Pandas API

If hvplot and pandas are both installed, then we can use the pandas.options.plotting.backend to control the output of pd.DataFrame.plot and pd.Series.plot. This notebook is meant to recreate the pandas visualization docs.

import numpy as np
import pandas as pd

pd.options.plotting.backend = 'holoviews'

Basic Plotting: plot

The plot method on Series and DataFrame is just a simple wrapper around hvplot():

ts = pd.Series(np.random.randn(1000), index=pd.date_range('1/1/2000', periods=1000))
ts = ts.cumsum()
ts.plot()

If the index consists of dates, they will be sorted (as long as sort_date=True) and formatted the x-axis nicely as per above.

On DataFrame, plot() is a convenience to plot all of the columns with labels:

df = pd.DataFrame(np.random.randn(1000, 4), index=ts.index, columns=list('ABCD'))
df = df.cumsum()
df.plot()

You can plot one column versus another using the x and y keywords in plot():

df3 = pd.DataFrame(np.random.randn(1000, 2), columns=['B', 'C']).cumsum()
df3['A'] = pd.Series(list(range(len(df))))
df3.plot(x='A', y='B')

Other Plots

Plotting methods allow for a handful of plot styles other than the default line plot. These methods can be provided as the kind keyword argument to plot(). These include:

• bar or barh for bar plots

• hist for histogram

• box for boxplot

• kde or density for density plots

• area for area plots

• scatter for scatter plots

• hexbin for hexagonal bin plots

• ~~pie for pie plots~~

For example, a bar plot can be created the following way:

df.iloc[5].plot(kind='bar')

You can also create these other plots using the methods DataFrame.plot. instead of providing the kind keyword argument. This makes it easier to discover plot methods and the specific arguments they use:

In addition to these kinds, there are the DataFrame.hist(), and DataFrame.boxplot() methods, which use a separate interface.

Finally, there are several plotting functions in hvplot.plotting that take a Series or DataFrame as an argument. These include:

• Scatter Matrix

• Andrews Curves

• Parallel Coordinates

• Lag Plot

• ~~Autocorrelation Plot~~

• ~~Bootstrap Plot~~

• ~~RadViz~~

Plots may also be adorned with errorbars or tables.

Bar plots

For labeled, non-time series data, you may wish to produce a bar plot:

import holoviews as hv

df.iloc[5].plot.bar() * hv.HLine(0).opts(color='k')

Calling a DataFrame’s plot.bar() method produces a multiple bar plot:

df2 = pd.DataFrame(np.random.rand(10, 4), columns=['a', 'b', 'c', 'd'])
df2.plot.bar()

To produce a stacked bar plot, pass stacked=True:

df2.plot.bar(stacked=True)

To get horizontal bar plots, use the barh method:

df2.plot.barh(stacked=True)

Histograms

Histogram can be drawn by using the DataFrame.plot.hist() and Series.plot.hist() methods.

df4 = pd.DataFrame({'a': np.random.randn(1000) + 1, 'b': np.random.randn(1000),
                    'c': np.random.randn(1000) - 1}, columns=['a', 'b', 'c'])

df4.plot.hist(alpha=0.5)

Using the matplotlib backend, histograms can be stacked using stacked=True; stacking is not yet supported in the holoviews backend. Bin size can be changed by bins keyword.

# Stacked not supported
df4.plot.hist(stacked=True, bins=20)

WARNING:param.main: Stacking for histograms is not yet implemented in holoviews. Use bar plots if stacking is required.

You can pass other keywords supported by matplotlib hist. For example, horizontal and cumulative histogram can be drawn by invert=True and cumulative=True.

df4['a'].plot.hist(invert=True, cumulative=True)

The existing interface DataFrame.hist to plot histogram still can be used.

df['A'].diff().hist()

DataFrame.hist() plots the histograms of the columns on multiple subplots:

df.diff().hist(color='k', alpha=0.5, bins=50, subplots=True, width=300).cols(2)

The by keyword can be specified to plot grouped histograms:

# by does not support arrays, instead the array should be added as a column
data = pd.Series(np.random.randn(1000))
data = pd.DataFrame({'data': data, 'by_column': np.random.randint(0, 4, 1000)})
data.hist(by='by_column', width=300, subplots=True).cols(2)

Box Plots

Boxplot can be drawn calling Series.plot.box() and DataFrame.plot.box(), or DataFrame.boxplot() to visualize the distribution of values within each column.

