=== Alex Martelli - programming for scientists and engineers ===

- engineering vs. science
- why care about the details?
- do black boxes rule? No, nature and numbers are not in the business of making your life simpler by ensuring computations are well conditioned and behaved
- Acton, real computing made real: good computing routines are problem-specific
	- example: computing difference of two large numbers, such as sqrt(x+1) - sqrt(x); recast as 1/[sqrt(x+1) + sqrt(x)]
	- software has bugs (reliable software has bugs that bite you rarely)
	- if numbers are important, compute them twice; different compilers, different computers, optimization levels
	- numerical software packages always hide deficiencies
- but: in many disciplines, sampling bias kills you (meta-issue: publication bias); how good are measurements in the first place?
- Spolsky: all nontrivial abstractions, to some degree, are leaky
	- so black boxes are useful, but leaky
	- need to know what's inside; need to care; need to learn what to do about it
- today can code fast, elegantly, clearly, numerically efficiently, but have to care
- black boxes are necessary evil

- good computing tools are not about replacing human intelligence, but rather about enabling its application
- wrong computations always have consequences, even in research


=== Travis Vaught, Jarrod Millman - state of scipy ===

- numeric (1995), scipy (2001), matplotlib (2003), numarray (2003), numpy (2006)
- numpy book now in the public domain
- numpy recent changes: MaskedArrays, IO code, test coverage, numscons, coding style
- blog: planet.scipy.org


=== Stéfan Van der Walt - numpy/scipy documentation ===

- wiki system to allow community to contribute
- scipy documentation editor by Pauli Virtonen
- workflow: docstring needs writing -> needs editing -> needs reviewing -> needs proofing -> proofed
- found bugs
- numpy.doc infrastructure

- reference guide using Sphinx
- docstrings extracted from numpy, put in intermediate representation, sent to sphinx

- pydocweb editor available in launchpad

- what about propagating changes in the source?
	- changes to docstrings and new functions are pulled in with appropriate status
	- changes to code are _not_ pulled in

- what about development documentation? (info in SVN, etc.)


=== Michael Droettboom - matplotlib and Sphinx ===

- sphinx:
	- printed (PDF) and online (HTML) generated from same source
	- all reStructuredText
	- more attractive
	- searchable, indexed, cross-referenced
- can pull in docstrings using autodoc plugin
- matplotlib.pyplot allows inserting inline plots and latex formulas
	- math engine is quite complete, supports also stix...
	- ...but sphinx has dvipng and jsMath plugins
- intersphinx links

- presentation given from a bookmark folder in web browser


=== Lightning talks ===

- Fernando Perez: ipython 1.0 out soon, many improvements (WX and Cocoa shells), fold in ipython1 features