How to overlay seaborn heatmap on matplotlib figure

Question

I am trying to overlay a heatmap on top of a matplotlib figure of a football pitch.

This is the image of the matplotlib pitch created by the code block below:


import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import matplotlib.patches as plt_p
import numpy as np

def draw_pitch(ax):
    # size of the pitch is 120, 80
    #Create figure

    #Pitch Outline & Centre Line
    plt.plot([0,0],[0,80], color="black")
    plt.plot([0,120],[80,80], color="black")
    plt.plot([120,120],[80,0], color="black")
    plt.plot([120,0],[0,0], color="black")
    plt.plot([60,60],[0,80], color="black")

    #Left Penalty Area
    plt.plot([14.6,14.6],[57.8,22.2],color="black")
    plt.plot([0,14.6],[57.8,57.8],color="black")
    plt.plot([0,14.6],[22.2,22.2],color="black")

    #Right Penalty Area
    plt.plot([120,105.4],[57.8,57.8],color="black")
    plt.plot([105.4,105.4],[57.8,22.5],color="black")
    plt.plot([120, 105.4],[22.5,22.5],color="black")

    #Left 6-yard Box
    plt.plot([0,4.9],[48,48],color="black")
    plt.plot([4.9,4.9],[48,32],color="black")
    plt.plot([0,4.9],[32,32],color="black")

    #Right 6-yard Box
    plt.plot([120,115.1],[48,48],color="black")
    plt.plot([115.1,115.1],[48,32],color="black")
    plt.plot([120,115.1],[32,32],color="black")

    #Prepare Circles
    centreCircle = plt.Circle((60,40),8.1,color="black",fill=False)
    centreSpot = plt.Circle((60,40),0.71,color="black")
    leftPenSpot = plt.Circle((9.7,40),0.71,color="black")
    rightPenSpot = plt.Circle((110.3,40),0.71,color="black")

    #Draw Circles
    ax.add_patch(centreCircle)
    ax.add_patch(centreSpot)
    ax.add_patch(leftPenSpot)
    ax.add_patch(rightPenSpot)

    #Prepare Arcs
    # arguments for arc
    # x, y coordinate of centerpoint of arc
    # width, height as arc might not be circle, but oval
    # angle: degree of rotation of the shape, anti-clockwise
    # theta1, theta2, start and end location of arc in degree
    leftArc = plt_p.Arc((9.7,40),height=16.2,width=16.2,angle=0,theta1=310,theta2=50,color="black")
    rightArc = plt_p.Arc((110.3,40),height=16.2,width=16.2,angle=0,theta1=130,theta2=230,color="black")

    #Draw Arcs
    ax.add_patch(leftArc)
    ax.add_patch(rightArc)

fig=plt.figure()
fig.set_size_inches(7, 5)
ax=fig.add_subplot(1,1,1)
draw_pitch(ax)
plt.axis('off')
plt.show()

As recommended on previous posts, I have tried to pass in the ax argument into sns.heatmap() and change the alpha to increase the transparency of the heatmap. However, the heatmap still covers the entire figure and football pitch is not visible.

When running the below code, I get the following result:

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import matplotlib.patches as plt_p
import numpy as np

#DUMMY DATA 
df_test = pd.DataFrame(np.array([[43.2, 22.4, 0], [-5.1,-53.2,1], [33.5,-19.2,0],
                                 [23.2, 32.4, 1], [-5.3,-53.2,1], [33.5,-69.2,0],
                                 [53.2, -42.4, 0], [-5.4,-53.2,0], [-3.5,-39.2,0],
                                 [63.2, 62.4, 1], [-52,-53.2,0], [37.5,-11.2,1],
                                 [113.2, 72.4, 0], [-34.2,-53.2,0], [42.5,-119.2,1]]),
                  columns=['x', 'y', 'outcome'])

