=== Stefan van der Walt - numpy tricks ===

- broadcasting
	- to combine arrays of different sizes
	- simple example is scalar broadcasting
	- x, y = np.ogrid[1:100,1:100] 		# each vector one-dimensional, x column, y row 
	- g = x + 1j*y						# create grid by broadcasting
	- better than np.mgrid, which generates two 2-d arrays
	- broadcasting compares dimensions starting from last
	- can add dimension with "None" index to allow broadcasting
- fancy indexing
	- with another array, not with scalars and slices
	- extract diagonal from 3x3 array
		- a[[1,2,3],[1,2,3]]
	- array indices are broadcast against one another...
	- ...then slices used as is
- structured arrays
	- dt = np.dtype([('time',np.uint64),('position',[('az',np.float32),
													 ('el',np.float32)])])
													
	- a = np.fromfile(filename,dtype=dt)	# from binary
											# loadtxt would do similarly (see manpage)
	- endianness: np.dtype('<i4'), np.dtype('>i4')
- strided tricks
	- a.strides gives the bytes that must be skipped in memory to move between elements along each direction
	- np.stride_tricks.as_strided(x,shape=(3,2,5),strides=(20,20,4))
- subclassing
	- __new__, __array_finalize__
	- see http://www.scipy.org/Subclasses

- more hidden features:
	- np.save, np.load
	- np.fromregex
	- np.lib.Arrayterator
	- np.emath
	- np.unravel_index
	- np.lexsort
	- np.iinfo
	- np.issubdtype
	- np.linalg.matrix_power


=== Eric Jones - Traits ===

- ETS
- Chaco: complex, interactive 2D plots
- Kiva: 2D primitives
- Mayavi: 3D visualization with VTK
- Enable: object-based 2D drawing canvas
- Envisage: application plugin framework

- Traits add five characteristics to Python object attributes
	- initialization
	- validation
	- delegation
	- notification	-> provides interactivity
	- visualization	-> scales well

>>> from enthought.traits.api import HasTraits, Float
>>> 
>>> class Rectangle(HasTraits):
>>> 	width = Float			# width = Float(1.0) would set default
>>> 	height = Float
>>> 
>>> rect = Rectangle(width = 2.0)

- casting traits such as CFloat will use Python's float() to convert
- Any trait will behave as an untyped Python variable

- "Property NAME" trait: will call _get_NAME(self), _set_NAME(self)
- can pass "depends_on=[...,...]" to Property to let is know when it changes
- then the decorator @cached_property can be used on _get methods to cache values

- can define a default view as a readonly editor:
>>> view = View('width','height',Item('area',style='readonly'))
- can also pass a different view to edit_traits()
- views don't need to be specific to a class, as long as Items match

- all Traits automatically support the Listener pattern for function and methods called when a trait changes
- static listener is something like _volume_changed(self,old,new)
- dynamic observer is set with obj.on_trait_change(observerf,name='volume')
	- disconnect it with obj.on_trait_change(observerf,name='volume',remove=True)
- can also use a decorator "@on_trait_change('reverb,volume')" (really one string?) to create listener
- listeners can also be run on a different thread with "dispatch='new'"

- Instance: use compound traits as elementary traits
>>> class Person(HasTraits):
>>>		...
>>> class Family(HasTraits):
>>>		dad = Instance(Person,args=(),kw={'first_name':'Jane'})

- delegation: trait can Delegate a particular trait to one of its compound traits
>>> class Child(Person):
>>>		parent = Instance(Person,args=())
>>>		last_name = Delegate('parent','last_name')

- it's possible to traitify numpy arrays


=== Peter Wang - Interactive plotting with Chaco ===

- several backends: Wx, Quartz, GL (beta)
- nice interactive capabilities, edit elements (colormaps, etc.)

- "script-oriented" plotting is available through enthought.chaco.shell
	- static view of data, concept of main plots

- "application-oriented plot"
	- general scheme: make ArrayPlotData object, make Plot object, run plot in Plot object
	- can run multiple plots (renderers) in the same Plot
	- can arrange Plots in Containers
	- using traits, can dynamically change Plot features (such as Plot.renderer.color) and ArrayPlotData
	- have standard set of tools
		- as simple as plot.tools.append(PanTool(plot)), ZoomTool, DragZoom
	- can create custom tools that see mouse movements and keys


=== Prabhu Ramachandran, Gael Varoquaux - 3D visualization with TVTK and Mayavi ===