devs/arviz/pull/1417))
 * Remove ticks and spines in `plot_violin`  ([1426 ](https://github.com/arviz-devs/arviz/pull/1426))
 * Use circular KDE function and fix tick labels in circular `plot_trace` ([1428](https://github.com/arviz-devs/arviz/pull/1428))
+*  Fix `pair_plot` for mixed discrete and continuous variables ([1434](https://github.com/arviz-devs/arviz/pull/1434))
 
 ### Deprecation
 

arviz/plots/backends/bokeh/pairplot.py

@@ -9,14 +9,14 @@
 from ....rcparams import rcParams
 from ...distplot import plot_dist
 from ...kdeplot import plot_kde
-from ...plot_utils import _scale_fig_size, calculate_point_estimate
+from ...plot_utils import _scale_fig_size, calculate_point_estimate, vectorized_to_hex
 from .. import show_layout
 from . import backend_kwarg_defaults
 
 
 def plot_pair(
     ax,
-    infdata_group,
+    plotters,
     numvars,
     figsize,
     textsize,
@@ -28,7 +28,7 @@ def plot_pair(
     colorbar,  # pylint: disable=unused-argument
     divergences,
     diverging_mask,
-    divergences_kwargs,  # pylint: disable=unused-argument
+    divergences_kwargs,
     flat_var_names,
     backend_kwargs,
     marginal_kwargs,
@@ -123,8 +123,17 @@ def plot_pair(
     if reference_values_kwargs is None:
         reference_values_kwargs = {}
 
-    reference_values_kwargs.setdefault("line_color", "red")
-    reference_values_kwargs.setdefault("line_width", 5)
+    reference_values_kwargs.setdefault("line_color", "black")
+    reference_values_kwargs.setdefault("fill_color", vectorized_to_hex("C2"))
+    reference_values_kwargs.setdefault("line_width", 1)
+    reference_values_kwargs.setdefault("size", 10)
+
+    if divergences_kwargs is None:
+        divergences_kwargs = {}
+    divergences_kwargs.setdefault("line_color", "black")
+    divergences_kwargs.setdefault("fill_color", vectorized_to_hex("C1"))
+    divergences_kwargs.setdefault("line_width", 1)
+    divergences_kwargs.setdefault("size", 10)
 
     dpi = backend_kwargs.pop("dpi")
     max_plots = (
@@ -151,17 +160,18 @@ def plot_pair(
     if point_estimate_marker_kwargs is None:
         point_estimate_marker_kwargs = {}
 
-    point_estimate_marker_kwargs.setdefault("line_width", markersize)
-    point_estimate_kwargs.setdefault("line_color", "orange")
-    point_estimate_kwargs.setdefault("line_width", 3)
+    point_estimate_marker_kwargs.setdefault("size", markersize)
+    point_estimate_marker_kwargs.setdefault("color", "black")
+    point_estimate_kwargs.setdefault("line_color", "black")
+    point_estimate_kwargs.setdefault("line_width", 2)
     point_estimate_kwargs.setdefault("line_dash", "solid")
 
     tmp_flat_var_names = None
     if len(flat_var_names) == len(list(set(flat_var_names))):
-        source_dict = dict(zip(flat_var_names, [list(post) for post in infdata_group]))
+        source_dict = dict(zip(flat_var_names, [list(post[-1].flatten()) for post in plotters]))
     else:
         tmp_flat_var_names = ["{}__{}".format(name, str(uuid4())) for name in flat_var_names]
-        source_dict = dict(zip(tmp_flat_var_names, [list(post) for post in infdata_group]))
+        source_dict = dict(zip(tmp_flat_var_names, [list(post[-1].flatten()) for post in plotters]))
     if divergences:
         divergenve_name = "divergences_{}".format(str(uuid4()))
         source_dict[divergenve_name] = np.array(diverging_mask).astype(bool).astype(int).astype(str)
@@ -250,7 +260,7 @@ def get_width_and_height(jointplot, rotate):
 
