| 1 | """ |
|---|
| 2 | NeuroTools.random |
|---|
| 3 | ===================== |
|---|
| 4 | |
|---|
| 5 | A set of classes representing statistical distributions, with an interface that |
|---|
| 6 | is compatible with the ParameterSpace class in the parameters module. |
|---|
| 7 | |
|---|
| 8 | Classes |
|---|
| 9 | ------- |
|---|
| 10 | |
|---|
| 11 | GammaDist - gamma.pdf(x,a,b) = x**(a-1)*exp(-x/b)/gamma(a)/b**a |
|---|
| 12 | NormalDist - normal distribution |
|---|
| 13 | UniformDist - uniform distribution |
|---|
| 14 | |
|---|
| 15 | """ |
|---|
| 16 | |
|---|
| 17 | from NeuroTools import check_dependency |
|---|
| 18 | |
|---|
| 19 | import numpy, numpy.random |
|---|
| 20 | |
|---|
| 21 | |
|---|
| 22 | class ParameterDist(object): |
|---|
| 23 | |
|---|
| 24 | def __init__(self,**params): |
|---|
| 25 | self.params = params |
|---|
| 26 | self.dist_name = 'ParameterDist' |
|---|
| 27 | |
|---|
| 28 | def __repr__(self): |
|---|
| 29 | if len(self.params)==0: |
|---|
| 30 | return '%s()'% (self.dist_name,) |
|---|
| 31 | s = '%s('% (self.dist_name,) |
|---|
| 32 | for key in self.params: |
|---|
| 33 | s+='%s=%s,' % (key,str(self.params[key])) |
|---|
| 34 | return s[:-1]+')' |
|---|
| 35 | |
|---|
| 36 | def next(self,n=1): |
|---|
| 37 | raise NotImplementedError('This is an abstract base class and cannot be used directly') |
|---|
| 38 | |
|---|
| 39 | def from_stats(self,vals,bias=0.0,expand=1.0): |
|---|
| 40 | self.__init__(mean=numpy.mean(vals)+bias, std=numpy.std(vals)*expand) |
|---|
| 41 | |
|---|
| 42 | def __eq__(self, o): |
|---|
| 43 | # should we track the state of the rng and return False if it is different between self and o? |
|---|
| 44 | if (type(self) == type(o) and |
|---|
| 45 | self.dist_name == o.dist_name and |
|---|
| 46 | self.params == o.params): |
|---|
| 47 | return True |
|---|
| 48 | else: |
|---|
| 49 | return False |
|---|
| 50 | |
|---|
| 51 | class GammaDist(ParameterDist): |
|---|
| 52 | """ |
|---|
| 53 | gamma.pdf(x,a,b) = x**(a-1)*exp(-x/b)/gamma(a)/b**a |
|---|
| 54 | |
|---|
| 55 | Yields strictly positive numbers. |
|---|
| 56 | Generally the distribution is implemented by scipy.stats.gamma.pdf(x/b,a)/b |
|---|
| 57 | For more info, in ipython type: |
|---|
| 58 | >>> ? scipy.stats.gamma |
|---|
| 59 | |
|---|
| 60 | """ |
|---|
| 61 | |
|---|
| 62 | def __init__(self,mean=None,std=None,repr_mode='ms',**params): |
|---|
| 63 | """ |
|---|
| 64 | repr_mode specifies how the dist is displayed, |
|---|
| 65 | either mean,var ('ms', the default) or a,b ('ab') |
|---|
| 66 | """ |
|---|
| 67 | |
|---|
| 68 | if check_dependency('scipy'): |
|---|
| 69 | self.next = self._next_scipy |
|---|
| 70 | |
|---|
| 71 | self.repr_mode = repr_mode |
|---|
| 72 | if 'm' in params and mean==None: |
|---|
| 73 | mean = params['m'] |
|---|
| 74 | if 's' in params and std==None: |
|---|
| 75 | std = params['s'] |
|---|
| 76 | |
|---|
| 77 | # both mean and std not specified |
|---|
