Fill radial bars of barplot with gradient

Question

I am trying to generate a radial (circular) barplot displaying various color ranges in order to compare them. Using a radial barplot allows to better apprehend the circular nature of the hue value and to represent out-of-range ranges such as [340°, 10°]

To illustrate with this image, the idea is to replace the black color of the bars of the barplot on the left by the parts of the chromatic circle (displayed on the right) corresponding.

This is the code that I use to generate the radial barplot :

import matplotlib.pyplot as plt
from numpy import pi

start_values = [0, 3*pi/4, pi/6]
range_values = [pi/3, 3*pi/10, pi/2]

ax = plt.subplot(projection='polar')

for i in range(len(start_values)):
    ax.barh(i, range_values[i], left=start_values[i], color=(0, 0, 0))

I wanted to apply Matplotlib's hsv color map but the color parameter of barh() does not accept it.

On the internet I have only found techniques to apply gradients or cmaps to regular barplots, with rectangular bars (like this one: How to fill matplotlib bars with a gradient? However, these techniques do not work with circular bars.

Answer 1

Not the most elegant version but it does the job

import matplotlib.pyplot as plt
from numpy import pi
import colorsys

start_values = [0, 3*pi/4, pi/6]
range_values = [pi/3, 3*pi/10, pi/2]

ax = plt.subplot(projection='polar')

n = 100

for i in range(len(start_values)):
  ax.barh(i, [range_values[i]/n]*n,
          left=[start_values[i]+j*range_values[i]/n for j in range(n)],
          color=[colorsys.hsv_to_rgb((start_values[i]+(j+0.5)*range_values[i]/n)/(2*pi),1,1) for j in range(n)])

plt.show()

output :

Answer 2

Thanks Pierre, I used your technique to build a graph that also represents saturation and value as a gradient in the direction of the radius. Here is what it returns for the example given (with saturation and value ranges added)

Code:

def radius_barplot(lower_color, upper_color, ax, index=None, angle_precision=1, full_color=0.9, plot_index=(1, 1, 1)):
    
    ax = plt.subplot(plot_index[0], plot_index[1], plot_index[2], projection="polar")
    
    #print("Building graph")
    if index is None:
        if isinstance(lower_color[0], pd.Series):
            index = lower_color[0].index
            lower_color = [parameter.tolist() for parameter in lower_color]
            upper_color = [parameter.tolist() for parameter in upper_color]
        else:
            index= range(len(lower_color[0]))
            
    pi2 = 2*pi
    hue_only = 1 - full_color

    for i in range(len(index)):

        #print("Frame n°", i+1, sep='')
        min_hue, max_hue = lower_color[0][i] / 179 * 2 * pi, upper_color[0][i] / 179 * 2 * pi
        min_sat, min_val = lower_color[1][i] / 255, lower_color[2][i] / 255
        range_hue, range_sat, range_val = max_hue - min_hue, round(upper_color[1][i] - lower_color[1][i]), round(upper_color[2][i] - lower_color[2][i])
        n_height = min(max(range_sat, range_val), 20)
        pas_height = 1*full_color/n_height
        large_pas_angle, large_pas_height = angle_precision*1.5/180*pi, pas_height*1.25
        pas_sat, pas_val = range_sat / 255 / n_height, range_val / 255 / n_height

        #print("Angle range: [", round(min_hue/pi*180), "; ", round((max_hue)/pi*180), "]", sep='')
        
        # For each degree of the angle
        for angle in np.arange(min_hue, max_hue, angle_precision/180*pi):
            
            # To avoid an overflow on the end of the interval
            if angle + large_pas_angle > max_hue: large_pas_angle = max_hue - angle
                
            # Keep a positive angle for the hue value
            if angle < 0: angle = 2*pi + angle
                        
            # For each line inside the current bar
            if full_color > 0:
                for height in range(n_height):
                    """
                    print("Height: [", round(i + height*pas_height, 2), ", ", round(i + height*pas_height + pas_height, 2),
                          "], angle: ", round(angle/pi*180, 3),
                          "°: color [", round(angle*360), ", ", round((min_sat + height*pas_sat)*100), ", ", round((min_val + height*pas_val)*100), "]", sep='')
                    """
                
                    ax.barh(y=i + height*pas_height,
                            width=large_pas_angle, left=angle, height=large_pas_height, align='edge',
                            color=colorsys.hsv_to_rgb(angle/pi2,
                                                      min_sat + height*pas_sat,
                                                      min_val + height*pas_val)
                    )
            
            
            # To display the reference hue (with full saturation and value)
            ax.barh(y=i + full_color,
                    width=large_pas_angle, left=angle, height=hue_only, align='edge',
                    color=(colorsys.hsv_to_rgb(angle/pi2, 1, 1)))
            # """
        ax.set_rgrids(range(len(index)), labels=index)

radius_barplot(
    ((0, (3*pi/4)/(2*pi)*179, (pi/6)/(2*pi)*179), (0, 200, 50), (0, 50, 200)),  # Minimal values of colors ranges
    (((pi/3)/(2*pi)*179, (21*pi/20)/(2*pi)*179, (2*pi/3)/(2*pi)*179), (255, 250, 100), (255, 100, 250)),  # Maximal values of colors ranges
    axes)

Graph:

The main problem of this technique is its complexity: returning this graph took 1 minute and 45 seconds on my computer and getting the graph I am interested in takes almost 5 minutes.

So if someone has a better, more optimal method to obtain a similar result I would still be interested.