User Guide

Array Conversion

The function bob.core.convert() allows you to convert objects of type numpy.ndarray between different types, with range compression or decompression. For example, here we demonstrate a conversion using default ranges. In this type of conversion, our implementation will assume that the source array contains values within the range of uint8_t numbers and will expand it to the range of uint16_t numbers, as desired by the programmer:

>>> x = numpy.array([0,255,0,255,0,255], 'uint8').reshape(2,3)
>>> x
array([[  0, 255,   0],
       [255,   0, 255]], dtype=uint8)
>>> bob.core.convert(x, 'uint16')
array([[    0, 65535,     0],
       [65535,     0, 65535]], dtype=uint16)

The user can optionally specify source, destination ranges or both. For example:

>>> x = numpy.array([0, 10, 20, 30, 40], 'uint8')
>>> bob.core.convert(x, 'float64', source_range=(0,40), dest_range=(0.,1.))
array([ 0.  ,  0.25,  0.5 ,  0.75,  1.  ])

Any range not specified is assumed to default on the type range.

Random Number Generation

You can build a new random number generator (RNG) of type bob.core.random.mt19937 using one of two possible ways:

1. Use the default constructor, which initializes with the default seed:

   >>> bob.core.random.mt19937()
   bob.core.random.mt19937()
   

2. Pass a seed while initializing:

   >>> rng = bob.core.random.mt19937(34)
   

RNGs can be compared for equality. The == operator checks if both generators are on the exact same state and would generate the same sequence of numbers when exposed to the same distributions. For example:

>>> rng1 = bob.core.random.mt19937(111)
>>> rng2 = bob.core.random.mt19937(111)
>>> rng1 == rng2
True
>>> rng3 = bob.core.random.mt19937(12)
>>> rng1 == rng3
False

The seed can be re-initialized at any point in time, which can be used to sync two RNGs:

>>> rng3.seed(111)
>>> rng1 == rng3
True

Distributions skew numbers produced by the RNG so they look like the parameterized distribution. By calling a distribution with an RNG, one effectively generates random numbers:

>>> rng = bob.core.random.mt19937()
>>> # creates an uniform distribution of integers inside [0, 10]
>>> u = bob.core.random.uniform(int, 0, 10)
>>> u(rng) 
8

At our reference guide (see below), you will find more implemented distributions you can use on your programs. To simplify the task of generating random numbers, we provide a class that mimics the behavior of boost::random::variate_generator, in Python:

>>> ugen = bob.core.random.variate_generator(rng, u)
>>> ugen() 
6

You can also pass an optional shape when you call the variate generator, in which case it generates a numpy.ndarray of the specified size:

>>> ugen((3,3)) 
array([[ 3,  1,  6],
       [ 3,  2,  6],
       [10, 10, 10]])