def gumbel(shape, dtype=default_override_or(np.float32), loc=0.0, scale=1.0, seed=auto_select, name=''):
"""gumbel(shape, dtype=default_override_or(np.float32), loc=0.0, scale=1.0, seed=auto_select, name='')
Generates samples from the Gumbel distribution with location `loc` and scale `scale`.
Args:
shape (tuple): shape of the output (entries are independent random draws)
dtype (np.float32 or np.float64): data type. Default is np.float32.
loc (float): location of the distribution
scale (float): scale of the distribution
seed (int): pseudo random number generator seed (default: automatically select a unique seed)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`~cntk.ops.functions.Function`
Examples:
>>> g = C.random.gumbel((2,3), seed=98052)
>>> g.eval(device=C.cpu()) # explicitly setting cpu because this is tested on multiple platforms; leave it unspecified in your code
array([[-0.987713, -0.522298, 0.425918],
[-1.019599, 5.435177, 1.586071]], dtype=float32)
See also:
`The Gumbel-Max Trick
<https://hips.seas.harvard.edu/blog/2013/04/06/the-gumbel-max-trick-for-discrete-distributions/>`_.
"""
from cntk.cntk_py import gumbel_random
shape, dtype = sanitize_random_args(shape, dtype)
return gumbel_random(shape, dtype, loc, scale, seed, name)
def uniform(shape, dtype=default_override_or(np.float32), low=0.0, high=1.0, seed=auto_select, name=''):
"""uniform(shape, dtype=default_override_or(np.float32), low=0.0, high=1.0, seed=auto_select, name='')
Generates samples from the uniform distribution in the interval [`low`,`high`).
Args:
shape (tuple): shape of the output (entries are independent random draws)
dtype (np.float32 or np.float64): data type. Default is np.float32.
low (float): lower end of the range of the random numbers
high (float): upper end of the range of the random numbers
seed (int): pseudo random number generator seed (default: automatically select a unique seed)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`~cntk.ops.functions.Function`
Examples:
>>> u = C.random.uniform((2,3), seed=98052)
>>> u.eval(device=C.cpu()) # explicitly setting cpu because this is tested on multiple platforms; leave it unspecified in your code
array([[ 0.931785, 0.814722, 0.479606],
[ 0.937468, 0.004351, 0.185131]], dtype=float32)
"""
from cntk.cntk_py import uniform_random
shape, dtype = sanitize_random_args(shape, dtype)
return uniform_random(shape, dtype, low, high, seed, name)
def gumbel(shape,
dtype=default_override_or(np.float32),
loc=0.0,
scale=1.0,
seed=auto_select,
name=''):
"""gumbel(shape, dtype=default_override_or(np.float32), loc=0.0, scale=1.0, seed=auto_select, name='')
Generates samples from the Gumbel distribution with location `loc` and scale `scale`.
Args:
shape (tuple): shape of the output (entries are independent random draws)
dtype (np.float32 or np.float64): data type. Default is np.float32.
loc (float): location of the distribution
scale (float): scale of the distribution
seed (int): pseudo random number generator seed (default: automatically select a unique seed)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`~cntk.ops.functions.Function`
Examples:
>>> g = C.random.gumbel((2,3), seed=98052)
>>> g.eval(device=C.cpu()) # explicitly setting cpu because this is tested on multiple platforms; leave it unspecified in your code
array([[-0.987713, -0.522298, 0.425918],
[-1.019599, 5.435177, 1.586071]], dtype=float32)
See also:
`The Gumbel-Max Trick
<https://hips.seas.harvard.edu/blog/2013/04/06/the-gumbel-max-trick-for-discrete-distributions/>`_.
"""
from cntk.cntk_py import gumbel_random
shape, dtype = sanitize_random_args(shape, dtype)
return gumbel_random(shape, dtype, loc, scale, seed, name)
def normal(shape,
dtype=default_override_or(np.float32),
mean=0.0,
scale=1.0,
seed=auto_select,
name=''):
"""normal(shape, dtype=default_override_or(np.float32), mean=0.0, scale=1.0, seed=auto_select, name='')
Generates samples from the normal distribution with mean `mean` and standard deviation `scale`.
