dexplot 0.0.7

Simple plotting library for both long and wide data integrated with DataFrames

Dexplot

A Python library for making data visualizations.

The current aim of Dexplot is to make data visualization creation in Python more robust and straightforward. Dexplot is built on top of Matplotlib and accepts Pandas DataFrames as inputs.

Installation

pip install dexplot

Goals

The primary goals for Dexplot are:

• Maintain a very consistent API with as few functions as necessary to make the desired statistical plots
• Allow the user to tweak the plots without digging into Matplotlib

Tidy Data from Pandas

Dexplot only accepts Pandas DataFrames as input for its plotting functions that are in "tidy" form.

Sample plots

Dexplot currently maintains two primary functions, aggplot which is used to aggregate data and jointplot, which is used to plot raw values from two variables against each other. heatmap is another function available that produces just a single heatmap.

aggplot can create five different kinds of plots.

• bar
• line
• box
• hist
• kde

jointplot can create four different kinds of plots

• scatter
• line
• 2D kde
• bar

There are 7 primary parameters to aggplot:

• agg - Name of column to be aggregated. If it is a column with string/categorical values, then the counts or relative frequency percentage will be returned.
• groupby - Name of column whose unique values will form independent groups. This is used in a similar fashion as the group by SQL clause.
• data - The Pandas DataFrame
• hue - The name of the column to further group the data within a single plot
• row - The name of the column who's unique values split the data in to separate rows
• col - The name of the column who's unique values split the data in to separate columns
• kind - The kind of plot to create. One of the five strings from above.

jointplot uses x and y instead of groupby and agg.

City of Houston Data

To get started, we will use City of Houston employee data collected from the year 2016. It contains public information from about 1500 employees and is located in Dexplot's GitHub repository.

import pandas as pd
import dexplot as dxp

emp = pd.read_csv('notebooks/data/employee.csv')
emp.head()

	title	dept	salary	race	gender	experience	experience_level
0	POLICE OFFICER	Houston Police Department-HPD	45279.0	White	Male	1	Novice
1	ENGINEER/OPERATOR	Houston Fire Department (HFD)	63166.0	White	Male	34	Veteran
2	SENIOR POLICE OFFICER	Houston Police Department-HPD	66614.0	Black	Male	32	Veteran
3	ENGINEER	Public Works & Engineering-PWE	71680.0	Asian	Male	4	Novice
4	CARPENTER	Houston Airport System (HAS)	42390.0	White	Male	3	Novice

Plotting the average salary by department

The agg parameter is very important and is what will be aggregated (summarized by a single point statistic, like the mean or median). It is the first parameter and only parameter you must specify (besides data). If this column is numeric, then by default, the mean of it will be calculated. Here, we specify the groupby parameter, who's unique values form the independent groups and label the x-axis.

dxp.aggplot(agg='salary', groupby='dept', data=emp)

<matplotlib.axes._subplots.AxesSubplot at 0x1190d2128>

Make horizontal with the orient parameter

The orient parameter controls whether the plot will be horizontal or vertical. By default it is set to 'h'.

dxp.aggplot(agg='salary', groupby='dept', data=emp, orient='h')

<matplotlib.axes._subplots.AxesSubplot at 0x1192f7160>

Controlling the figure size

One of the goals of Dexplot is to not have you dip down into the details of Matplotlib. We can use the figsize parameter to change the size of our plot.

dxp.aggplot(agg='salary', groupby='dept', data=emp, orient='h', figsize=(8, 4))

<matplotlib.axes._subplots.AxesSubplot at 0x119377b00>

Adding another dimension with hue

The hue parameter may be used to further subdivide each unique value in the groupby column. Notice that long tick labels are automatically wrapped.

dxp.aggplot(agg='salary', groupby='dept', data=emp, hue='gender')

<matplotlib.axes._subplots.AxesSubplot at 0x1193b1208>

Aggregating a String/Categorical column

It is possible to use a string/categorical column as the aggregating variable. In this instance, the counts of the unique values of that column will be returned. Because this is already doing a groupby, you cannot specify a groupby column in this instance. Let's get the count of employees by race.

dxp.aggplot(agg='race', data=emp, figsize=(8, 4))

