User Tools

Site Tools


other:python:jyp_steps

JYP's recommended steps for learning python

If you don't know which python distribution to use and how to start the python interpreter, you should first read the Working with Python page

As can be expected, there is a lot of online python documentation available, and it's easy to get lost. You can always use google to find an answer to your problem, and you will probably end up looking at lots of answers on Stack Overflow or a similar site. But it's always better to know where you can find some good documentation… and to spend some time to read the documentation

This page tries to list some python for the scientist related resources, in a suggested reading order. Do not print anything (or at least not everything), but it's a good idea to download all the pdf files in the same place, so that you can easily open and search the documents

JYP's introduction to python

Part 1

You can start using python by reading the Bien démarrer avec python tutorial that was used during a 2013 IPSL python class:

  • this tutorial is in French (my apologies for the lack of translation, but it should be easy to understand)
  • it's an introduction to python (and programming) for the climate scientist: after reading this tutorial, you should be able to do most of the things you usually do in a shell script
    • python types, tests, loops, reading a text file
    • the tutorial is very detailed about string handling, because strings offer an easy way to practice working with indices (indexing and slicing), before indexing numpy arrays. And our usual pre/post-processing scripts often need to do a lot of string handling in order to generate the file/variable/experiment names
  • after reading this tutorial, you should practice with the following:

Part 2

Once you have done your first steps, you should read Plus loin avec Python (start at page 39, the previous pages are an old version of what was covered in Part 1 above)

  • this tutorial is in French (sorry again)
  • after reading this tutorial, you will be able to do more than you can do in a shell script, in an easier way
    • advanced string formatting
    • creating functions and using modules
    • working with file paths and handling files without calling external Linux programs
      (e.g. using os.remove(file_name) instead of rm $file_name)
    • using command-line options for scripts, or using configuration files
    • calling external programs

The official python documentation

You do not need to read all the python documentation at this step, but it is really well made and you should at least have a look at it. The Tutorial is very good, and you should have a look at the table of content of the Python Standard Library. There is a lot in the default library that can make your life easier

Python 2.7

Python 3

Scientific Python Lectures

Summary: One document to learn numerics, science, and data with Python

Note: this used to be called Scipy Lecture Notes

Where: pdf - html

This is a really nice and useful document that is regularly updated and used for the EuroScipy tutorials.

This document will teach you lots of things about python, numpy and matplotlib, debugging and optimizing scripts, and about using python for statistics, image processing, machine learning, washing dishes (this is just to check if you have read this page), etc…

Numpy and Scipy

Summary: Python provides ordered objects (e.g. lists, strings, basic arrays, …) and some math operators, but you can't do real heavy computation with these. Numpy makes it possible to work with multi-dimensional data arrays, and using array syntax and masks (instead of explicit nested loops and tests) and the apropriate numpy functions will allow you to get performance similar to what you would get with a compiled program! Scipy adds more scientific functions

Where: html and pdf documentation

Getting started

  1. always remember that indices start at 0 and that the last element of an array is at index -1!
    First learn about indexing and slicing by manipulating strings, as shown in Part 1 above (try 'This document by JY is awesome!'[::-1] and 'This document by JY is awesome!'[slice(None, None, -1)]) 8-)
  2. if you are a Matlab user (but the references are interesting for others as well), you can read the following:
    1. NumPy for MATLAB users (nice, but does not seem to be maintained any more)
  3. read the really nice numpy Quickstart tutorial
  4. have a quick look at the full documentation to know where things are
    1. Numpy User Guide
    2. Numpy Reference Guide
    3. Scipy Reference Guide

Beware of the array view side effects

When you take a slice of an array, you get a View : an array that has a new shape but that still shares its data with the first array.

That is not a problem when you only read the values, but if you change the values of the View, you change the values of the first array (and vice-versa)! If that is not what want, do not forget to make a copy of the data before working on it!