For instance, here is a boxplot representing five trials of 10 observations of a uniform random variable on [0,1).

df = pd.DataFrame(np.random.rand(10, 5), columns=['A', 'B', 'C', 'D', 'E'])
df.plot.box()

Using the matplotlib backend, boxplot can be colorized by passing color keyword. You can pass a dict whose keys are boxes, whiskers, medians and caps.

This behavior is not supported in the holoviews backend.

You can pass other keywords supported by holoviews boxplot. For example, horizontal and custom-positioned boxplot can be drawn by invert=True and positions keywords.

# positions not supported
df.plot.box(invert=True, positions=[1, 4, 5, 6, 8])

WARNING:param.main: positions option not found for box plot with bokeh; similar options include: ['projection']

See the boxplot method and the matplotlib boxplot documentation for more.

The existing interface DataFrame.boxplot to plot boxplot still can be used.

df = pd.DataFrame(np.random.rand(10, 5))
df.boxplot()

You can create a stratified boxplot using the groupby keyword argument to create groupings. For instance,

df = pd.DataFrame(np.random.rand(10,2), columns=['Col1', 'Col2'])

df['X'] = pd.Series(['A','A','A','A','A','B','B','B','B','B'])

df.boxplot(col='X')

You can also pass a subset of columns to plot, as well as group by multiple columns:

df = pd.DataFrame(np.random.rand(10,3), columns=['Col1', 'Col2', 'Col3'])

df['X'] = pd.Series(['A','A','A','A','A','B','B','B','B','B'])
df['Y'] = pd.Series(['A','B','A','B','A','B','A','B','A','B'])

df.boxplot(y=['Col1','Col2'], col='X', row='Y')

df_box = pd.DataFrame(np.random.randn(50, 2))
df_box['g'] = np.random.choice(['A', 'B'], size=50)
df_box.loc[df_box['g'] == 'B', 1] += 3
df_box.boxplot(row='g')

For more control over the ordering of the levels, we can perform a groupby on the data before plotting.

df_box.groupby('g').boxplot()

Area Plot

You can create area plots with Series.plot.area() and DataFrame.plot.area(). Area plots are not stacked by default. To produce stacked area plot, each column must be either all positive or all negative values.

When input data contains NaN, it will be automatically filled by 0. If you want to drop or fill by different values, use dataframe.dropna() or dataframe.fillna() before calling plot.

df = pd.DataFrame(np.random.rand(10, 4), columns=['a', 'b', 'c', 'd'])
df.plot.area(alpha=0.5)

To produce an stacked plot, pass stacked=True.

df.plot.area(stacked=True)

Scatter Plot

Scatter plot can be drawn by using the DataFrame.plot.scatter() method. Scatter plot require that x and y be specified using the x and y keywords.

df = pd.DataFrame(np.random.rand(50, 4), columns=['a', 'b', 'c', 'd'])
df.plot.scatter(x='a', y='b')

To plot multiple column groups in a single axes, repeat plot method and then use the overlay operator (*) to combine the plots. It is recommended to specify color and label keywords to distinguish each groups.

plot_1 = df.plot.scatter(x='a', y='b', color='DarkBlue', label='Group 1')
plot_2 = df.plot.scatter(x='c', y='d', color='DarkGreen', label='Group 2')

plot_1 * plot_2

The keyword c may be given as the name of a column to provide colors for each point:

df.plot.scatter(x='a', y='b', c='c', s=50)

You can pass other keywords supported by matplotlib scatter. The example below shows a bubble chart using a column of the DataFrame as the bubble size.

df.plot.scatter(x='a', y='b', s=df['c']*500)

The same effect can be accomplished using the scale option.

df.plot.scatter(x='a', y='b', s='c', scale=25)

Hexagonal Bin Plot

You can create hexagonal bin plots with DataFrame.plot.hexbin(). Hexbin plots can be a useful alternative to scatter plots if your data are too dense to plot each point individually.

df = pd.DataFrame(np.random.randn(1000, 2), columns=['a', 'b'])
df['b'] = df['b'] + np.arange(1000)

df.plot.hexbin(x='a', y='b', gridsize=25, width=500, height=400)

A useful keyword argument is gridsize; it controls the number of hexagons in the x-direction, and defaults to 100. A larger gridsize means more, smaller bins.