<matplotlib.axes._subplots.AxesSubplot at 0x119377cf8>

Using hue with a String/Categorical column

Using a groupby is not allowed when a string/categorical column is being aggregated. But, we can still sub-divide the groups further by specifying hue.

dxp.aggplot(agg='race', data=emp, hue='dept')

<matplotlib.axes._subplots.AxesSubplot at 0x11b7d1588>

Getting the relative frequency percentage with normalize

It is possible to turn the raw counts into percentages by passing a value to normalize. Let's find the percentage of all employees by race.

dxp.aggplot(agg='race', data=emp, normalize='all', figsize=(8, 4))

<matplotlib.axes._subplots.AxesSubplot at 0x11b7f1e10>

You can normalize over any variable

The parameter normalize can be one of the values passed to the parameters 'agg', 'hue', 'row', 'col', or a tuple containing any number of these or 'all'. For instance, in the following plot, you can normalize by either race or dept.

dxp.aggplot(agg='race', data=emp, hue='dept', normalize='race')

<matplotlib.axes._subplots.AxesSubplot at 0x11bb0d048>

Data normalized by race

As you can see, the data was normalized by race. For example, from the graph, we can tell that about 30% of black employees were members of the police department. We can also normalize by department. From the graph, about 10% of the Health & Human Services employees were Asian.

dxp.aggplot(agg='race', data=emp, hue='dept', normalize='dept')

<matplotlib.axes._subplots.AxesSubplot at 0x11bf4f0b8>

Stacked Bar Plots

All bar plots that have use the hue variable, can be stacked. Here, we stack the maximum salary by department grouped by race.

dxp.aggplot(agg='salary', data=emp, hue='dept', groupby='race', aggfunc='max', stacked=True)

<matplotlib.axes._subplots.AxesSubplot at 0x11b7d1208>

Stacking counts

The raw counts of each department by experience level are stacked here.

dxp.aggplot(agg='experience_level', data=emp, hue='dept', aggfunc='max', stacked=True)

<matplotlib.axes._subplots.AxesSubplot at 0x11c41b0f0>

Stacking relative frequencies

The relative frequencies of each department by each race and experience level.

dxp.aggplot(agg='experience_level', data=emp, hue='dept', row='race', 
            normalize=('race', 'experience_level'), wrap=3, stacked=True)

(<Figure size 864x720 with 5 Axes>,)

Other kinds of plots line, box, hist, and kde

aggplot is capable of making four other kinds of plots. The line plot is very similar to the bar plot but simply connects the values together. Let's go back to a numeric column and calculate the median salary by department across each gender.

dxp.aggplot(agg='salary', data=emp, groupby='dept', hue='gender', kind='line', aggfunc='median')

<matplotlib.axes._subplots.AxesSubplot at 0x11c994eb8>

aggfunc can take any string value that Pandas can

There are more than a dozen string values that aggfunc can take. These are simply passed to Pandas groupby method which does the aggregation.

All plots can be both vertical and horizontal

We can rotate all plots with orient.

dxp.aggplot(agg='salary', data=emp, groupby='dept', hue='gender', kind='line', aggfunc='median', orient='h')

<matplotlib.axes._subplots.AxesSubplot at 0x11cd2ad68>

Boxplots

Here is the same data plotted as a box plot. This isn't actually an aggregation, so the aggfunc parameter is meaningless here. Instead, all the values of the particular group are plotted.

dxp.aggplot(agg='salary', data=emp, groupby='dept', hue='gender', kind='box', orient='h')

<matplotlib.axes._subplots.AxesSubplot at 0x11d0c27f0>

Histograms and KDE's

As with boxplots, histograms and kdes do not function with aggfunc as they aren't aggregating but simply displaying all the data for us. Also, it is not possible to use both groupby and agg with these plots.

dxp.aggplot(agg='salary', data=emp, groupby='dept', kind='hist', orient='v')

<matplotlib.axes._subplots.AxesSubplot at 0x11d37c780>

dxp.aggplot(agg='salary', data=emp, groupby='dept', kind='kde', orient='v')

<matplotlib.axes._subplots.AxesSubplot at 0x11d5ee748>

Splitting into separate plots

The row and col parameters can be used to split the data into separate plots. Each unique value of row or col will create a new plot. A one-item tuple consisting of the entire Figure is returned.