             if j == i and marginals:
                 rotate = numvars == 2 and j == 1
-                var1_dist = infdata_group[i]
+                var1_dist = plotters[i][-1].flatten()
                 plot_dist(
                     var1_dist,
                     ax=ax[j, i],
@@ -272,8 +282,8 @@ def get_width_and_height(jointplot, rotate):
                         ax[j, i].circle(var1, var2, source=source)
 
                 if "kde" in kind:
-                    var1_kde = infdata_group[i]
-                    var2_kde = infdata_group[j + marginals_offset]
+                    var1_kde = plotters[i][-1].flatten()
+                    var2_kde = plotters[j + marginals_offset][-1].flatten()
                     plot_kde(
                         var1_kde,
                         var2_kde,
@@ -285,8 +295,8 @@ def get_width_and_height(jointplot, rotate):
                     )
 
                 if "hexbin" in kind:
-                    var1_hexbin = infdata_group[i]
-                    var2_hexbin = infdata_group[j + marginals_offset]
+                    var1_hexbin = plotters[i][-1].flatten()
+                    var2_hexbin = plotters[j + marginals_offset][-1].flatten()
                     ax[j, i].grid.visible = False
                     ax[j, i].hexbin(
                         var1_hexbin,
@@ -298,17 +308,14 @@ def get_width_and_height(jointplot, rotate):
                     ax[j, i].circle(
                         var1,
                         var2,
-                        line_color="black",
-                        fill_color="orange",
-                        line_width=1,
-                        size=10,
                         source=source,
                         view=source_div,
+                        **divergences_kwargs,
                     )
 
                 if point_estimate:
-                    var1_pe = infdata_group[i]
-                    var2_pe = infdata_group[j]
+                    var1_pe = plotters[i][-1].flatten()
+                    var2_pe = plotters[j][-1].flatten()
                     pe_x = calculate_point_estimate(point_estimate, var1_pe)
                     pe_y = calculate_point_estimate(point_estimate, var2_pe)
                     ax[j, i].square(pe_x, pe_y, **point_estimate_marker_kwargs)
@@ -330,7 +337,7 @@ def get_width_and_height(jointplot, rotate):
 
                         ax[j - 1, i].add_layout(ax_vline)
 
-                        pe_last = calculate_point_estimate(point_estimate, infdata_group[-1])
+                        pe_last = calculate_point_estimate(point_estimate, plotters[-1][-1])
                         ax_pe_vline = Span(
                             location=pe_last,
                             dimension="height",

arviz/plots/backends/matplotlib/pairplot.py

@@ -15,7 +15,7 @@
 
 def plot_pair(
     ax,
-    infdata_group,
+    plotters,
     numvars,
     figsize,
     textsize,
@@ -73,6 +73,7 @@ def plot_pair(
         marginal_kwargs = {}
 
     point_estimate_kwargs = matplotlib_kwarg_dealiaser(point_estimate_kwargs, "fill_between")
+    point_estimate_kwargs.setdefault("color", "k")
 
     if kind != "kde":
         kde_kwargs.setdefault("contourf_kwargs", {})
@@ -105,14 +106,15 @@ def plot_pair(
 
     reference_values_kwargs = matplotlib_kwarg_dealiaser(reference_values_kwargs, "plot")
 
-    reference_values_kwargs.setdefault("color", "C3")
+    reference_values_kwargs.setdefault("color", "C2")
+    reference_values_kwargs.setdefault("markeredgecolor", "k")
     reference_values_kwargs.setdefault("marker", "o")
 
     point_estimate_marker_kwargs = matplotlib_kwarg_dealiaser(
         point_estimate_marker_kwargs, "scatter"
     )
     point_estimate_marker_kwargs.setdefault("marker", "s")
-    point_estimate_marker_kwargs.setdefault("color", "C1")
+    point_estimate_marker_kwargs.setdefault("color", "k")
 
     # pylint: disable=too-many-nested-blocks
     if numvars == 2:
@@ -127,8 +129,8 @@ def plot_pair(
         point_estimate_marker_kwargs.setdefault("s", markersize + 50)
 