By default, a histogram of the counts around each (x, y) point is computed. You can specify alternative aggregations by passing values to the C and reduce_C_function arguments. C specifies the value at each (x, y) point and reduce_C_function is a function of one argument that reduces all the values in a bin to a single number (e.g. mean, max, sum, std). In this example the positions are given by columns a and b, while the value is given by column z. The bins are aggregated with numpy’s max function.

df = pd.DataFrame(np.random.randn(1000, 2), columns=['a', 'b'])
df['b'] = df['b'] = df['b'] + np.arange(1000)
df['z'] = np.random.uniform(0, 3, 1000)

df.plot.hexbin(x='a', y='b', C='z', reduce_function=np.max, gridsize=25, width=500, height=400)

Density Plot

You can create density plots using the Series.plot.kde() and DataFrame.plot.kde() methods.

ser = pd.Series(np.random.randn(1000))

ser.plot.kde()

Pie plot

NOT SUPPORTED

Plotting Tools

These functions can be imported from hvplot.plotting and take a Series or DataFrame as an argument.

Scatter Matrix Plot

You can create a scatter plot matrix using the scatter_matrix function:

import pandas as pd
import numpy as np

from hvplot import scatter_matrix

df = pd.DataFrame(np.random.randn(1000, 4), columns=['a', 'b', 'c', 'd'])

scatter_matrix(df, alpha=0.2, diagonal='kde')

Since scatter matrix plots may generate a large number of points (the one above renders 120,000 points!), you may want to take advantage of the power provided by Datashader to rasterize the off-diagonal plots into a fixed-resolution representation:

df = pd.DataFrame(np.random.randn(10000, 4), columns=['a', 'b', 'c', 'd'])

scatter_matrix(df, rasterize=True, dynspread=True)

Andrews Curves

Andrews curves allow one to plot multivariate data as a large number of curves that are created using the attributes of samples as coefficients for Fourier series. By coloring these curves differently for each class it is possible to visualize data clustering. Curves belonging to samples of the same class will usually be closer together and form larger structures.

from bokeh.sampledata import iris
from hvplot import andrews_curves

data = iris.flowers

andrews_curves(data, 'species')

Parallel Coordinates

Parallel coordinates is a plotting technique for plotting multivariate data. It allows one to see clusters in data and to estimate other statistics visually. Using parallel coordinates points are represented as connected line segments. Each vertical line represents one attribute. One set of connected line segments represents one data point. Points that tend to cluster will appear closer together.

from hvplot import parallel_coordinates

parallel_coordinates(data, 'species')

Lag Plot

Lag plots are used to check if a data set or time series is random. Random data should not exhibit any structure in the lag plot. Non-random structure implies that the underlying data are not random.

from hvplot import lag_plot

data = pd.Series(0.1 * np.random.rand(1000) +
                 0.9 * np.sin(np.linspace(-99 * np.pi, 99 * np.pi, num=1000)))

lag_plot(data, width=400, height=400)

Autocorrelation Plot

NOT SUPPORTED

Bootstrap Plot

NOT SUPPORTED

RadViz

NOT SUPPORTED

Plot Formatting

General plot style arguments

Most plotting methods have a set of keyword arguments that control the layout and formatting of the returned plot:

ts.plot(c='k', line_dash='dashed', label='Series')

For each kind of plot (e.g. line, bar, scatter) any additional arguments keywords are passed along to the corresponding holoviews object (hv.Curve, hv.Bar, hv.Scatter). These can be used to control additional styling, beyond what pandas provides.

Controlling the Legend

You may set the legend argument to False to hide the legend, which is shown by default. You can also control the placement of the legend using the same legend argument set to one of: ‘top_right’, ‘top_left’, ‘bottom_left’, ‘bottom_right’, ‘right’, ‘left’, ‘top’, or ‘bottom’.

df = pd.DataFrame(np.random.randn(1000, 4), 
                  index=ts.index, columns=list('ABCD'))
df = df.cumsum()

df.plot(legend=False)

df.plot(legend='top_left')

Scales

You may pass logy to get a log-scale Y axis, similarly use logx to get a log-scale on the X axis or logz to get a log-scale on the color axis.

ts = pd.Series(np.random.randn(1000), 
               index=pd.date_range('1/1/2000', periods=1000))

ts = np.exp(ts.cumsum())

ts.plot(logy=True)

df = pd.DataFrame(np.random.randn(1000, 2), columns=['a', 'b'])
df['b'] = df['b'] + np.arange(1000)
df.plot.hexbin(x='a', y='b', gridsize=25, width=500, height=400, logz=True)

Plotting on a Secondary Y-axis

NOT SUPPORTED

Suppressing tick resolution adjustment

pandas includes automatic tick resolution adjustment for regular frequency time-series data. For limited cases where pandas cannot infer the frequency information (e.g., in an externally created twinx), you can choose to suppress this behavior for alignment purposes.

