zeppelin/docs/interpreter/python.md

layout	title	description	group
page	Python 2 & 3 Interpreter for Apache Zeppelin	Python is a programming language that lets you work quickly and integrate systems more effectively.	interpreter

{% include JB/setup %}

Python 2 & 3 Interpreter for Apache Zeppelin

Overview

Zeppelin supports python language which is very popular in data analytics and machine learning.

Name	Class	Description
%python	PythonInterpreter	Vanilla python interpreter, with least dependencies, only python environment installed is required
%python.ipython	IPythonInterpreter	Provide more fancy python runtime via IPython, almost the same experience like Jupyter. It requires more things, but is the recommended interpreter for using python in Zeppelin, see below
%python.sql	PythonInterpreterPandasSql	Provide sql capability to query data in Pandas DataFrame via pandasql

Configuration

Property	Default	Description
zeppelin.python	python	Path of the installed Python binary (could be python2 or python3). You should set this property explicitly if python is not in your $PATH(example: /usr/bin/python).
zeppelin.python.maxResult	1000	Max number of dataframe rows to display.
zeppelin.python.useIPython	true	When this property is true, %python would be delegated to %python.ipython if IPython is available, otherwise IPython is only used in %python.ipython.
zeppelin.yarn.dist.archives		Comma separated list of archives to be extracted into the working directory of interpreter. e.g. You can specify conda pack archive files via this property in python's yarn mode. It could be either files in local filesystem or files on hadoop compatible file systems
zeppelin.interpreter.conda.env.name		conda environment name, aka the folder name in the working directory of interpreter

Vanilla Python Interpreter (%python)

The vanilla python interpreter provides basic python interpreter feature, only python installed is required.

Matplotlib integration

The vanilla python interpreter can display matplotlib figures inline automatically using the matplotlib:

%python

import matplotlib.pyplot as plt
plt.plot([1, 2, 3])

The output of this command will by default be converted to HTML by implicitly making use of the %html magic. Additional configuration can be achieved using the builtin z.configure_mpl() method. For example,

z.configure_mpl(width=400, height=300, fmt='svg')
plt.plot([1, 2, 3])

Will produce a 400x300 image in SVG format, which by default are normally 600x400 and PNG respectively. In the future, another option called angular can be used to make it possible to update a plot produced from one paragraph directly from another (the output will be %angular instead of %html). However, this feature is already available in the pyspark interpreter. More details can be found in the included "Zeppelin Tutorial: Python - matplotlib basic" tutorial notebook.

If Zeppelin cannot find the matplotlib backend files (which should usually be found in $ZEPPELIN_HOME/interpreter/lib/python) in your PYTHONPATH, then the backend will automatically be set to agg, and the (otherwise deprecated) instructions below can be used for more limited inline plotting.

If you are unable to load the inline backend, use z.show(plt):

%python

import matplotlib.pyplot as plt
plt.figure()
(.. ..)
z.show(plt)
plt.close()

The z.show() function can take optional parameters to adapt graph dimensions (width and height) as well as output format (png or optionally svg).

%python

z.show(plt, width='50px')
z.show(plt, height='150px', fmt='svg')

IPython Interpreter (%python.ipython) (recommended)

IPython is more powerful than the vanilla python interpreter with extra functionality. You can use IPython with Python2 or Python3 which depends on which python you set in zeppelin.python.

For non-anaconda environment

Prerequisites

- Jupyter `pip install jupyter`
- grpcio `pip install grpcio`
- protobuf `pip install protobuf`

For anaconda environment (zeppelin.python points to the python under anaconda)

Prerequisites

- grpcio `pip install grpcio`
- protobuf `pip install protobuf`

In addition to all the basic functions of the vanilla python interpreter, you can use all the IPython advanced features as you use it in Jupyter Notebook.

e.g.

Use IPython magic

%python.ipython

#python help
range?

#timeit
%timeit range(100)

Use matplotlib

%python.ipython

%matplotlib inline
import matplotlib.pyplot as plt

print("hello world")
data=[1,2,3,4]
plt.figure()
plt.plot(data)

Colored text output

More types of visualization

e.g. IPython supports hvplot

Better code completion

By default, Zeppelin would use IPython in %python if IPython prerequisites are meet, otherwise it would use vanilla Python interpreter in %python. If you don't want to use IPython via %python, then you can set zeppelin.python.useIPython as false in interpreter setting.

Pandas integration

Apache Zeppelin Table Display System provides built-in data visualization capabilities. Python interpreter leverages it to visualize Pandas DataFrames though similar z.show() API, same as with Matplotlib integration.

Example:

%python

import pandas as pd
rates = pd.read_csv("bank.csv", sep=";")
z.show(rates)

SQL over Pandas DataFrames

There is a convenience %python.sql interpreter that matches Apache Spark experience in Zeppelin and enables usage of SQL language to query Pandas DataFrames and visualization of results though built-in Table Display System.

Prerequisites

• Pandas pip install pandas
• PandaSQL pip install -U pandasql

Here's one example:

• first paragraph

%python

import pandas as pd
rates = pd.read_csv("bank.csv", sep=";")

• next paragraph

%python.sql

SELECT * FROM rates WHERE age < 40

Using Zeppelin Dynamic Forms

You can leverage Zeppelin Dynamic Form inside your Python code.

Example :

%python

### Input form
print(z.input("f1","defaultValue"))

### Select form
print(z.select("f2",[("o1","1"),("o2","2")],"o1"))

### Checkbox form
print("".join(z.checkbox("f3", [("o1","1"), ("o2","2")],["o1"])))

ZeppelinContext API

Python interpreter create a variable z which represent ZeppelinContext for you. User can use it to do more fancy and complex things in Zeppelin.

