Module asymmetree.tools.Sampling
Expand source code
# -*- coding: utf-8 -*-
import numpy as np
from scipy.special import zeta
__author__ = 'David Schaller'
class Sampler:
"""Campling from from a variety of distributions."""
def __init__(self, params, minimum=None, maximum=None, discrete=False,
shift=0):
"""
Parameters
----------
params : float or tuple
The distribution and its parameters, see documentation for
available options.
minimum : float, optional
Minimum value to be sampled. The default is None.
maximum : float, optional
Maximum value to be sampled. The default is None.
discrete : bool, optional
If True, round sampled values in case of a continuous
distribution. The default is False.
shift : float, optional
Shift distribution along x-axis. The default is 0, i.e., the
distribution is not shifted.
"""
self._params = params
self._discrete = discrete
self._min = minimum
self._max = maximum
self._shift = shift
# function to be defined in _check_distribution
self.draw = None
self._check_distribution()
def __call__(self):
"""Draw a value from the specified distribution."""
return self.draw()
def _check_distribution(self):
if self._min is not None and not isinstance(self._min, (int, float)):
raise ValueError('minimum must be a number')
if self._max is not None and not isinstance(self._max, (int, float)):
raise ValueError('maximum must be a number')
if not isinstance(self._shift, (int, float)):
raise ValueError('shift value must be a number')
if isinstance(self._params, (int, float)):
self._distr = 'constant'
self.draw = self._draw_constant
self._exp_val = (round(self._params) if self._discrete else
self._params)
elif isinstance(self._params, (tuple, list)) and len(self._params) >= 2:
if self._params[0] == 'constant':
val = self._params[1]
if not isinstance(val, (int, float)):
raise ValueError('constant value must be a number')
self._distr = 'constant'
self.draw = self._draw_constant
self._exp_val = round(val) if self._discrete else val
elif self._params[0] == 'uniform':
a = self._params[1]
if not len(self._params) >= 3:
raise ValueError('uniform distr. requires 2 parameters')
b = self._params[2]
if (not isinstance(a, (int, float)) or
not isinstance(b, (int, float)) or b < a):
raise ValueError('parameters a and b for uniform distr.'\
'must be numbers and a <= b')
self._distr = 'uniform'
self.draw = self._draw_uniform
self._exp_val = (a + b) / 2
self._a = a
self._b = b
elif self._params[0] == 'discrete_uniform':
a = self._params[1]
if not len(self._params) >= 3:
raise ValueError('discrete uniform distr. requires 2 '\
'parameters')
b = self._params[2]
if (not isinstance(a, int) or
not isinstance(b, int) or b < a):
raise ValueError('parameters a and b for discrete uniform '\
'distr. must be ints and a <= b')
self._distr = 'discrete_uniform'
self.draw = self._draw_discrete_uniform
self._exp_val = (a + b) / 2
self._a = a
self._b = b
elif self._params[0] == 'gamma':
shape = self._params[1]
if not len(self._params) >= 3:
raise ValueError('gamma distr. requires 2 parameters')
scale = self._params[2]
if (not isinstance(shape, float) or shape <= 0.0 or
not isinstance(scale, float) or scale <= 0.0):
raise ValueError('scale and shape parameters for gamma '\
'distribution must be floats >0.0')
self._distr = 'gamma'
self.draw = self._draw_gamma
self._exp_val = shape * scale
self._shape = shape
self._scale = scale
elif self._params[0] == 'gamma_mean':
mean = self._params[1]
if not isinstance(mean , (int, float)) or mean <= 0.0:
raise ValueError('mean of gamma distribution must be '\
'a number >0.0')
self._distr = 'gamma'
self.draw = self._draw_gamma
self._exp_val = mean
self._shape = 1.0
self._scale = mean / self._shape
elif self._params[0] == 'exponential':
rate = self._params[1]
if not isinstance(mean , float) or rate < 0.0:
raise ValueError('rate must be a float >=0.0')
self._distr = 'exponential'
self.draw = self._draw_exponential
self._exp_val = float('inf') if rate == 0.0 else 1/rate
self._rate = rate
elif self._params[0] == 'zipf':
a = self._params[1]
if not isinstance(a, float) or a <= 1.0:
raise ValueError('parameter a for zipf distr. must be a '\
'float >1')
# expected value is infinite for a <= 2
if a > 2.0:
self._exp_val = zeta(a - 1.0) / zeta(a)
# use some valid value when exp. value is infinite
elif self._min is not None:
self._exp_val = self._min
else:
self._exp_val = 0.0
self._distr = 'zipf'
self.draw = self._draw_zipf
self._a = a
elif self._params[0] == 'negative_binomial':
r = self._params[1]
if not len(self._params) >= 3:
raise ValueError('negative binomial distr. requires 2 '\
'parameters')
q = self._params[2]
if not isinstance(r, int) or r < 1:
raise ValueError('parameter r must be an int and >0')
if not isinstance(q, float) or q <= 0 or q >= 1.0:
raise ValueError('parameter q must be a float >0 and <1')
self._distr = 'negative_binomial'
self.draw = self._draw_neg_bin
self._exp_val = r * q / (1 - q)
self._r = r
self._q = q
else:
raise ValueError("length distribution '{}' is not "\
"supported".format(self._params[0]))
else:
raise ValueError('could not parse distribution')