Args:
shape (tuple): shape of the output (entries are independent random draws)
dtype (np.float32 or np.float64): data type. Default is np.float32.
mean (float): mean of the distribution
scale (float): scale (standard deviation) of the distribution
seed (int): pseudo random number generator seed (default: automatically select a unique seed)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`~cntk.ops.functions.Function`
Examples:
>>> z = C.random.normal((2,3), seed=98052)
>>> z.eval(device=C.cpu()) # explicitly setting cpu because this is tested on multiple platforms; leave it unspecified in your code
array([[ 1.803254, 0.995395, -0.631974],
[-1.73672 , 0.005615, -0.340025]], dtype=float32)
"""
from cntk.cntk_py import normal_random
shape, dtype = sanitize_random_args(shape, dtype)
return normal_random(shape, dtype, mean, scale, seed, name)
def uniform(shape,
dtype=default_override_or(np.float32),
low=0.0,
high=1.0,
seed=auto_select,
name=''):
"""uniform(shape, dtype=default_override_or(np.float32), low=0.0, high=1.0, seed=auto_select, name='')
Generates samples from the uniform distribution in the interval [`low`,`high`).
Args:
shape (tuple): shape of the output (entries are independent random draws)
dtype (np.float32 or np.float64): data type. Default is np.float32.
low (float): lower end of the range of the random numbers
high (float): upper end of the range of the random numbers
seed (int): pseudo random number generator seed (default: automatically select a unique seed)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`~cntk.ops.functions.Function`
Examples:
>>> u = C.random.uniform((2,3), seed=98052)
>>> u.eval(device=C.cpu()) # explicitly setting cpu because this is tested on multiple platforms; leave it unspecified in your code
array([[ 0.931785, 0.814722, 0.479606],
[ 0.937468, 0.004351, 0.185131]], dtype=float32)
"""
from cntk.cntk_py import uniform_random
shape, dtype = sanitize_random_args(shape, dtype)
return uniform_random(shape, dtype, low, high, seed, name)
def bernoulli(shape,
dtype=default_override_or(np.float32),
mean=0.5,
seed=auto_select,
name=''):
"""bernoulli(shape, dtype=default_override_or(np.float32), mean=0.5, seed=auto_select, name='')
Generates samples from the Bernoulli distribution with success probability `mean`.
Args:
shape (tuple): shape of the output (entries are independent random draws)
dtype (np.float32 or np.float64): data type. Default is np.float32.
mean (float): success probability
seed (int): pseudo random number generator seed (default: automatically select a unique seed)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`~cntk.ops.functions.Function`
Examples:
>>> b = C.random.bernoulli((2,3), seed=98052)
>>> b.eval(device=C.cpu()) # explicitly setting cpu because this is tested on multiple platforms; leave it unspecified in your code
array([[ 1., 1., 0.],
[ 1., 0., 0.]], dtype=float32)
"""
from cntk.cntk_py import bernoulli_random
shape, dtype = sanitize_random_args(shape, dtype)
return bernoulli_random(shape, dtype, mean, seed, name)
def bernoulli(shape, dtype=default_override_or(np.float32), mean=0.5, seed=auto_select, name=''):
"""bernoulli(shape, dtype=default_override_or(np.float32), mean=0.5, seed=auto_select, name='')
Generates samples from the Bernoulli distribution with success probability `mean`.
Args:
shape (tuple): shape of the output (entries are independent random draws)
dtype (np.float32 or np.float64): data type. Default is np.float32.
mean (float): success probability
seed (int): pseudo random number generator seed (default: automatically select a unique seed)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`~cntk.ops.functions.Function`
Examples:
>>> b = C.random.bernoulli((2,3), seed=98052)
>>> b.eval(device=C.cpu()) # explicitly setting cpu because this is tested on multiple platforms; leave it unspecified in your code
array([[ 1., 1., 0.],
[ 1., 0., 0.]], dtype=float32)
"""
from cntk.cntk_py import bernoulli_random
shape, dtype = sanitize_random_args(shape, dtype)
return bernoulli_random(shape, dtype, mean, seed, name)
def normal(shape, dtype=default_override_or(np.float32), mean=0.0, scale=1.0, seed=auto_select, name=''):
"""normal(shape, dtype=default_override_or(np.float32), mean=0.0, scale=1.0, seed=auto_select, name='')
Generates samples from the normal distribution with mean `mean` and standard deviation `scale`.
Args:
shape (tuple): shape of the output (entries are independent random draws)
dtype (np.float32 or np.float64): data type. Default is np.float32.
mean (float): mean of the distribution
scale (float): scale (standard deviation) of the distribution
seed (int): pseudo random number generator seed (default: automatically select a unique seed)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`~cntk.ops.functions.Function`
Examples:
>>> z = C.random.normal((2,3), seed=98052)
>>> z.eval(device=C.cpu()) # explicitly setting cpu because this is tested on multiple platforms; leave it unspecified in your code
array([[ 1.803254, 0.995395, -0.631974],
[-1.73672 , 0.005615, -0.340025]], dtype=float32)
"""
from cntk.cntk_py import normal_random
shape, dtype = sanitize_random_args(shape, dtype)
return normal_random(shape, dtype, mean, scale, seed, name)