heatmap

This section provides examples of how to use the heatmap function. At a minimum, the heatmap function requires the following keywords:

  • df: a pandas DataFrame

  • x: the name of the DataFrame column containing the x-axis data

  • y: the name of the DataFrame column containing the y-axis data

  • z: the name of the DataFrame column containing the z-axis data

Heatmaps in fivecentplots can display both categorical and non-categorical data on either a uniform or non-uniform grid.

Setup

Imports

[1]:

%load_ext autoreload
%autoreload 2
%matplotlib inline
import fivecentplots as fcp
import pandas as pd
import numpy as np
import os, sys, pdb
osjoin = os.path.join
db = pdb.set_trace

Sample data

[2]:

df = pd.read_csv(osjoin(os.path.dirname(fcp.__file__), 'tests', 'fake_data_heatmap.csv'))
df.head()

[2]:
Player Category Average
0 Lebron James Points 27.5
1 Lebron James Assists 9.1
2 Lebron James Rebounds 8.6
3 Lebron James Blocks 0.9
4 James Harden Points 30.4

Set theme

[3]:

#fcp.set_theme('gray')
#fcp.set_theme('white')

Other

[4]:

SHOW = False

Categorical heatmap

First consider a case where both the x and y DataFrame columns contain categorical data values:

No data labels

[5]:

fcp.heatmap(df=df, x='Category', y='Player', z='Average', cbar=True, show=SHOW)
_images/heatmap_14_0.png

Note that for heatmaps the x tick labels are rotated 90° by default. This can be overridden via the keyword tick_labels_major_x_rotation.

With data labels

[6]:

fcp.heatmap(df=df, x='Category', y='Player', z='Average', cbar=True, data_labels=True,
            heatmap_font_color='#aaaaaa', show=SHOW, tick_labels_major_y_edge_width=0, ws_ticks_ax=5)
_images/heatmap_17_0.png

Cell size

The size of the heatmap cell will default to a width of 60 pixels unless: (1) the keyword heatmap_cell_size (or cell_size when directly supplying the value to the function call) is specified; or (2) ax_size is explicitly defined. Note that for a heatmap the cells are always square with width=height.

[7]:

fcp.heatmap(df=df, x='Category', y='Player', z='Average', cbar=True, data_labels=True,
            heatmap_font_color='#aaaaaa', show=SHOW, tick_labels_major_y_edge_width=0, ws_ticks_ax=5, cell_size=100)
_images/heatmap_20_0.png

Non-uniform data

A major difference between heatmaps and contour plots is that contour plots assume that the x and y DataFrame column values are numerical and continuous. With a heatmap, we can cast numerical data into categorical form. Note that any missing values get mapped as nan values are not not plotted.

[8]:

# Read the contour DataFrame
df2 = pd.read_csv(osjoin(os.path.dirname(fcp.__file__), 'tests', 'fake_data_contour.csv'))

[9]:

fcp.heatmap(df=df2, x='X', y='Y', z='Value', row='Batch', col='Experiment',
            cbar=True, show=SHOW, share_z=True, ax_size=[400, 400],
            data_labels=False, label_rc_font_size=12, filter='Batch==103', cmap='viridis')
_images/heatmap_24_0.png

Note that the x-axis width is not 400px as specified by the keyword ax_scale. This occurs because the data set does not have as many values on the x-axis as on the y-axis. fivecentplots applies the axis size to the axis with the most items and scales the other axis accordingly.

imshow alternative

We can also use fcp.heatmap to display images (similar to imshow in matplotlib). Here we will take a random image from the world wide web, place it in a pandas DataFrame, and display.

[10]:

# Read an image
import imageio
url = 'https://s4827.pcdn.co/wp-content/uploads/2011/04/low-light-iphone4.jpg'
imgr = imageio.imread(url)

# Convert to grayscale
r, g, b = imgr[:,:,0], imgr[:,:,1], imgr[:,:,2]
gray = 0.2989 * r + 0.5870 * g + 0.1140 * b

# Convert image data to pandas DataFrame
img = pd.DataFrame(gray)
img.head()

[10]:
0 1 2 3 4 5 6 7 8 9 ... 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591
0 0.9999 0.0000 0.0000 2.2278 5.2275 5.2275 5.2275 4.2276 2.2278 2.2278 ... 0.8859 1.7718 2.7717 4.7715 4.7715 2.7717 2.0599 2.0599 2.0599 1.1740
1 0.9999 3.9996 2.9997 1.2279 5.2275 9.2271 7.2273 3.2277 2.2278 2.2278 ... 0.8859 0.8859 2.7717 1.7718 1.7718 3.7716 2.0599 0.5870 0.5870 1.1740
2 6.9993 4.9995 3.9996 2.2278 11.2269 18.2262 9.2271 0.2280 2.2278 1.2279 ... 6.7713 6.7713 2.7717 1.7718 2.7717 4.7715 4.0597 1.1740 0.5870 2.0599
3 13.9986 0.9999 1.9998 2.9997 7.2273 15.2265 9.2271 4.2276 1.2279 0.2280 ... 6.7713 6.7713 2.7717 4.7715 7.7712 8.7711 10.0591 9.0592 6.0595 3.0598
4 6.9993 0.0000 12.9987 14.9985 4.2276 5.2275 5.2275 4.2276 1.2279 0.2280 ... 1.7718 0.8859 3.7716 6.7713 10.7709 13.7706 16.0585 15.0586 10.0591 5.0596

5 rows × 2592 columns

Display the image as a colored heatmap:

[11]:

fcp.heatmap(img, cmap='inferno', cbar=True, ax_size=[600, 600])
_images/heatmap_30_0.png

Now let’s enhance the contrast of the same image by limiting our color range to the mean pixel value +/- 3 * sigma:

[12]:

uu = img.stack().mean()
ss = img.stack().std()
fcp.heatmap(img, cmap='inferno', cbar=True, ax_size=[600, 600], zmin=uu-3*ss, zmax=uu+3*ss)
_images/heatmap_32_0.png

We can also crop the image by specifying range value for x and y. Unlike imshow, the actual row and column values displayed on the x- and y-axis are preserved after the zoom (not reset to 0, 0):

[13]:

fcp.heatmap(img, cmap='inferno', cbar=True, ax_size=[600, 600], xmin=1400, xmax=2000, ymin=500, ymax=1000)
_images/heatmap_34_0.png

private eyes are watching you…