Here is the default behavior, notice how the x-axis tick labeling is performed:

df = pd.DataFrame(np.random.randn(1000, 4), 
                  index=ts.index, columns=list('ABCD'))
df = df.cumsum()
df.A.plot()

Formatters can be set using format strings, by declaring bokeh TickFormatters, or using custom functions. See HoloViews Tick Docs for more information.

from bokeh.models.formatters import DatetimeTickFormatter

formatter = DatetimeTickFormatter(months='%b %Y')

df.A.plot(yformatter='$%.2f', xformatter=formatter)

Subplots

Each Series in a DataFrame can be plotted on a different axis with the subplots keyword:

df = pd.DataFrame(np.random.randn(1000, 4), 
                  index=ts.index, columns=list('ABCD'))
df = df.cumsum()
df.plot(subplots=True, height=150).cols(1)

Controlling layout and targeting multiple axes

The layout of subplots can be specified using .cols(n).

df.plot(subplots=True, shared_axes=False, width=150).cols(3)

Plotting with error bars

Plotting with error bars is not supported in DataFrame.plot() and Series.plot(). To add errorbars, users should fall back to using hvplot directly.

import hvplot.pandas  # noqa

df3 = pd.DataFrame({'data1': [3, 2, 4, 3, 2, 4, 3, 2],
                    'data2': [6, 5, 7, 5, 4, 5, 6, 5]})
mean_std = pd.DataFrame({'mean': df3.mean(), 'std': df3.std()})
mean_std

	mean	std
data1	2.875	0.834523
data2	5.375	0.916125

(mean_std.plot.bar(y='mean', alpha=0.7) * \
 mean_std.hvplot.errorbars(x='index', y='mean', yerr1='std')
).opts(show_legend=False).redim.range(mean=(0, 6.5))

Plotting tables

The matplotlib backend includes support for the table keyword in DataFrame.plot() and Series.plot(). This same effect can be accomplished with holoviews by using hvplot.table and creating a layout of the resulting plots.

df = pd.DataFrame(np.random.rand(5, 3), columns=['a', 'b', 'c'])

(df.plot(xaxis=False, legend='top_right') + \
 df.T.hvplot.table()).cols(1)

Colormaps

A potential issue when plotting a large number of columns is that it can be difficult to distinguish some series due to repetition in the default colors. To remedy this, DataFrame plotting supports the use of the colormap argument, which accepts either a colormap or a string that is a name of a colormap.

To use the cubehelix colormap, we can pass colormap='cubehelix'.

df = pd.DataFrame(np.random.randn(1000, 10), index=ts.index)
df = df.cumsum()
df.plot(colormap='cubehelix')

Colormaps can also be used with other plot types, like bar charts:

dd = pd.DataFrame(np.random.randn(10, 10)).applymap(abs)
dd = dd.cumsum()
dd.plot.bar(colormap='Greens')

/tmp/ipykernel_3859/236712583.py:1: FutureWarning: DataFrame.applymap has been deprecated. Use DataFrame.map instead.
  dd = pd.DataFrame(np.random.randn(10, 10)).applymap(abs)

Parallel coordinates charts:

from bokeh.sampledata import iris
from hvplot import parallel_coordinates, andrews_curves

data = iris.flowers
parallel_coordinates(data, 'species', colormap='gist_rainbow')

Andrews curves charts:

andrews_curves(data, 'species', colormap='winter')

Plotting directly with holoviews

In some situations it may still be preferable or necessary to prepare plots directly with hvplot or holoviews, for instance when a certain type of plot or customization is not (yet) supported by pandas. Series and DataFrame objects behave like arrays and can therefore be passed directly to holoviews functions without explicit casts.

import holoviews as hv

price = pd.Series(np.random.randn(150).cumsum(),
                  index=pd.date_range('2000-1-1', periods=150, freq='B'), name='price')
ma = price.rolling(20).mean()
mstd = price.rolling(20).std()

price.plot(c='k') * ma.plot(c='b', label='mean') * \
hv.Area((mstd.index, ma - 2 * mstd, ma + 2 * mstd), 
        vdims=['y', 'y2']).opts(color='b', alpha=0.2)

This web page was generated from a Jupyter notebook and not all interactivity will work on this website.