dxp.aggplot(agg='salary', data=emp, groupby='experience_level', kind='kde', orient='v', row='dept')

(<Figure size 720x1152 with 6 Axes>,)

Use the wrap parameter to make new rows/columns

Set the wrap parameter to an integer to determine where a new row/column will be formed.

dxp.aggplot(agg='salary', data=emp, groupby='experience_level', kind='box', orient='v', row='dept', wrap=3)

(<Figure size 864x720 with 6 Axes>,)

wrap works for both row or col

dxp.aggplot(agg='salary', data=emp, groupby='experience_level', kind='box', orient='v', col='dept', wrap=5)

(<Figure size 1296x576 with 6 Axes>,)

Use both row and col for a entire grid

By using both row and col, you can maximize the number of variables you divide the data into.

dxp.aggplot(agg='salary', data=emp, groupby='gender', kind='kde', row='dept', col='experience_level')

(<Figure size 1008x1152 with 18 Axes>,)

Normalize by more than one variable

Before, we normalized by just a single variable. It is possible to normalize by multiple variables with a tuple. Here we normalize by department and gender. Adding up all the blue bars for each department should add to 1.

dxp.aggplot(agg='dept', data=emp, hue='gender', kind='bar', row='race', normalize=('dept', 'gender'))

(<Figure size 720x1008 with 5 Axes>,)

Normalize by three variables

Here we normalize by race, experience level, and gender. Each set of orange/blue bars within each plot will add to 1.

dxp.aggplot(agg='dept', data=emp, hue='gender', kind='bar', row='race', 
            col='experience_level', normalize=('gender', 'experience_level', 'race'), orient='h')

(<Figure size 1008x1008 with 15 Axes>,)

Joint Plots

joinplot works differently than aggplot in that no aggregation takes place. It plots the raw values between two variables. It can split the data into groups or new plots with hue, row, and col. The default plot is a scatter plot, but you can also provide a string value to the kind parameter to make line, kde, or bar plots.

dxp.jointplot('experience', 'salary', data=emp)

<matplotlib.axes._subplots.AxesSubplot at 0x120b9af60>

Split data in the same plot with hue

dxp.jointplot('experience', 'salary', data=emp, hue='gender')

<matplotlib.axes._subplots.AxesSubplot at 0x12171e6d8>

Plot a regression line by setting fit_reg equal to True

By default it plots the 95% confidence interval around the mean.

dxp.jointplot('experience', 'salary', data=emp, hue='gender', fit_reg=True)

<matplotlib.axes._subplots.AxesSubplot at 0x1218e6c18>

Further split the data into separate plots with row and col

dxp.jointplot('experience', 'salary', data=emp, hue='gender', row='dept', wrap=3, fit_reg=True)

(<Figure size 864x720 with 6 Axes>,)

dxp.jointplot('experience', 'salary', data=emp, hue='gender', row='dept', col='experience_level')

(<Figure size 1008x1152 with 18 Axes>,)

Use the s parameter to change the size of each marker

Let s equal a column name containing numeric values to set each marker size individually. We need to create another numeric variable first since the dataset only contains two.

import numpy as np
emp['num'] = np.random.randint(10, 300, len(emp))

dxp.jointplot('experience', 'salary', data=emp, hue='gender', row='dept', wrap=3, s='num')

(<Figure size 864x720 with 6 Axes>,)

Line Plots

df_stocks = pd.read_csv('notebooks/data/stocks.csv', parse_dates=['date'])
df_stocks.head()

<style scoped> .dataframe tbody tr th:only-of-type { vertical-align: middle; }

.dataframe tbody tr th {
    vertical-align: top;
}

.dataframe thead th {
    text-align: right;
}

</style>

	date	close	symbol	percent_gain	year	month
0	2013-10-07	63.7997	aapl	0.0	2013	10
1	2013-10-07	96.6579	cvx	0.0	2013	10
2	2013-10-07	35.0541	txn	0.0	2013	10
3	2013-10-07	19.4912	csco	0.0	2013	10
4	2013-10-07	310.0300	amzn	0.0	2013	10

dxp.jointplot(x='date', y='percent_gain', data=df_stocks, hue='symbol', kind='line')