#CREATES THE HEATMAP OVERLAY ON THE FOOTBALL PITCH
def pass_comp_map(df):
    df['x_bands'] = pd.qcut(df['x'],4,labels=False)
    df['y_bands'] = pd.qcut(df['y'],3,labels=False)
    df_pass = df[['x_bands','y_bands','outcome']]
    df_sum = df_pass.groupby(['x_bands','y_bands'], as_index=False).sum() # get total number of completed passes 
    df_count = df_pass.groupby(['x_bands','y_bands'], as_index=False).count() #get total number passes
    df_agg = pd.merge(df_sum, df_count['outcome'].to_frame(), how ='left',left_index=True,right_index=True)
    df_agg['pass_comp'] = df_agg['outcome_x'] / df_agg['outcome_y']
    data = df_agg[['x_bands','y_bands','pass_comp']]
    data_pivot = data.pivot_table(index='y_bands', columns='x_bands', values='pass_comp')
    data_pivot = data_pivot.fillna(0)

    #OVERLAY FIGIURE CREATED HERE
    fig=plt.figure()
    fig.set_size_inches(7, 5)
    ax=fig.add_subplot(1,1,1)
    draw_pitch(ax)
    plt.axis('off')
    sns.heatmap(data_pivot,cbar=False, xticklabels=False, yticklabels=False,annot=True,alpha = 0.5,ax=ax)
    plt.show()

pass_comp_map(df_test)

How do I make it such that the heatmap is overlayed on the football pitch, but the football pitch is still visible?

I would suggest you to familiarize yourself with the different abstraction levels at which you can interact with matplotlib (e.g. [here](https://stackoverflow.com/questions/43482191/matplotlib-axes-plot-vs-pyplot-plot)). When you want to work with overlays it is crucial to understand and have full control over the axes in your figure — Florian Drawitsch, Feb 06 '19 at 13:05
In short: Don't use a seaborn heatmap. Instead use matplotlib's `imshow` or `pcolormesh` instead. They allow you to define the scales of the plot. Then choose the same scale as your football field. — ImportanceOfBeingErnest, Feb 06 '19 at 15:27

score 5 · Answer 1 · answered Feb 06 '19 at 17:35

As commented already, I would recommend not to use a seaborn.heatmap, because it's pretty much impossible to scale it to the size of the field.

If you make no further changes to your code, just replace the line sns.heatmap(...) by

ax.imshow(data_pivot.values, zorder=0, aspect="auto", extent=(0,120,0,80), 
          cmap=sns.cubehelix_palette(light=1, as_cmap=True))

The plot already looks like

You can still play with alpha or use a different colormap etc. To also have the heatmap annotated, you can do so via

scale = np.array([120,80])
ax.imshow(data_pivot.values, zorder=0, aspect="auto", extent=(0,scale[0],0,scale[1]), 
          cmap=sns.cubehelix_palette(light=1, as_cmap=True), origin="lower")
offs = np.array([scale[0]/data_pivot.values.shape[1], scale[1]/data_pivot.values.shape[0]])
for pos, val in np.ndenumerate(data_pivot.values):
    ax.annotate(f"{val:.2f}", xy=np.array(pos)[::-1]*offs+offs/2, ha="center", va="center")
ax.invert_yaxis()

score 1 · Answer 2 · answered Feb 06 '19 at 13:08

1

Have a look at your scales, the pitch is on a completely different scale to the heatmap. If you zoom out you will see something that looks a bit like this:

Zoom out of plot

answered Feb 06 '19 at 13:08

Simon Rogers

344
1
10

Thanks, what's the best way to rescale? I tried to change plt.ylim() and plt.xlim() which made the heatmap bigger but I still couldn't see the football pitch [heatmap](https://imgur.com/1PcV5z1) – James Feb 06 '19 at 14:16
I think the easiest thing to do would be to modify your pitch so instead of going [0-120, 0-80], they go [0-4, 0-3]. I don't know how to scale the seaborn heatmap as it plots as categories. So you might have to look at other ways of doing the heatmap (e.g. pcolormesh) – Simon Rogers Feb 06 '19 at 15:26

score 1 · Accepted Answer · answered Feb 06 '19 at 15:28

As demonstrated in @simon-rogers' answer, you have an issue of scale between your pitch drawing and your heatmap.