Views are a good thing most of the time, so only make a copy of your data when needed, because otherwise copying a big array will just be a waste of CPU and computer memory. Anyway, it is always better to understand what you are doing… :-P

Check the example below and the copies and views part of the quickstart tutorial.

>>> import numpy as np
>>> a = np.arange(30).reshape((3,10))
>>> a
array([[ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9],
       [10, 11, 12, 13, 14, 15, 16, 17, 18, 19],
       [20, 21, 22, 23, 24, 25, 26, 27, 28, 29]])
 
>>> b = a[1, :]
>>> b
array([10, 11, 12, 13, 14, 15, 16, 17, 18, 19])
 
>>> b[3:7] = 0
>>> b
array([10, 11, 12,  0,  0,  0,  0, 17, 18, 19])
 
>>> a
array([[ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9],
       [10, 11, 12,  0,  0,  0,  0, 17, 18, 19],
       [20, 21, 22, 23, 24, 25, 26, 27, 28, 29]])
 
>>> a[:, 2:4] = -1
>>> a
array([[ 0,  1, -1, -1,  4,  5,  6,  7,  8,  9],
       [10, 11, -1, -1,  0,  0,  0, 17, 18, 19],
       [20, 21, -1, -1, 24, 25, 26, 27, 28, 29]])
 
>>> b
array([10, 11, -1, -1,  0,  0,  0, 17, 18, 19])
 
>>> c = a[1, :].copy()
>>> c
array([10, 11, -1, -1,  0,  0,  0, 17, 18, 19])
 
>>> c[:] = 9
>>> c
array([9, 9, 9, 9, 9, 9, 9, 9, 9, 9])
 
>>> b
array([10, 11, -1, -1,  0,  0,  0, 17, 18, 19])
 
>>> a
array([[ 0,  1, -1, -1,  4,  5,  6,  7,  8,  9],
       [10, 11, -1, -1,  0,  0,  0, 17, 18, 19],
       [20, 21, -1, -1, 24, 25, 26, 27, 28, 29]])

Extra numpy information

You can also check the numpy section of the Useful python stuff page

Using NetCDF files with Python

What is NetCDF?

  • If you are working with climate model output data, there is a good chance that your input array data will be stored in a NetCDF file!
  • There may be different ways of dealing with NetCDF files, depending on which python distribution you have access to

CliMAF and C-ESM-EP

People using CMIPn and model data on the IPSL servers can easily search and process NetCDF files using:

xarray

xarray makes working with labelled multi-dimensional arrays in Python simple, efficient, and fun! […] It is particularly tailored to working with netCDF files

Note: more packages (than listed below) may be listed in the Extra packages list page

  • xCDAT: xarray extended with Climate Data Analysis Tools
  • xoa: xarray-based ocean analysis library

netCDF4

netCDF4 is a Python interface to the netCDF C library

cdms2

  • cdms2 is unfortunately not maintained anymore and is slowly being phased out in favor of a combination of xarray and xCDAT

cdms2 can read/write netCDF files (and read grads dat+ctl files) and provides a higher level interface than netCDF4. cdms2 is available in the CDAT distribution, and can theoretically be installed independently of CDAT (e.g. it will be installed when you install CMOR in conda). When you can use cdms2, you also have access to cdtime, that is very useful for handling time axis data.

How to get started:

  1. read JYP's cdms tutorial, starting at page 54
    1. the tutorial is in French (soooorry!)
    2. you have to replace cdms with cdms2, and MV with MV2 (sooorry about that, the tutorial was written when CDAT was based on Numeric instead of numpy to handle array data)
  2. read the official cdms documentation (link may change)

Matplotlib

The full content of this matplotlib section has been moved to
Working with matplotlib (JYP version)
after becoming too big to manage here


Note: Plotting maps with matplotlib+cartopy (examples provided by JYP)

Summary: there are lots of python libraries that you can use for plotting, but Matplotlib has become a de facto standard

Where: Matplotlib web site

Help on stack overflow: matplotlib help

Basemap

Basemap is going to be slowly phased out, in favor of cartopy
More information in this:

Summary: Basemap is an extension of Matplotlib that you can use for plotting maps, using different projections

Where: Basemap web site

Help on stack overflow: basemap help

How to use basemap?