self._exp_val += self._shift
# expected value must be between min. and max.
if ((self._min is not None and self._exp_val < self._min) or
(self._max is not None and self._exp_val > self._max)):
raise ValueError('expected value must be >= minimum and'\
'<= maximum')
def _draw_constant(self):
return self._exp_val
def _draw_uniform(self):
while True:
x = np.random.uniform(low=self._a, high=self._b) + self._shift
if self._discrete:
x = round(x)
if ((not self._min or x >= self.min) and
(not self._max or x <= self._max)):
return x
def _draw_discrete_uniform(self):
while True:
x = np.random.randint(self._a, high=self._b) + self._shift
if ((not self._min or x >= self.min) and
(not self._max or x <= self._max)):
return x
def _draw_gamma(self):
while True:
x = np.random.gamma(self._shape, scale=self._scale) + self._shift
if self._discrete:
x = round(x)
if ((not self._min or x >= self.min) and
(not self._max or x <= self._max)):
return x
def _draw_exponential(self):
while True:
if self._rate == 0.0:
x = float('inf')
else:
x = np.random.exponential(scale=1/self._rate) + self._shift
if self._discrete:
x = round(x)
if ((not self._min or x >= self.min) and
(not self._max or x <= self._max)):
return x
def _draw_zipf(self):
while True:
x = np.random.zipf(self._a) + self._shift
if ((not self._min or x >= self.min) and
(not self._max or x <= self._max)):
return x
def _draw_neg_bin(self):
while True:
x = np.random.negative_binomial(self._r, 1 - self._q) + self._shift
if ((not self._min or x >= self.min) and
(not self._max or x <= self._max)):
return xClasses
class Sampler (params, minimum=None, maximum=None, discrete=False, shift=0)-
Campling from from a variety of distributions.
Parameters
params:floatortuple- The distribution and its parameters, see documentation for available options.
minimum:float, optional- Minimum value to be sampled. The default is None.
maximum:float, optional- Maximum value to be sampled. The default is None.
discrete:bool, optional- If True, round sampled values in case of a continuous distribution. The default is False.
shift:float, optional- Shift distribution along x-axis. The default is 0, i.e., the distribution is not shifted.
Expand source code
class Sampler: """Campling from from a variety of distributions.""" def __init__(self, params, minimum=None, maximum=None, discrete=False, shift=0): """ Parameters ---------- params : float or tuple The distribution and its parameters, see documentation for available options. minimum : float, optional Minimum value to be sampled. The default is None. maximum : float, optional Maximum value to be sampled. The default is None. discrete : bool, optional If True, round sampled values in case of a continuous distribution. The default is False. shift : float, optional Shift distribution along x-axis. The default is 0, i.e., the distribution is not shifted. """ self._params = params self._discrete = discrete self._min = minimum self._max = maximum self._shift = shift # function to be defined in _check_distribution self.draw = None self._check_distribution() def __call__(self): """Draw a value from the specified distribution.""" return self.draw() def _check_distribution(self): if self._min is not None and not isinstance(self._min, (int, float)): raise ValueError('minimum must be a number') if self._max is not None and not isinstance(self._max, (int, float)): raise ValueError('maximum must be a number') if not isinstance(self._shift, (int, float)): raise ValueError('shift value must be a number') if isinstance(self._params, (int, float)): self._distr = 'constant' self.draw = self._draw_constant self._exp_val = (round(self._params) if self._discrete else self._params) elif isinstance(self._params, (tuple, list)) and len(self._params) >= 2: if self._params[0] == 'constant': val = self._params[1] if not isinstance(val, (int, float)): raise ValueError('constant value must be a number') self._distr = 'constant' self.draw = self._draw_constant self._exp_val = round(val) if self._discrete else val elif self._params[0] == 'uniform': a = self._params[1] if not len(self._params) >= 3: raise ValueError('uniform distr. requires 2 parameters') b = self._params[2] if (not isinstance(a, (int, float)) or not isinstance(b, (int, float)) or b < a): raise ValueError('parameters a and b for uniform distr.'