The author has 2 main goals:
1) To improve the productivity of scientists familiar with specific software systems (especially Matlab, Maple, and Mathematica) by teaching them to "glue" applications together.
2) To advocate Python as the preferred "glue" language. In his own words, "I hope to convince computational scientists having experience with Perl that Python is a preferable alternative, especially for large long-term projects."
He has certainly done a creditable job. As an expert in computational differential equations, he neglects neither efficiency nor correctness, while stressing both simplicity and reliability. In this sense, he has done a great service to the Python community.
The question is: What justifies the purchase of his book?
The answer is: Chapters 4, 9, and 10.
Very convincing arguments.
2. Getting Started With Python Scripting--38pp
3. Basic Python--56pp
A too-quick tutorial. Go to python dot org instead.
4. Numerical Computing in Python--48pp
Stellar explanations of vectorized array operations.
5. Combining Python with Fortran, C, and C++--36pp
Details use of Fortran2Py and SWIG. Mentions many alternatives.
6. Introduction to GUI Programming--70pp
Useful examples of Tkinter/pmw widgets.
7. Web Interfaces and CGI Programming--24pp
Good source of ideas.
8. Advanced Python--132pp
Deep and extensive. Includes: option parsing, regular expressions, data persistence and compression, object-oriented programming, exceptions, generic programming, efficiency.
9. Fortran Programming with NumPy Arrays--32pp
All about efficiency and re-use.
10. C and C++ Programming with NumPy Arrays--40pp
More about efficiency. NumPy C API, C++ objects, and SCXX.
11. More Advanced GUI Programming--73pp
Tedious discussion of both Web and standalone GUIs. BLT, canvas, cgi.
12. Tools and Examples--70pp
Excellent examples of PDE solvers, with a powerful GUI, but quite long and tedious.
A. Setting up the Required Software Environment--16pp
Wonderfully specific installation instructions!
B. Elements of Software Engineering--50pp
Python's strength! Very practical advice on modularity, documentation, coding style, regression-testing, version-control.
+ Downloadable py4cs package, esp. numpytools module
+ Great advice everywhere, e.g. CGI checklist, Pythonic programming, and trouble-shooting.
+ Concrete evidence for most assertions.
+ Very attractive presentation. Sturdy, high-quality cover, binding and pages. Brief, elegant code fragments (except in Chapter 12). Readable prose. No wasted space.
+ Available as 5MB pdf file, after purchase of hardcopy. Very nice.
+ Slides, installation instructions, and errata also at web site. Very professional.
- Not enough tables to be a useful manual.
- On p.428(#7) he points out that handling a raised exception is very slow. However, when I time his example with a positive argument, the try-except version is 20% faster (b/c the if clause is skipped), so he is actually giving bad advice for the general case. Luckily, he contradicts himself later, on page 685: "Exceptions should be used instead of if-else tests." The best advice: Avoid common exceptions in inner loops.
- The 10-page index is not as great as it at first seems. (See Martelli's Python in a Nutshell for a better one.)
- Pure interface functions should 'raise NotImplementedError', rather than 'return'.
- Exceptions should never be trapped mindlessly with 'except:'. That would hide your own SyntaxErrors!
- Too many exercises. (It's published as a textbook.) Since there are no answers, the exercises are useless for non-students. (See Lutz's Learning Python for effective exercises with answers.)
This contains the best information on numerical programming in Python that I've seen. Though expensive, it could easily be your only Python book, given the excellent online documenation already available.