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)
    • If you have too much trouble understanding this French Tutorial, you can read the first 6 chapters of the Tutorial in the official Python documentation and chapters 1.2.1 to 1.2.5 in the Scipy Lecture Notes. Once you have read these, you can try to read the French tutorial again
  • 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

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

cdms2 and netCDF4

There is a good chance that your input array data will come from a file in the NetCDF format.

Depending on which python distribution you are using, you can use the cdms2 or or netCDF4 modules to read the data.

cdms2

Summary: 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 UV-CDAT distribution, and can theoretically be installed independently of UV-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)

netCDF4

Summary: netCDF4 can read/write netCDF files and is available in most python distributions

Where: http://unidata.github.io/netcdf4-python/

Matplotlib

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

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 Python package for advanced map generation with a simple matplotlib interface and Iris is a Python package for analysing and visualising meteorological and oceanographic data sets

Where: Cartopy and Iris web sites

Examples:

Help on stack overflow: cartopy help

Maps and projections resources

About projections

Libraries

3D resources

Data file formats

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

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.

Pandas

Summary: pandas is a library providing high-performance, easy-to-use data structures and data analysis tools

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:

Scipy Lecture Notes

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

Where: pdf - html

This is a really nice and useful document that is regularly updated and used for the EuroScipy tutorials. You will learn more things about python, numpy and matplotlib, debugging and optimizing scripts, and also learn about using python for statistics, image processing, machine learning, washing dishes (this is just to check if you have read this page), etc…

Quick Reference and cheat sheets

Misc tutorials

  • PyFormat: With this site we try to show you the most common use-cases covered by the old and new style string formatting API with practical examples

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

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

The official Porting Python 2 Code to Python 3 page gives the required information to make the transition from python 2 to python 3. It is still safe to use Python 2.7, so there is no rush to change to Python 3.

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





[ PMIP3 Wiki Home ] - [ Help! ] - [ Wiki syntax ]

other/python/jyp_steps.txt · Last modified: 2019/11/18 12:04 by jypeter