\ 'must be numbers and a <= b') self._distr = 'uniform' self.draw = self._draw_uniform self._exp_val = (a + b) / 2 self._a = a self._b = b elif self._params[0] == 'discrete_uniform': a = self._params[1] if not len(self._params) >= 3: raise ValueError('discrete uniform distr. requires 2 '\ 'parameters') b = self._params[2] if (not isinstance(a, int) or not isinstance(b, int) or b < a): raise ValueError('parameters a and b for discrete uniform '\ 'distr. must be ints and a <= b') self._distr = 'discrete_uniform' self.draw = self._draw_discrete_uniform self._exp_val = (a + b) / 2 self._a = a self._b = b elif self._params[0] == 'gamma': shape = self._params[1] if not len(self._params) >= 3: raise ValueError('gamma distr. requires 2 parameters') scale = self._params[2] if (not isinstance(shape, float) or shape <= 0.0 or not isinstance(scale, float) or scale <= 0.0): raise ValueError('scale and shape parameters for gamma '\ 'distribution must be floats >0.0') self._distr = 'gamma' self.draw = self._draw_gamma self._exp_val = shape * scale self._shape = shape self._scale = scale elif self._params[0] == 'gamma_mean': mean = self._params[1] if not isinstance(mean , (int, float)) or mean <= 0.0: raise ValueError('mean of gamma distribution must be '\ 'a number >0.0') self._distr = 'gamma' self.draw = self._draw_gamma self._exp_val = mean self._shape = 1.0 self._scale = mean / self._shape elif self._params[0] == 'exponential': rate = self._params[1] if not isinstance(mean , float) or rate < 0.0: raise ValueError('rate must be a float >=0.0') self._distr = 'exponential' self.draw = self._draw_exponential self._exp_val = float('inf') if rate == 0.0 else 1/rate self._rate = rate elif self._params[0] == 'zipf': a = self._params[1] if not isinstance(a, float) or a <= 1.0: raise ValueError('parameter a for zipf distr. must be a '\ 'float >1') # expected value is infinite for a <= 2 if a > 2.0: self._exp_val = zeta(a - 1.0) / zeta(a) # use some valid value when exp. value is infinite elif self._min is not None: self._exp_val = self._min else: self._exp_val = 0.0 self._distr = 'zipf' self.draw = self._draw_zipf self._a = a elif self._params[0] == 'negative_binomial': r = self._params[1] if not len(self._params) >= 3: raise ValueError('negative binomial distr. requires 2 '\ 'parameters') q = self._params[2] if not isinstance(r, int) or r < 1: raise ValueError('parameter r must be an int and >0') if not isinstance(q, float) or q <= 0 or q >= 1.0: raise ValueError('parameter q must be a float >0 and <1') self._distr = 'negative_binomial' self.draw = self._draw_neg_bin self._exp_val = r * q / (1 - q) self._r = r self._q = q else: raise ValueError("length distribution '{}' is not "\ "supported".format(self._params[0])) else: raise ValueError('could not parse distribution') self._exp_val += self._shift # expected value must be between min. and max. if ((self._min is not None and self._exp_val < self._min) or (self._max is not None and self._exp_val > self._max)): raise ValueError('expected value must be >= minimum and'\ '<= maximum') def _draw_constant(self): return self._exp_val def _draw_uniform(self): while True: x = np.random.uniform(low=self._a, high=self._b) + self._shift if self._discrete: x = round(x) if ((not self._min or x >= self.min) and (not self._max or x <= self._max)): return x def _draw_discrete_uniform(self): while True: x = np.random.randint(self._a, high=self._b) + self._shift if ((not self._min or x >= self.min) and (not self._max or x <= self._max)): return x def _draw_gamma(self): while True: x = np.random.gamma(self._shape, scale=self._scale) + self._shift if self._discrete: x = round(x) if ((not self._min or x >= self.min) and (not self._max or x <= self._max)): return x def _draw_exponential(self): while True: if self._rate == 0.0: x = float('inf') else: x = np.random.exponential(scale=1/self._rate) + self._shift if self._discrete: x = round(x) if ((not self._min or x >= self.min) and (not self._max or x <= self._max)): return x def _draw_zipf(self): while True: x = np.random.zipf(self._a) + self._shift if ((not self._min or x >= self.min) and (not self._max or x <= self._max)): return x def _draw_neg_bin(self): while True: x = np.random.negative_binomial(self._r, 1 - self._q) + self._shift if ((not self._min or x >= self.min) and (not self._max or x <= self._max)): return x