Matplotlib 2.x By Example

For quick plotting without having to set colors for each data series, Matplotlib uses a list of colors called the default property cycle, whereby each series is assigned one of the default colors in the cycle. In Matplotlib 2.0, the list has been changed from the original red, green, blue, cyan, magenta, yellow, and black, noted as ['b', 'g', 'r', 'c', 'm', 'y', 'k'], to the current category10 color palette introduced by the Tableau software. As implied by the name, the new palette has 10 distinct colors suitable for categorical display. The list can be accessed by importing Matplotlib and calling matplotlib.rcParams['axes.prop_cycle'] in Python.

Colormaps are useful in showing gradient. The yellow to blue "viridis" colormap is now the default one in Matplotlib 2.0. This perceptually uniform colormap better represents the transition of numerical values visually than the classic “jet” scheme. This is a comparison between two colormaps:

Besides defaulting to a perceptually continuous colormap, qualitative colormaps are now available for grouping values into categories:

Points in a scatter plot have a larger default size and no longer have a black edge, giving clearer visuals. Different colors in the default color cycle will be used for each data series if the color is not specified:

While previous versions set the legend in the upper-right corner, Matplotlib 2.0 sets the legend location as "best" by default. It automatically avoids overlapping of the legend with the data. The legend box also has rounded corners, lighter edges, and a partially transparent background to keep the focus of the readers on the data. The curve of square numbers in the classic and current default styles demonstrates the case:

Dash patterns in line styles can now scale with the line width to display bolder dashes for clarity:

From the documentation (https://matplotlib.org/users/dflt_style_changes.html#plot)

Just like the dots in the scatter plot shown before, most filled elements ("artists", which we will explain more in Chapter 2, Figure Aesthetics) no longer have a black edge by default, making the graphics less cluttered:

The default font is now changed from "Bitstream Vera Sans" to "DejaVu Sans". The current font supports additional international, math, and symbol characters, including emojis.

The alpha channel, which specifies the degree of transparency, is now fully supported in Matplotlib 2.0.

Matplotlib 2.0 now resamples images with less memory and less data type conversion.

It is claimed that the speed of text rendering by the Agg backend is increased by 20%. We will discuss more on backends in Chapter 6, Adding Interactivity and Animating Plots.

To generate a video output of animated plots, a more efficient codec, H.264, is now used by default in place of MPEG-4. As H.264 has a higher compression rate, the smaller output file size permits longer video record time and reduces the time and network data needed to load them. Real-time playback of H.264 videos is generally more fluent and in better quality than those encoded in MPEG-4.

New parameters are added, such as date.autoformatter.year for date time string formatting.

Style files are no longer allowed to configure settings unrelated to the style to prevent unexpected consequences. These parameters include the following:

'interactive', 'backend', 'backend.qt4', 'webagg.port', 'webagg.port_retries', 'webagg.open_in_browser', 'backend_fallback', 'toolbar', 'timezone', 'datapath', 'figure.max_open_warning', 'savefig.directory', tk.window_focus', 'docstring.hardcopy'

The Axes properties axisbg and axis_bgcolor are replaced by facecolor to keep the keywords consistent.

List is a basic Python data type for storing a collection of values. A list is created by putting element values inside a square bracket. To reuse our list, we can give it a name and store it like this:

evens = [2,4,6,8,10]

When we want to get a series in a greater range, for instance, to get more data points for our curve of squares to make it smoother, we may use the Python range() function:

evens = range(2,102,2)

This command will give us all even numbers from 2 to 100 (both inclusive) and store it in a list named evens.

Very often, we deal with more complex data. If you need a matrix with multiple columns or want to perform mathematical operations over all elements in a collection, then numpy is for you:

import numpy as np

We abbreviated numpy to np by convention, keeping our code succinct.

np.array() converts a supported data type, a list in this case, into a Numpy array. To produce a numpy array from our evens list, we do the following:

np.array(evens)

A pandas dataframe is useful when we have some non-numerical labels or values in our matrix. It does not require homogeneous data, unlike Numpy. Columns can be named. There are also functions such as melt() and pivot_table() that add convenience in reshaping the table to facilitate analysis and plotting.

To convert a list into a pandas dataframe, we do the following:

import pandas as pd
pd.DataFrame(evens)

You can also convert a numpy array into a pandas dataframe.

We can initialize an empty list, read the file line by line, split each line, and append the second element to our list:

evens = []
with open as f:
    for line in f.readlines():
        evens.append(line.split()[1])

It is simple when we have a file with only two columns, and only one column to read, but it can get more tedious when we have an extended table containing thousands of columns and rows and we want to convert them into a Numpy matrix later.

Numpy provides a standard one-liner solution:

import numpy as np
np.loadtxt(‘evens.txt’,delimiter=’\t’,usecols=1,dtype=np.int32)

The first parameter is the path of the data file. The delimiter parameter specifies the string used to separate values, which is a tab here. Because numpy.loadtxt() by default separate values separated by any whitespace into columns by default, this argument can be omitted here. We have set it for demonstration.

For usecols and dtype that specify which columns to read and what data type each column corresponds to, you may pass a single value to each, or a sequence (such as list) for reading multiple columns.

Numpy also by default skips lines starting with #, which typically marks comment or header lines. You may change this behavior by setting the comment parameter.

Similar to Numpy, pandas offers an easy way to load text files into a pandas dataframe:

import pandas as pd
pd.read_csv(usecols=1)

Here the separation can be denoted by either sep or delimiter, which is set as comma , by default (CSV stands for comma-separated values).

There is a long list of less commonly used options available as to determine how different data formats, data types, and errors should be handled. You may refer to the documentation at http://pandas.pydata.org/pandas-docs/stable/generated/pandas.read_csv.html. Besides flat CSV files, Pandas also has other built-in functions for reading other common data formats, such as Excel, JSON, HTML, HDF5, SQL, and Google BigQuery.

To stay focused on data visualization, we will not dig deep into the methods of data cleaning in this book, but this is a survival skill set very helpful in data science. If interested, you can check out resources on data handling with Python.

Compared to JPEG, another common image file format, PNG, has the advantage of allowing a transparent background. PNG is widely supported by most image viewers and handlers.

A PDF is a standard document format, which you don't have to worry about the availability of readers. However, most Office software do not support the import of PDF as image.

SVG is a vector graphics format that can be scaled without losing details. Hence, better quality can be achieved with a smaller file size. It goes well on the web with HTML5. It may not be supported by some primitive image viewers.

Postscript is a page description language for electronic publishing. It is useful for batch processing images to publish.