Seaborn's heatmap will be drawn on an axe with limits: [0-number of columns] in x and [0-number of lines] in y. With your example dataframe, the resulting plot is therefore 4x3, while your pitch is 120x80.

The solution is to draw the pitch at a 4x3 scale. Better yet, draw the pitch with parametrized dimensions so it can fit various sized heatmaps.

I've started working on the function, but I didn't have time to figure out the ratios for the circles, I leave that as an exercise to you .

def draw_pitch(ax, width=120, height=80):
    # size of the pitch is width, height
    #Create figure

    #Pitch Outline & Centre Line
    plt.plot([0,0],[0,height], color="black")
    plt.plot([0,width],[height,height], color="black")
    plt.plot([width,width],[height,0], color="black")
    plt.plot([width,0],[0,0], color="black")
    plt.plot([width/2,width/2],[0,height], color="black")

    #Left Penalty Area
    plt.plot([width*0.12,width*0.12],[height*0.72,height*0.28],color="black")
    plt.plot([0,width*0.12],[height*0.72,height*0.72],color="black")
    plt.plot([0,width*0.12],[height*0.28,height*0.28],color="black")

    #Right Penalty Area
    plt.plot([width,width*0.88],[height*0.72,height*0.72],color="black")
    plt.plot([width*0.88,width*0.88],[height*0.72,height*0.28],color="black")
    plt.plot([width, width*0.88],[height*0.28,height*0.28],color="black")

    #Left 6-yard Box
    plt.plot([0,width*0.04],[height*0.6,height*0.6],color="black")
    plt.plot([width*0.04,width*0.04],[height*0.6,height*0.4],color="black")
    plt.plot([0,width*0.04],[height*0.4,height*0.4],color="black")

    #Right 6-yard Box
    plt.plot([width,width*0.96],[height*0.6,height*0.6],color="black")
    plt.plot([width*0.96,width*0.96],[height*0.6,height*0.4],color="black")
    plt.plot([width,width*0.96],[height*0.4,height*0.4],color="black")

    #Prepare Circles
    centreCircle = plt.Circle((width/2,40),8.1,color="black",fill=False)
    centreSpot = plt.Circle((width/2,40),0.71,color="black")
    leftPenSpot = plt.Circle((9.7,40),0.71,color="black")
    rightPenSpot = plt.Circle((110.3,40),0.71,color="black")

    #Draw Circles
    ax.add_patch(centreCircle)
    ax.add_patch(centreSpot)
    ax.add_patch(leftPenSpot)
    ax.add_patch(rightPenSpot)

    #Prepare Arcs
    # arguments for arc
    # x, y coordinate of centerpoint of arc
    # width, height as arc might not be circle, but oval
    # angle: degree of rotation of the shape, anti-clockwise
    # theta1, theta2, start and end location of arc in degree
    leftArc = plt_p.Arc((9.7,40),height=16.2,width=16.2,angle=0,theta1=310,theta2=50,color="black")
    rightArc = plt_p.Arc((110.3,40),height=16.2,width=16.2,angle=0,theta1=130,theta2=230,color="black")

    #Draw Arcs
    ax.add_patch(leftArc)
    ax.add_patch(rightArc)

Why fitting the pitch to the heatmap and not vice versa? Seems much more natural to me; else, if you have more data, the pitch will grow?! — ImportanceOfBeingErnest, Feb 06 '19 at 15:37
It would be easier for sure. I couldn't find a way to specify the extend of the heatmap using seaborn, so I went for scaling the pitch instead. The point is to keep the pitch the correct size for the dataframe, but that only works well if the heatmap has a ration 4/3, otherwise things will look weird. — Diziet Asahi, Feb 06 '19 at 15:41

How to overlay seaborn heatmap on matplotlib figure

3 Answers3