API	Description
z.put(key, value)	Put object value with identifier key to distributed resource pool of Zeppelin, so that it can be used by other interpreters
z.get(key)	Get object with identifier key from distributed resource pool of Zeppelin
z.remove(key)	Remove object with identifier key from distributed resource pool of Zeppelin
z.getAsDataFrame(key)	Get object with identifier key from distributed resource pool of Zeppelin and converted into pandas dataframe. The object in the distributed resource pool must be table type, e.g. jdbc interpreter result.
z.angular(name, noteId = None, paragraphId = None)	Get the angular object with identifier name
z.angularBind(name, value, noteId = None, paragraphId = None)	Bind value to angular object with identifier name
z.angularUnbind(name, noteId = None)	Unbind value from angular object with identifier name
z.show(p)	Show python object p in Zeppelin, if it is pandas dataframe, it would be displayed in Zeppelin's table format, others will be converted to string
z.textbox(name, defaultValue="")	Create dynamic form Textbox name with defaultValue
z.select(name, options, defaultValue="")	Create dynamic form Select name with options and defaultValue. options should be a list of Tuple(first element is key, the second element is the displayed value) e.g. z.select("f2",[("o1","1"),("o2","2")],"o1")
z.checkbox(name, options, defaultChecked=[])	Create dynamic form Checkbox `name` with options and defaultChecked. options should be a list of Tuple(first element is key, the second element is the displayed value) e.g. z.checkbox("f3", [("o1","1"), ("o2","2")],["o1"])
z.noteTextbox(name, defaultValue="")	Create note level dynamic form Textbox
z.noteSelect(name, options, defaultValue="")	Create note level dynamic form Select
z.noteCheckbox(name, options, defaultChecked=[])	Create note level dynamic form Checkbox
z.run(paragraphId)	Run paragraph
z.run(noteId, paragraphId)	Run paragraph
z.runNote(noteId)	Run the whole note

Run Python in yarn cluster

Zeppelin supports to run python interpreter in yarn cluster which means the python interpreter runs in the yarn container. This can achieve better multi-tenant for python interpreter especially when you already have a hadoop yarn cluster.

But there's one critical problem to run python in yarn cluster: how to manage the python environment in yarn container. Because yarn cluster is a distributed cluster environemt which is composed many nodes, and your python interpreter can start in any node. It is not practical to manage python environment in each nodes.

So in order to run python in yarn cluster, we would suggest you to use conda to manage your python environment, and Zeppelin can ship your codna environment to yarn container, so that each python interpreter can has its own python environment.

Step 1

We would suggest you to use conda pack to create archives of conda environments, and ship it to yarn container. Otherwise python interpreter will use the python executable in PATH of yarn container.

Here's one example of yml file which could be used to generate a conda environment with python 3 and some useful python libraries.

• Create yml file for conda environment, write the following content into file env_python_3.yml

name: python_3_env
channels:
  - conda-forge
  - defaults
dependencies:
  - python=3.7 
  - pycodestyle
  - numpy
  - pandas
  - scipy
  - grpcio
  - protobuf
  - pandasql
  - ipython
  - ipykernel
  - jupyter_client
  - panel
  - pyyaml
  - seaborn
  - plotnine
  - hvplot
  - intake
  - intake-parquet
  - intake-xarray
  - altair
  - vega_datasets
  - pyarrow

• Create conda environment via this yml file using either conda or mamba

conda env create -f env_python_3.yml

mamba env create -f python_3_env

• Pack the conda environment using either conda

conda pack -n python_3

Step 2

Specify the following properties to enable yarn mode for python interpreter, and specify the correct python environment.

zeppelin.interpreter.launcher yarn
zeppelin.yarn.dist.archives /home/hadoop/python_3.tar.gz#environment
zeppelin.interpreter.conda.env.name environment

zeppelin.yarn.dist.archives is the python conda environment tar which is created in step 1. This tar will be shipped to yarn container and untar in the working directory of yarn container. environment in /home/hadoop/python_3.tar.gz#environment is the folder name after untar. This folder name should be the same as zeppelin.interpreter.conda.env.name.

Python environments (used for non-yarn mode)

Default

By default, PythonInterpreter will use python command defined in zeppelin.python property to run python process. The interpreter can use all modules already installed (with pip, easy_install...)

Conda

Conda is an package management system and environment management system for python. %python.conda interpreter lets you change between environments.

Usage

• get the Conda Information:

  %python.conda info
  

• list the Conda environments:

  %python.conda env list
  

• create a conda enviornment:

  %python.conda create --name [ENV NAME]
  

• activate an environment (python interpreter will be restarted):

  %python.conda activate [ENV NAME]
  

• deactivate

  %python.conda deactivate
  

• get installed package list inside the current environment

  %python.conda list
  

• install package

  %python.conda install [PACKAGE NAME]
  

• uninstall package

  %python.conda uninstall [PACKAGE NAME]
  

Docker

%python.docker interpreter allows PythonInterpreter creates python process in a specified docker container.

Usage

• activate an environment

  %python.docker activate [Repository]
  %python.docker activate [Repository:Tag]
  %python.docker activate [Image Id]
  

• deactivate

  %python.docker deactivate
  

Here is an example

# activate latest tensorflow image as a python environment
%python.docker activate gcr.io/tensorflow/tensorflow:latest

Technical description

For in-depth technical details on current implementation please refer to python/README.md.

Some features not yet implemented in the vanilla Python interpreter

• Interrupt a paragraph execution (cancel() method) is currently only supported in Linux and MacOs. If interpreter runs in another operating system (for instance MS Windows) , interrupt a paragraph will close the whole interpreter. A JIRA ticket (ZEPPELIN-893) is opened to implement this feature in a next release of the interpreter.
• Progression bar in webUI (getProgress() method) is currently not implemented.