Matplotlib is a library to visualize data. This section is about using matplotlib in the context of building ML algorithms. It does not cover matplotlib content to build reports or presentations.
Despite their name, they are used to plot any kind of one-dimensional signal. For example: time series or sound waves.
import numpy as np
import matplotlib.pyplot as plt
# Dummy data
x = np.linspace(0, 20, 1000) # [0, ... 1000 points, 20]
y = np.sin(x) + 0.2 * x
plt.plot(x, y)
plt.xlabel('input')
plt.ylabel('output')
plt.title('my plot'); # semi-colon suppresses the output of plt.title() in the jupyter notebook result as it is irrelevant
# plt.show() # Needed if you are not inside a notebook to show the graph
Plot 2D data to visualise the geometric relationships between them. e.g. clustering plots.
# Dummy data
X = np.random.randn(200, 2)
X[:50] += 3
Y = np.zeros(200)
Y[:50] = 1
plt.scatter(X[:,0], X[:,1], c=Y);
# first argument = first dimension, second arg = second dimension
# c means color.
# - It is optional.
# - It must be a one dimensional array of n-samples with integers representing a label
Lets us see the distribution of our data.
X1 = np.random.randn(5000) # Dummy data
plt.hist(X1, bins=50);
# Dummy Data
X1 = np.random.randn(5000)
X2 = np.random.randn(5000) + 3
plt.hist(X1, bins=50, alpha=0.5, label="data1")
plt.hist(X2, bins=50, alpha=0.5, label="data2")
plt.xlabel("Data")
plt.ylabel("Count")
plt.title("Multiple histograms")
# the `label` attributes don't get shown unless legend is specified
plt.legend(loc='upper right');
# To open images into python we can use a library called Pillow
from PIL import Image
im = Image.open('img/section-3-matplotlib/dummy-image.jpeg')
type(im) # PIL.JpegImagePlugin.JpegImageFile
# PIL images can be converted to their np array representation
image_as_array = np.array(im)
image_as_array.shape # (354, 536, 3) image h, image w, RGB channels
# Use imshow to show images
plt.imshow(image_as_array)
plt.imshow(im) # Both produce the same result
# It is very common to work with black and white images in computer vision tasks.
# A common way of converting a color image to B&W is to take the average of the RGB dimensions of the image
bw_image_as_array = image_as_array.mean(axis=2)
bw_image_as_array.shape # (354, 536) => This has lost the 3rd dimension
# The first command may show the image in some weird shades of green. This is because it is up to matplot lib to
# assign which colour to show for an "image" that only has one channel.
# The cmap argument is used to control the color map that matplotlib will use.
plt.imshow(bw_image_as_array)
plt.imshow(bw_image_as_array, cmap='gray')
A heatmap is nothing more than an image visualization of 2D data.
dummy_data = np.arange(0, 200 * 200, 1).reshape((200,200))
plt.imshow(dummy_data, cmap="inferno")
More complete instructions in the docs.
It is very common to combine several types of plots to show data.
The combination linear + scatter plot is particularly common en machine learning to plot things like
regression results or decision boundaries in classifiers.
# Dummy data
x = np.linspace(-10, 20, 100)
y = 3*x + 2
y2 = y + np.random.normal(0, 10, size=y.shape)
y2_color = y2 > y
# Just keep adding things to the same plot
plt.plot(x,y)
plt.plot(x, y-20)
plt.scatter(x, y2, c=y2_color);
import numpy as np
import matplotlib.pyplot as plt
x = np.linspace(-10, 10, 100)
y1 = x
y2 = x ** 2
y3 = np.sin(x)
y4 = np.abs(x)
# The simplest way of doing it
plt.subplot(2, 2, 1) # rows, columns, active position
plt.plot(x, y1)
plt.subplot(2, 2, 2)
plt.plot(x, y2)
plt.subplot(2, 2, 3)
plt.plot(x, y3)
plt.subplot(2, 2, 4)
plt.plot(x, y4)
# Another way of doing it that gives you more control
fig, ax = plt.subplots(2, 2, figsize=(10,7))
fig.tight_layout()
ax[0,0].plot(x, y1)
ax[0,1].plot(x, y2)
ax[1,0].plot(x, y3)
ax[1,1].plot(x, y4)
Go to the pandas summary > plotting with Pandas for information on how to build this same plots with Pandas and how to build some other plots like boxplots and scatter_matrix.