  1. look at the examples
  2. read some documentation!
    1. the really nice basemap tutorial seems much better than the official documentation below

Cartopy + Iris

Summary:

  • Cartopy is a matplolib-based Python package designed for geospatial data processing in order to produce maps and other geospatial data analyses
  • Iris is a powerful, format-agnostic, community-driven Python package for analysing and visualising Earth science data.

Where: Cartopy and Iris web sites

Examples:

Help on stack overflow: Cartopy help - Iris help

Maps and projections resources

About projections

Libraries

3D plots resources

Data analysis

EDA (Exploratory Data Analysis) ?

The EDA concept seems to apply to time series (and tabular data), which is not exactly the case of full climate model output data
    • The method of studying and exploring record sets to apprehend their predominant traits, discover patterns, locate outliers, and identify relationships between variables. EDA is normally carried out as a preliminary step before undertaking extra formal statistical analyses or modeling.

Easy to use datasets

If you need standard datasets for testing, example, demos, …

  • CMIP6 data on ESGF
    • Example : orog_fx_IPSL-CM6A-LR_piControl_r1i1p1f1_gr.nc:

Pandas

Summary: pandas is a fast, powerful, flexible and easy to use open source data analysis and manipulation tool

Where: Pandas web site

JYP's comment: pandas is supposed to be quite good for loading, processing and plotting time series, without writing custom code. It is very convenient for processing tables in xlsx files (or csv, etc…). You should at least have a quick look at:

statsmodels

statsmodels is a Python module that provides classes and functions for the estimation of many different statistical models, as well as for conducting statistical tests, and statistical data exploration.

Note: check the example in the Statistics in Python tutorial

scikit-learn

scikit-learn is a Python library for machine learning, and is one of the most widely used tools for supervised and unsupervised machine learning. Scikit–learn provides an easy-to-use, consistent interface to a large collection of machine learning models, as well as tools for model evaluation and data preparation

Note: check the example in scikit-learn: machine learning in Python

scikit-image

scikit-image is a collection of algorithms for image processing in Python

Note: check the example in scikit-image: image processing

YData Profiling

YData Profiling: a leading package for data profiling, that automates and standardizes the generation of detailed reports, complete with statistics and visualizations.

D-Tale

D-Tale brings you an easy way to view & analyze Pandas data structures. It integrates seamlessly with ipython notebooks & python/ipython terminals.

Sweetviz

Sweetviz is pandas based Python library that generates beautiful, high-density visualizations to kickstart EDA (Exploratory Data Analysis) with just two lines of code.

AutoViz

AutoViz: the One-Line Automatic Data Visualization Library. Automatically Visualize any dataset, any size with a single line of code

Data file formats

  • We list below some resources about non-NetCDF data formats that can be useful

The shelve package

The built-in shelve package, can be easily used for storing data (python objects like lists, dictionaries, numpy arrays that are not too big, …) on disk and retrieving them later

Use case:

  1. Use a script do to the heavy data pre-processing and store the (intermediate) results in a file using shelve, or update the results
  2. Use another script for plotting the results stored with shelve. This way you don't have to wait for the pre-processing step to finish each time you want to improve your plot(s)

Warning:

  • read the documentation and the example carefully (it's quite small)
    • if you get the impression that the data is not saved correctly, re-read the parts about updating correctly the content of the shelve file
    • you should be able to store most python objects in a shelve file, but it is safer to make tests
  • do not forget to close the output file
  • if you are dealing with big arrays and want to avoid performance issues, you should use netCDF files for storing the intermediate results

json files

More and more applications use json files as configuration files or as a mean to use text files to exchange data (through serialization/deserialization ).

json files look basically like a list of (nested) python dictionaries that would have been dumped to a text file

  • json module documentation
  • example script: /home/users/jypeter/CDAT/Progs/Devel/beaugendre/nc2json.py
  • A compact (not easy to read…) json file can be pretty-printed with
    cat file.json | python -m json.tool | less

LiPD files

BagIt files

BagIt, a set of hierarchical file layout conventions for storage and transfer of arbitrary digital content.