         # Flatten data
-        x = infdata_group[0].flatten()
-        y = infdata_group[1].flatten()
+        x = plotters[0][-1].flatten()
+        y = plotters[1][-1].flatten()
         if ax is None:
             if marginals:
                 # Instantiate figure and grid
@@ -177,13 +179,13 @@ def plot_pair(
                 ax = np.atleast_2d(ax)[0, 0]
 
         if "scatter" in kind:
-            ax.plot(infdata_group[0], infdata_group[1], **scatter_kwargs)
+            ax.plot(x, y, **scatter_kwargs)
         if "kde" in kind:
-            plot_kde(infdata_group[0], infdata_group[1], ax=ax, **kde_kwargs)
+            plot_kde(x, y, ax=ax, **kde_kwargs)
         if "hexbin" in kind:
             hexbin = ax.hexbin(
-                infdata_group[0],
-                infdata_group[1],
+                x,
+                y,
                 gridsize=gridsize,
                 **hexbin_kwargs,
             )
@@ -195,8 +197,8 @@ def plot_pair(
 
         if divergences:
             ax.plot(
-                infdata_group[0][diverging_mask],
-                infdata_group[1][diverging_mask],
+                x[diverging_mask],
+                y[diverging_mask],
                 **divergences_kwargs,
             )
 
@@ -256,10 +258,10 @@ def plot_pair(
             )
         hexbin_values = []
         for i in range(0, vars_to_plot):
-            var1 = infdata_group[i]
+            var1 = plotters[i][-1].flatten()
 
             for j in range(0, vars_to_plot):
-                var2 = infdata_group[j + not_marginals]
+                var2 = plotters[j + not_marginals][-1].flatten()
                 if i > j:
                     if ax[j, i].get_figure() is not None:
                         ax[j, i].remove()
@@ -312,7 +314,7 @@ def plot_pair(
 
                         if marginals:
                             ax[j - 1, i].axvline(pe_x, **point_estimate_kwargs)
-                            pe_last = calculate_point_estimate(point_estimate, infdata_group[-1])
+                            pe_last = calculate_point_estimate(point_estimate, plotters[-1][-1])
                             ax[-1, -1].axvline(pe_last, **point_estimate_kwargs)
 
                         ax[j, i].scatter(pe_x, pe_y, **point_estimate_marker_kwargs)

arviz/plots/pairplot.py

@@ -4,10 +4,15 @@
 
 import numpy as np
 
-from ..data import convert_to_dataset, convert_to_inference_data
+from ..data import convert_to_dataset
 from ..rcparams import rcParams
 from ..utils import _var_names, get_coords
-from .plot_utils import get_plotting_function, xarray_to_ndarray
+from .plot_utils import (
+    get_plotting_function,
+    xarray_to_ndarray,
+    filter_plotters_list,
+    xarray_var_iter,
+)
 
 
 def plot_pair(
@@ -187,12 +192,13 @@ def plot_pair(
         coords = {}
 
     # Get posterior draws and combine chains
-    data = convert_to_inference_data(data)
-    grouped_data = convert_to_dataset(data, group=group)
-    var_names = _var_names(var_names, grouped_data, filter_vars)
-    flat_var_names, infdata_group = xarray_to_ndarray(
-        get_coords(grouped_data, coords), var_names=var_names, combined=True
+    dataset = convert_to_dataset(data, group=group)
+    var_names = _var_names(var_names, dataset, filter_vars)
+    plotters = filter_plotters_list(
+        list(xarray_var_iter(get_coords(dataset, coords), var_names=var_names, combined=True)),
+        "plot_pair",
     )
+    flat_var_names = [plotter[0] for plotter in plotters]
 
     divergent_data = None
     diverging_mask = None
@@ -224,7 +230,7 @@ def plot_pair(
             )
 
     if gridsize == "auto":
-        gridsize = int(len(infdata_group[0]) ** 0.35)
+        gridsize = int(dataset.dims["draw"] ** 0.35)
 
     numvars = len(flat_var_names)
 
@@ -233,7 +239,7 @@ def plot_pair(
 
     pairplot_kwargs = dict(
         ax=ax,
-        infdata_group=infdata_group,
+        plotters=plotters,
         numvars=numvars,
         figsize=figsize,
         textsize=textsize,