| 78 | if (mean,std)==(None,None): |
|---|
| 79 | if 'a' in params: |
|---|
| 80 | a = params['a'] |
|---|
| 81 | else: |
|---|
| 82 | a = 1.0 |
|---|
| 83 | if 'b' in params: |
|---|
| 84 | b = params['b'] |
|---|
| 85 | else: |
|---|
| 86 | b = 1.0 |
|---|
| 87 | else: |
|---|
| 88 | if mean==None: |
|---|
| 89 | mean = 0.0 |
|---|
| 90 | if std==None: |
|---|
| 91 | std=1.0 |
|---|
| 92 | a = mean**2/std**2 |
|---|
| 93 | b = mean/a |
|---|
| 94 | ParameterDist.__init__(self,a=a,b=b) |
|---|
| 95 | self.dist_name = 'GammaDist' |
|---|
| 96 | |
|---|
| 97 | def _next_scipy(self,n=1): |
|---|
| 98 | import scipy.stats |
|---|
| 99 | return scipy.stats.gamma.rvs(self.params['a'],size=n)*self.params['b'] |
|---|
| 100 | def _next_no_scipy(self,n=1): |
|---|
| 101 | raise Exception('Error scipy was not found at import time. GammaDist realization disabled.') |
|---|
| 102 | |
|---|
| 103 | next = _next_no_scipy |
|---|
| 104 | |
|---|
| 105 | def mean(self): |
|---|
| 106 | return self.params['a']*self.params['b'] |
|---|
| 107 | |
|---|
| 108 | def std(self): |
|---|
| 109 | return self.params['b']*numpy.sqrt(self.params['a']) |
|---|
| 110 | |
|---|
| 111 | def __repr__(self): |
|---|
| 112 | if self.repr_mode == 'ms': |
|---|
| 113 | return '%s(m=%f,s=%f)' % (self.dist_name,self.mean(),self.std()) |
|---|
| 114 | else: |
|---|
| 115 | return '%s(a=%f,b=%f)' % (self.dist_name,self.params['a'],self.params['b']) |
|---|
| 116 | |
|---|
| 117 | |
|---|
| 118 | class NormalDist(ParameterDist): |
|---|
| 119 | """ |
|---|
| 120 | normal distribution with parameters |
|---|
| 121 | mean + std |
|---|
| 122 | |
|---|
| 123 | """ |
|---|
| 124 | |
|---|
| 125 | def __init__(self,mean=0.0,std=1.0): |
|---|
| 126 | ParameterDist.__init__(self,mean=mean,std=std) |
|---|
| 127 | self.dist_name = 'NormalDist' |
|---|
| 128 | |
|---|
| 129 | def next(self,n=1): |
|---|
| 130 | return numpy.random.normal(loc=self.params['mean'],scale=self.params['std'],size=n) |
|---|
| 131 | |
|---|
| 132 | |
|---|
| 133 | class UniformDist(ParameterDist): |
|---|
| 134 | """ |
|---|
| 135 | uniform distribution with min,max |
|---|
| 136 | """ |
|---|
| 137 | |
|---|
| 138 | def __init__(self,min=0.0,max=1.0, return_type=float): |
|---|
| 139 | ParameterDist.__init__(self,min=min,max=max) |
|---|
| 140 | self.dist_name = 'UniformDist' |
|---|
| 141 | self.return_type = return_type |
|---|
| 142 | |
|---|
| 143 | def next(self,n=1): |
|---|
| 144 | vals = numpy.random.uniform(low=self.params['min'],high=self.params['max'],size=n) |
|---|
| 145 | if self.return_type != float: |
|---|
| 146 | vals = vals.astype(self.return_type) |
|---|
| 147 | return vals |
|---|
| 148 | |
|---|
| 149 | def from_stats(self,vals,bias=0.0,expand=1.0): |
|---|
| 150 | mn = numpy.min(vals) |
|---|
| 151 | mx = numpy.max(vals) |
|---|
| 152 | center = 0.5*(mx+mn)+bias |
|---|
| 153 | hw = 0.5*(mx-mn)*expand |
|---|
| 154 | self.__init__(min=center-hw,max=center+hw) |
|---|