Protocol Buffers

Protocol Buffers are (Google's) language-neutral, platform-neutral extensible mechanisms for serializing structured data

Quick Reference and cheat sheets

Miscellaneous Python stuff and tutorials

Some good coding tips

Debugging your code

There is only so much you can do with staring at your code in your favorite text editor, and adding print lines in your code (or using logging instead of print). The next step is to use the python debugger!

Debugging in text mode

  1. Start the script with: python -m pdb my_script.py
  2. Type run (or r) to go to the first line of the script
  3. Type continue (or c) to execute the script to the end, or till the first breakpoint or error is reached
  4. Use where (or w) to check the call stack that led to the current stop. Use up and down to navigate through the call stack and examine the values of the functions' parameters
  5. Type break NNN to stop at line NNN
  6. Use type(var) and print var to check the type and values of variables. You can also change the variables' values on the fly!
  7. Type run (or r) to restart the script
  8. Use next and step to execute some parts of the script line by line. If a code line calls a function:
    • next (or n) will execute a function and stop on the next line
    • step (or s) will stop at the first line inside the function
  9. Check the debugger commands for details, or type help in the debugger for using the built-in help

Using pydebug

Depending on the distribution, the editor and the programming environment you use, you may have access to a graphical version of the debugger. UV-CDAT users can use pydebug my_script.py

jupyter and notebook stuff

FIXME Misc notes, resources and links to organize later

  • jupyter {book}: Jupyter Book is an open source project for building beautiful, publication-quality books and documents from computational material.

Using a Python IDE

IDE = Integrated Development Environment

There are lots of ways to use Python and develop scripts, from using a lightweight approach (your favorite text editor with builtin python syntax highlighting, e.g. emacs and python -i myscript.py) to a full-fledged IDE. You'll find below some IDE related links

Spyder

Improving the performance of your code

You can already get a very efficient script by checking the following:

  • make sure that your script is not using too much memory (the amount depends on the computer you are using)! Your script should be scalable (e.g. keeps on working even when your data gets bigger), so it's a good idea to load only the data you need in memory (e.g. not all the time steps), and learn how to load chunks of data
  • make sure that you are using array/vector syntax and masks, instead of using explicit loops and tests. The numpy documentation is big, because there are lots of optimized functions to help you! If you are stuck, ask JY or somebody else who is used to numpy.

If your script is still not fast enough, there is a lot you can do to improve it, without resorting to parallelization (that may introduce extra bugs rather that extra performance). See the sections below

Hint: before optimizing your script, you should spent some time profiling it, in order to only spend time improving the slow parts of your script

Useful packages

  • Numexpr: Numexpr is a fast numerical expression evaluator for NumPy. With it, expressions that operate on arrays (like “3*a+4*b”) are accelerated and use less memory than doing the same calculation in Python.
  • PyTables: PyTables is a package for managing hierarchical datasets and designed to efficiently and easily cope with extremely large amounts of data

Tutorials by Ian Osvald

Python 2.7 vs Python 3

It is still safe to use Python 2.7, but you should consider upgrading to Python 3, unless some key modules you need are not compatible (yet) with Python 3

You should start writing code that will, when possible, work both in Python 2 and Python 3

Some interesting reading:

    • print is now a function. Use print('Hello')
    • You cannot test a difference with <> any longer! Use !=

What now?

You can do a lot more with python! But if you have read at least a part of this page, you should be able to find and use the modules you need. Make sure you do not reinvent the wheel! Use existing packages when possible, and make sure to report bugs or errors in the documentations when you find some

Out-of-date stuff

other/python/jyp_steps.txt · Last modified: 2024/08/27 12:51 by jypeter