| 1 |
""" |
| 2 |
Statistical Ensembles |
| 3 |
""" |
| 4 |
|
| 5 |
from csb.numeric import log, exp |
| 6 |
from abc import ABCMeta, abstractmethod |
| 7 |
|
| 8 |
|
| 9 |
class StatisticalEnsemble(object): |
| 10 |
|
| 11 |
__metaclass__ = ABCMeta |
| 12 |
|
| 13 |
def __call__(self, raw_energies): |
| 14 |
return exp(-self.energy(raw_energies)) |
| 15 |
|
| 16 |
def log_prob(self, raw_energies): |
| 17 |
return -self.energy(raw_energies) |
| 18 |
|
| 19 |
@abstractmethod |
| 20 |
def energy(self, raw_energies): |
| 21 |
""" |
| 22 |
Transforms the raw energies as if they were observed |
| 23 |
in this statistical ensemble |
| 24 |
""" |
| 25 |
pass |
| 26 |
|
| 27 |
def gradient(self, raw_energies): |
| 28 |
raise NotImplementedError() |
| 29 |
|
| 30 |
|
| 31 |
class BoltzmannEnsemble(StatisticalEnsemble): |
| 32 |
|
| 33 |
def __init__(self, beta=1.): |
| 34 |
|
| 35 |
self._beta = float(beta) |
| 36 |
|
| 37 |
@property |
| 38 |
def beta(self): |
| 39 |
""" |
| 40 |
Inverse temperature |
| 41 |
""" |
| 42 |
return self._beta |
| 43 |
@beta.setter |
| 44 |
def beta(self, value): |
| 45 |
value = float(value) |
| 46 |
if value <= 0.: |
| 47 |
raise ValueError("Inverse temperature {0} < 0".formate(value)) |
| 48 |
self._beta = value |
| 49 |
|
| 50 |
def energy(self, raw_energies): |
| 51 |
return raw_energies * self._beta |
| 52 |
|
| 53 |
class FermiEnsemble(BoltzmannEnsemble): |
| 54 |
|
| 55 |
def __init__(self, beta=1., e_max=0.): |
| 56 |
|
| 57 |
super(FermiEnsemble, self).__init__(beta) |
| 58 |
self._e_max = float(e_max) |
| 59 |
|
| 60 |
@property |
| 61 |
def e_max(self): |
| 62 |
""" |
| 63 |
Maximum energy |
| 64 |
""" |
| 65 |
return self._e_max |
| 66 |
@e_max.setter |
| 67 |
def e_max(self, value): |
| 68 |
self._e_max = float(value) |
| 69 |
|
| 70 |
def energy(self, raw_energies): |
| 71 |
|
| 72 |
from numpy import isinf |
| 73 |
|
| 74 |
if isinf(self.beta): |
| 75 |
m = (raw_energies >= self.e_max).astype('f') |
| 76 |
return - m * log(0.) |
| 77 |
else: |
| 78 |
x = 1 + exp(self.beta * (raw_energies - self.e_max)) |
| 79 |
return log(x) |
| 80 |
|
| 81 |
class TsallisEnsemble(StatisticalEnsemble): |
| 82 |
|
| 83 |
def __init__(self, q=1., e_min=0.): |
| 84 |
|
| 85 |
self._q = q |
| 86 |
self._e_min = e_min |
| 87 |
|
| 88 |
@property |
| 89 |
def q(self): |
| 90 |
""" |
| 91 |
q-analoge of the temperature |
| 92 |
""" |
| 93 |
return self._q |
| 94 |
@q.setter |
| 95 |
def q(self, value): |
| 96 |
if value <= 0.: |
| 97 |
raise ValueError("Inverse temperature {0} < 0".formate(value)) |
| 98 |
self._q = value |
| 99 |
|
| 100 |
@property |
| 101 |
def e_min(self): |
| 102 |
""" |
| 103 |
lower bound of the energy |
| 104 |
""" |
| 105 |
return self._e_min |
| 106 |
@e_min.setter |
| 107 |
def e_min(self, value): |
| 108 |
self._e_min = value |
| 109 |
|
| 110 |
def energy(self, raw_energies): |
| 111 |
q = self.q |
| 112 |
e_min = self.e_min |
| 113 |
|
| 114 |
if (q < 1 + 1e-10): |
| 115 |
return raw_energies * q |
| 116 |
else: |
| 117 |
return log(1 + (raw_energies - e_min) * (q - 1)) * q / (q - 1) + e_min |
| 118 |
|
| 119 |
|
| 120 |
class CompositeEnsemble(StatisticalEnsemble): |
| 121 |
|
| 122 |
def __init__(self, ensembles=[]): |
| 123 |
|
| 124 |
self._ensembles = ensembles |
| 125 |
|
| 126 |
@property |
| 127 |
def ensembles(self): |
| 128 |
""" |
| 129 |
Collection of statistical ensembles |
| 130 |
""" |
| 131 |
return self._ensembles |
| 132 |
@ensembles.setter |
| 133 |
def ensembles(self, value): |
| 134 |
if not isinstance(value, list): |
| 135 |
if len(value) > 0: |
| 136 |
if not isinstance(value[0], StatisticalEnsemble): |
| 137 |
raise ValueError("Not a list of statistical ensembles") |
| 138 |
else: |
| 139 |
self._enesmbles = value |
| 140 |
else: |
| 141 |
self._enesmbles = value |
| 142 |
|
| 143 |
def energy(self, raw_energies): |
| 144 |
return sum([self._ensembles[i].energy(raw_energies[i]) |
| 145 |
for i in range(len(self.ensembles))], 0) |
| 146 |
|
| 147 |
def gradient(self, raw_energies): |
| 148 |
return sum([self._ensembles[i].gradient(raw_energies[i]) |
| 149 |
for i in range(len(self.ensembles))], 0) |