<matplotlib.axes._subplots.AxesSubplot at 0x121ad34a8>

dxp.jointplot(x='date', y='percent_gain', data=df_stocks, kind='line', hue='symbol', row='year', wrap=3,
             sharex=False, sharey=False)

(<Figure size 864x720 with 6 Axes>,)

2D KDE Plots

dxp.jointplot('experience', 'salary', data=emp, kind='kde')

<matplotlib.axes._subplots.AxesSubplot at 0x12290e898>

dxp.jointplot('experience', 'salary', data=emp, kind='kde', row='dept', col='gender', sharex=False, sharey=False)

(<Figure size 864x1152 with 12 Axes>,)

Bar Plots for aggregated data

If your data is already aggregated, you can use jointplot to plot it.

df = emp.groupby('dept').agg({'salary':'mean'}).reset_index()
df

<style scoped> .dataframe tbody tr th:only-of-type { vertical-align: middle; }

.dataframe tbody tr th {
    vertical-align: top;
}

.dataframe thead th {
    text-align: right;
}

</style>

	dept	salary
0	Health & Human Services	51324.980583
1	Houston Airport System (HAS)	53990.368932
2	Houston Fire Department (HFD)	59960.441096
3	Houston Police Department-HPD	60428.745614
4	Parks & Recreation	39426.150943
5	Public Works & Engineering-PWE	50207.806452

dxp.jointplot('dept', 'salary', data=df, kind='bar')

<matplotlib.axes._subplots.AxesSubplot at 0x12512fe10>

Heatmaps

Heatmaps work with both tidy and aggregated data.

Frequency

When working with tidy data, passing it just x and y will plot the frequency of occurrences for all of the combinations of their unique values. Place the count as text in the box with annot. The default formatting has two decimals.

dxp.heatmap(x='dept', y='race', data=emp, annot=True, fmt='.0f')

(<Figure size 720x576 with 2 Axes>,)

Aggregating a variable with heatmaps

Set the agg parameter to aggregate a particular variable. Choose how you will aggregate with the aggfunc parameter, which takes any string that Pandas can. The default it the mean.

dxp.heatmap(x='dept', y='race', agg='salary', aggfunc='max', data=emp, annot=True, fmt='.0f')

(<Figure size 720x576 with 2 Axes>,)

Normalize heatmaps by row, column, or all data

You can normalize the data by row, column, or all data with. Use the string name of the column for row and column normalization. Below we find the total percentage of all combined years of experience normalized by race. For example, of all the total years of experience for White employees, 89% of those years are male.

dxp.heatmap(x='race', y='gender', agg='experience', aggfunc='sum', 
            data=emp, annot=True, fmt='.3f', normalize='race')

(<Figure size 720x576 with 2 Axes>,)

dxp.heatmap(x='race', y='dept', agg='experience', aggfunc='sum', 
            data=emp, annot=True, fmt='.3f', normalize='race', corr=True)

(<Figure size 720x576 with 2 Axes>,)

Heatmaps without aggregating data

If you pass just the DataFrame into heatmap then those raw values will be used to create the colors. Here we plot some random numbers from a normal distribution.

df = pd.DataFrame(np.random.randn(10, 5), columns=list('abcde'))
fig, = dxp.heatmap(data=df, annot=True)

Find correlations by setting corr equal to True

Setting the corr parameter to True computes the pairwise correlation matrix between the columns. Any string columns are discarded. Below, we use the popular Kaggle housing dataset.

housing = pd.read_csv('notebooks/data/housing.csv')
fig, = dxp.heatmap(data=housing, corr=True, figsize=(16, 16))

Comparison with Seaborn

If you have used the Seaborn library, then you should notice a lot of similarities. Much of Dexplot was inspired by Seaborn. Below is a list of the extra features in Dexplot not found in Seaborn

• The ability to graph relative frequency percentage and normalize over any number of variables
• Far fewer public functions. Only two at the moment
• No need for multiple functions to do the same thing. Seaborn has both countplot and barplot
• Ability to make grids with a single function instead of having to use a higher level function like catplot
• Pandas groupby methods are available as strings
• Both x/y-labels and titles are automatically wrapped so that they don't overlap
• The figure size (plus several other options) and available to change without dipping down into matplotlib
• No new types like FacetGrid. Only matplotlib objects are returned