numpy tutorial

- understanding numpy iterators
	- iterating through an array gets the rows/subarrays
	- iterating through .flat gets a flat C-ordering (last index in) run-through
    - ndenumerate returns index and object
    - also a way to convert flat index to tuple (and perhaps vice versa)

- dtype objects
    - reproduce standard types
	- construct generic (nested) types
	- sit in array header, tell us what's in the records; points to scalar type (Python type); have same attributes as arrays
    - 21 standard scalar types belong to hierarchy; some (int_, float_) enhance the corresponding Python type
    - the scalar types are classes; dtypes are a single class
    - byte order is handled in dtypes

- advanced array slicing
	- slicing returns another view of the same data
	- advanced slicing is indexing by array objects
    - b[[0,1],[2,4]] returns b[0,2] b[1,4]
    - this is element by element behavior; the MATLAB cross product behavior is given by ix_
    - b[:,[1,2]]     returns b[:,1] b[:,2]
    	- works by broadcasting
	- masked arrays: created boolean array, use it to index
	- a lot of this can be done with take	
	- always copies data (demarcation by simple and fancy indexing; integers and slicing objects are always simple)

- writing C code
	- numpy supports all the Numeric API
	- but all the dimensions and strides are integers the same size as the pointer on the platform
		- some calls remain that take int, and convert; a slightly different call would avoid the conversion
	- simple function to create array, PyArray_FROM_OTF
    - in book: using iterators, broadcast mechanism
    - numpy include in nonstandard location for compatibility with Python eggs, which have to be self-contained
    - numpy distutils will take that into account; there's also a get_include command

- ctypes
	- straightforward (but cumbersome) way of directly calling C shared libraries from Python
	- requires some checks to be done on the Python side
	- and requires some introspection on the library (get the method names, etc.)

- pyrex
	- lets you write code that mixes Python and C data types any way you want, and compiles it into a C extension for Python

- weave
	- very nice, will compile on the fly, speed up array computations
	- or inline C/C++ code

- memory mapped arrays
	- are useful for reading and/or modifying small segments of a large file with regular layout
	- limited to intsize-dependent filesize of platform
    - interesting also for files, generally

- misc
	- empty creates empty array, float by default
    - look into Python eggs (jars for Python; platform-dependent if they include C extensions)
    - look into numpy distutils
    - array interface between Numeric, numarray, numpy; also PIL (1.1.6)
		- if "nocopy" specified, no memory hit
	- pytables: managing hierarchical datasets
	- pysparse: sparse matrices (Numeric-based)