Python cheat sheet

This document is a memos about programming issue in pythons.

Standalone packaging

There are no more words needed for python’s fast reproduction and redistribution of source codes to developers and researchers. However, the executable package distribution is not the main scope of python. Unlikely many compile languages(C/C++, Rust, … ), the execution of python source code requires python environment and dependent packages. Have you ever seen any program requiring a compiler and libraries that the developer pawns off on users? Of course, some scientific programs or mass-size products, like games, demand some prerequisites but the most of dependencies are automatically installed. Moreover, for general users, constructing these environments is usually a merciless demand.

Pyinstaller is one solution of python distibution. There are other solutions about distibution of executable program.

See details of usage in official documents

This section describes some essential infos for fast referring.

Beaware using Libraries( especially scientific library )

Some mathematical routines, like matrix operations, are requiring additional modules or you have to implement yourself for your project. If the project requires stable, precise, and high speed operations and there are tones of types of operations, it is vital to use addtional libraries, considering fast impelmentation. However, those libraries commonly consist of large number of functionalities. It makes their size huge for including them in executable file with pyinstaller.

One example of those libraries is numpy. It is commonly included in many scientific applications and provides us various $n$-dimensional array modulation routines. However, if you only requires little matrix opeation, maybe 2 to 4?, including numpy to your project squanders lots of storage resources. At least, it requires 200MB in Windows, and a half a GB in Linux system.

C-Python interface

In some case, you would feel desire to write a core algorithm with C to accelerate overall speed of the implemeted routines. However, if the function is frequently called in python evironment to follow or to update the internal variables, you may not write such function with C. The transformation cost from python to C and reverse direction requires huge resources much more than you expected at first. To achieve a better performance, the implmented routine must minize data transfer during the execution. That means, C implemented part of the program could account for almost all parts of the routines.

Good examples are numpy and pandas, they create, manipulate the datatypes(ndarray, dataframe) inside C routines and never exchange the internal variables during the process.

Same argument also hold for binary operation. It is true that binary tricks with IEEE standard number system could provide better performance. However, Python is not a C and int(python) != int(C). Standard string class routines would show much faster performance than your bitwise implementation in most cases.

Tkinter

Icon issue in linux

tkinter provides two icon setting, iconbitmap and iconphoto. The iconbitmap is a icon on program windows where the left top side of windows. The iconphoto is a task bar indication icon. In Linux operating system, tkinter does not support .ico file format for iconbitmap. It only supports .xbm file format. It is meaningless to change the name of file. They are totally different file format. Gimp support xbm export of image. However, practically, iconbitmap routines are not work properly in many situation if gui application is on Linux system.