rvg
is a Python 3 package utility to create random values of any Python 3 data type.
Its main purpose is to help in applications where reliable -in terms of type safety- random values are needed (e.g. statistics, machine learning, general testing etc), and in specific layouts (e.g. "I want a numpy
structured array of random pairs of ints and floats and I want it now").
Sotiris Niarchos and George Papadopoulos
You can either:
master
branch will always mirror the latest release)pip
:
pip install rvg
For the time being, only numpy
types are supported. More specifically:
numpy
scalar data typesnumpy
arrays of scalar data typesnumpy
arrays of structured data typesAfter the alpha release, more features will be implemented, focusing mainly on Python 3 native types.
Right now, rvg
provides 2 interafaces for random values generation through the NumPyRVG
class:
from rvg import NumPyRVG
import numpy as np
# Interface 1
randuint = NumPyRVG(dtype=np.uint16)
# Interface 2
randsmall = NumPyRVG(limit=10) # same as limits=(-10, 10)
randbig = NumPyRVG(limits=(1e10, 1e100))
The functionalities of NumPyRVG
include the generation of:
numpy
scalar data types:>>> randsmall(np.uint8)
8
>>> randbig(np.double)
1.9296971162995923e+99
>>> res = [randsmall(t) for t in [np.int8, np.uint16, np.float32, np.double]]
>>> res
[7, 1, 3.0503626, 3.759943941132951]
>>> list(map(type, res))
[<class 'numpy.int8'>, <class 'numpy.uint16'>, <class 'numpy.float32'>, <class 'numpy.float64'>]
numpy
array data types from scalar types:>>> randuint((50, 100), shape=3)
array([79, 81, 85], dtype=uint16)
>>> randsmall(np.float16, shape=(4, 2))
array([[-7.23 , -9.31 ],
[-4.97 , -6.06 ],
[-5.19 , -3.344],
[-6.586, -3.133]], dtype=float16)
numpy
structured array data types (structured datatypes can be nested). Limits and shapes can be given in the form of a dictionary, describing all limits and/or shapes of each field of each level of the structured data type. An example follows:# Consider the struct definition below, in C:
##############################################
# typedef struct knode {
# int location;
# int indices [3];
# int keys [3];
# bool is_leaf;
# int num_keys;
# } knode;
##############################################
import numpy as np
from rvg import NumPyRVG
knode = np.dtype([
('location', int),
('indices', (int, 3)),
('keys', (int, 3)),
('is_leaf', int),
('num_keys', int)
])
knode_params = {
'location' : (0, 10),
'indices' : 42,
'keys' : 117,
'is_leaf' : (0, 2),
'num_keys' : (0, 256)
}
random_knode = NumPyRVG(dtype=knode)
knodes_array = random_knode(knode_params, 5)
print(knodes_array)
The output of the above script is a nested structured numpy
array consisting of 5 knode
structs, randomly initialized!
[(0, [-18, -11, 34], [ 89, 35, -57], 1, 189)
(6, [-35, 0, 4], [ -56, -65, 26], 1, 217)
(4, [-29, 40, 37], [ 93, -116, 91], 0, 38)
(2, [-28, -42, -36], [-101, 0, -43], 0, 82)
(0, [-14, -19, -13], [ -98, -46, -78], 1, 238)]
If the script was run in an interpreter, you could inspect its type as well:
>>> knodes_array
array([(0, [-18, -11, 34], [ 89, 35, -57], 1, 189),
(6, [-35, 0, 4], [ -56, -65, 26], 1, 217),
(4, [-29, 40, 37], [ 93, -116, 91], 0, 38),
(2, [-28, -42, -36], [-101, 0, -43], 0, 82),
(0, [-14, -19, -13], [ -98, -46, -78], 1, 238)],
dtype=[('location', '<i8'), ('indices', '<i8', (3,)), ('keys', '<i8', (3,)), ('is_leaf', '<i8'), ('num_keys', '<i8')])
The feature of nesting is not limited in arrays; you can create data types that are as complex as you want and/or need! See the relative test as an example of struct nesting.