What is HoloViews

HoloViews is a wrapper that makes it easier to use visualization tools such as matplotlib and Bokeh in broken terms. Visualization can be implemented with unified code and simple description using any backend (matplotlib, Bokeh, etc.).

The learning cost of visualization is not very high, so by using HoloViews ** "Let's concentrate on the original data analysis and other tasks instead of wasting time writing visualization code" ** It seems to be a feeling concept. In particular, the syntax of matplotlib is often esoteric, so many people may be wondering, "Isn't it a little simpler?"

Back end

The following visualization tools can be used as a backend.

matplotlib
Bokeh
Plotly(experimental)

Installation

When installing with conda

conda install -c ioam holoviews bokeh

If you use matplotlib or plotly for the backend described later, install it additionally.

conda install matplotlib plotly

When installing with pip

[2017/07/30 postscript] There seem to be some Options (http://holoviews.org/user_guide/Installing_and_Configuring.html), but try installing with'recommended'.

pip install numpy param
pip install 'holoviews[recommended]'

Install the package used for the backend. There is no problem if you exclude unused items from the installation target.

pip install matplotlib bokeh plotly

When installing the latest version

What is git? Those who say it is better to install with the above conda or pip.

git clone git://github.com/ioam/holoviews.git
cd holoviews
pip install -e .

Jupyter Notebook launch options

If the notebook version is 5.0, set the following startup options.

jupyter notebook --NotebookApp.iopub_data_rate_limit=100000000

In later versions, just jupyter notebook seems to be fine.

Backend settings

Specify the backend as an argument of the notebook_extension class.

import holoviews as hv

hv.extension('matplotlib')

When running on Jupyter Notebook, it can be set with cell magic.

%%output backend='matplotlib'

Try to output the graph

The following are all execution examples on Jupyter Notebook. matplotlib

import holoviews as hv

hv.extension('matplotlib')
curve = hv.Curve(([1, 2, 3], [1, 2, 4]))
curve

Bokeh

hv.extension('bokeh')
curve = hv.Curve(([1, 2, 3], [1, 2, 4]))
curve

Plotly

hv.extension('plotly')
curve = hv.Curve(([1, 2, 3], [1, 2, 4]))
curve

I was able to write the same code with any backend.

compose

I think many people say that arranging multiple figures with matplotlib was quite difficult. HoloViews allows you to arrange and overlay graphs simply by using operators.

Arrange with + operator

hv.extension('bokeh')
curve = hv.Curve(([1, 2, 3], [1, 2, 4]))
bars = hv.Bars((['a', 'b', 'c'], [3, 2, 1]))
curve + bars

* Overlap with operators

%%opts Scatter(color='green', size=10)
scatter = hv.Scatter(([1, 2, 3], [1, 2, 4]))
curve * scatter

Combine + and * operators

curve * scatter + bars

Features of HoloViews

HoloViews has various functions. If I have a chance, I would like to introduce them one by one.

Drawing live data
Processing and drawing of table data
Processing and drawing of multidimensional data
Extraction of drawing data
Event handling
Interactive and exploratory visualization
Large-scale data using Datashader

-> Continued: Basic graph of HoloViews in one liner

HoloViews may become the standard for Python visualization tools