Added support for categorical colormapping in bokeh backend

holoviews/plotting/bokeh/chart.py

@@ -58,10 +58,18 @@ def get_data(self, element, ranges=None, empty=False):
         mapping = dict(x=dims[xidx], y=dims[yidx])
         data = {}
 
+        xdim, ydim = dims[xidx], dims[yidx]
+        data[xdim] = [] if empty else element.dimension_values(xidx)
+        data[ydim] = [] if empty else element.dimension_values(yidx)
+        self._categorize_data(data, (xdim, ydim), element.dimensions())
+
         cdim = element.get_dimension(self.color_index)
         if cdim:
-            mapper = self._get_colormapper(cdim, element, ranges, style)
-            data[cdim.name] = [] if empty else element.dimension_values(cdim)
+            cdata = data[cdim.name] if cdim.name in data else element.dimension_values(cdim)
+            factors = list(np.unique(cdata)) if cdata.dtype.kind in 'OSU' else None
+            mapper = self._get_colormapper(cdim, element, ranges, style,
+                                           factors)
+            data[cdim.name] = cdata
             mapping['color'] = {'field': cdim.name,
                                 'transform': mapper}
 
@@ -85,10 +93,6 @@ def get_data(self, element, ranges=None, empty=False):
                     data[map_key] = np.sqrt(sizes)
                     mapping['size'] = map_key
 
-        xdim, ydim = dims[xidx], dims[yidx]
-        data[xdim] = [] if empty else element.dimension_values(xidx)
-        data[ydim] = [] if empty else element.dimension_values(yidx)
-        self._categorize_data(data, (xdim, ydim), element.dimensions())
         self._get_hover_data(data, element, empty)
         return data, mapping
 

holoviews/plotting/bokeh/element.py

@@ -12,7 +12,7 @@
 from bokeh.models.mappers import LinearColorMapper
 try:
     from bokeh.models import ColorBar
-    from bokeh.models.mappers import LogColorMapper
+    from bokeh.models.mappers import LogColorMapper, CategoricalColorMapper
 except ImportError:
     LogColorMapper, ColorBar = None, None
 from bokeh.models import LogTicker, BasicTicker
@@ -870,13 +870,11 @@ def _draw_colorbar(self, plot, color_mapper):
         self.handles['colorbar'] = color_bar
 
 
-    def _get_colormapper(self, dim, element, ranges, style):
+    def _get_colormapper(self, dim, element, ranges, style, factors=None):
         # The initial colormapper instance is cached the first time
         # and then only updated
         if dim is None:
             return None
-        low, high = ranges.get(dim.name, element.range(dim.name))
-        palette = mplcmap_to_palette(style.pop('cmap', 'viridis'))
         if self.adjoined:
             cmappers = self.adjoined.traverse(lambda x: (x.handles.get('color_dim'),
                                                          x.handles.get('color_mapper')))
@@ -887,25 +885,34 @@ def _get_colormapper(self, dim, element, ranges, style):
                 return cmapper
             else:
                 return None
-        colors = self.clipping_colors
-        if isinstance(low, (bool, np.bool_)): low = int(low)
-        if isinstance(high, (bool, np.bool_)): high = int(high)
-        opts = {'low': low, 'high': high}
-        color_opts = [('NaN', 'nan_color'), ('max', 'high_color'), ('min', 'low_color')]
-        for name, opt in color_opts:
-            color = colors.get(name)
-            if not color:
-                continue
-            elif isinstance(color, tuple):
-                color = [int(c*255) if i<3 else c for i, c in enumerate(color)]
-            opts[opt] = color
+
+        ncolors = None if factors is None else len(factors)
+        low, high = ranges.get(dim.name, element.range(dim.name))
+        palette = mplcmap_to_palette(style.pop('cmap', 'viridis'), ncolors)
+
+        color_rgb = lambda color: [int(c*255) if i<3 else c for i, c in enumerate(color)]
+        colors = {k: color_rgb(v) if isinstance(v, tuple) else v
+                  for k, v in self.clipping_colors}
+
+        if factors is None:
+            colormapper = LogColorMapper if self.logz else LinearColorMapper
+            if isinstance(low, (bool, np.bool_)): low = int(low)
+            if isinstance(high, (bool, np.bool_)): high = int(high)
+            opts = {'low': low, 'high': high}
+            color_opts = [('NaN', 'nan_color'), ('max', 'high_color'), ('min', 'low_color')]
+            opts.update({opt: colors[name] for name, opt in color_opts if name in colors})
+        else:
+            colormapper = CategoricalColorMapper
+            opts = dict(factors=factors)
+            if 'NaN' in colors:
+                opts['nan_color'] = colors['NaN']
+
         if 'color_mapper' in self.handles:
             cmapper = self.handles['color_mapper']
             cmapper.palette = palette
             cmapper.update(**opts)
         else:
-            colormapper = LogColorMapper if self.logz else LinearColorMapper
-            cmapper = colormapper(palette, **opts)
+            cmapper = colormapper(palette=palette, **opts)
             self.handles['color_mapper'] = cmapper
             self.handles['color_dim'] = dim
         return cmapper

holoviews/plotting/bokeh/util.py

@@ -65,14 +65,16 @@ def rgb2hex(rgb):
     return "#{0:02x}{1:02x}{2:02x}".format(*(int(v*255) for v in rgb))
 
 
-def mplcmap_to_palette(cmap):
+def mplcmap_to_palette(cmap, ncolors=None):
     """
     Converts a matplotlib colormap to palette of RGB hex strings."
     """
     if colors is None:
         raise ValueError("Using cmaps on objects requires matplotlib.")
     with abbreviated_exception():
         colormap = cm.get_cmap(cmap) #choose any matplotlib colormap here
+        if ncolors:
+            return [rgb2hex(colormap(i)) for i in np.linspace(0, 1, ncolors)]
         return [rgb2hex(m) for m in colormap(np.arange(colormap.N))]