class ActivityNode(ArrayNode):
_set_activity=None
decoder_size_warning=1000
sample_step_size=500
decoder_mode='NxN'
decoder_noise_use_gamma=False
decoder_noise_use_limit=False
decoder_noise_use_activity=False
lesion_size=0
lesion_cells=None
def configure(self,neurons,lif=True,saturation_range=(200,300),
t_ref=0.001,t_rc=0.01,J_threshold=1,activation_noise=0.1,apply_noise=False,
threshold_coverage=0.9,threshold_min=None,threshold_max=None,
basis_style='Sphere',sample_style='DefaultSampling',
sample_count=None,seed=None,code=None,force_new=False,use_hd=False,
basis=None,thresholds=None,saturations=None,alphas=None,Jbiases=None):
self.neurons=neurons
self.lif=lif
self.saturation_range=saturation_range
self.t_ref=t_ref
self.t_rc=t_rc
self.J_threshold=J_threshold
self.basis_style=basis_style
self.sample_style=sample_style
self.threshold_coverage=threshold_coverage
self.threshold_min=threshold_min
self.threshold_max=threshold_max
if sample_count is None:
sample_count=self.dimensions*500
if sample_count>5000: sample_count=5000
self.sample_count=sample_count
self.activation_noise=activation_noise
self.apply_noise=apply_noise
self.decoders={}
self.use_hd=use_hd
self.data_basis=basis
self.data_thresholds=thresholds
self.data_saturations=saturations
self.data_alphas=alphas
self.data_Jbiases=Jbiases
self.storage=Storage(self,seed=seed,code=code,force_new=force_new)
self.seed=self.storage.seed
self.initialize_node()
self.mode='rate'
if self.sample_count<self.neurons:
self.decoder_mode='NxS'
self.decoder_noise_use_limit=True
else:
self.decoder_mode='NxN'
self.decoder_noise_use_gamma=True
def set(self,value,calc_output=True):
ArrayNode.set(self,value,calc_output=False)
if self.mode=='rate':
if value is None:
self._set_activity=None
else:
c=self.array_to_current(self._set_array)+self.Jbias
self._set_activity=self.current_to_activity(c)
if calc_output: self._calc_output()
def array(self):
if self._array is None:
if self.mode!='rate': return ArrayNode.array(self)
self._array=self.activity_to_array(self._output)
return self._array
def _calc_output(self):
if self.mode!='rate': return ArrayNode._calc_output(self)
if self._set_activity is not None:
out=self._set_activity
else:
out=self.current_to_activity(self.array_to_current(self.accumulator.value())+self.Jbias)
if self.apply_noise:
out=self.add_activation_noise(out)
self._output=out
def _transmit_rate_direct(self,conn,dt):
x=self.activity_to_array(self._output,decoder=self.get_decoder(conn.func))
conn.pop2.accumulator.add(conn.apply_weight(x),conn.tau,dt)
def _transmit_rate_rate(self,conn,dt):
x=self.activity_to_array(self._output,decoder=self.get_decoder(conn.func))
conn.pop2.accumulator.add(conn.apply_weight(x),conn.tau,dt)
def _transmit_direct_rate(self,conn,dt):
conn.pop2.accumulator.add(conn.apply_func_weight(self._output),conn.tau,dt)
def initialize_node(self):
random=numpy.random.RandomState(seed=self.seed)
if self.data_alphas is not None and self.data_Jbiases is not None:
self.alpha=numpy.array([self.data_alphas[i%len(self.data_alphas)] for i in range(self.neurons)],dtype=numpy.dtype('float32'))
self.Jbias=numpy.array([self.data_Jbiases[i%len(self.data_Jbiases)] for i in range(self.neurons)],dtype=numpy.dtype('float32'))
else:
if self.data_saturations is None:
sr=self.saturation_range
sat=random.uniform(sr[0],sr[1],(self.neurons,1))
else:
N=len(self.data_saturations)
sat=numpy.array([[self.data_saturations[i%N] for i in range(self.neurons)]],dtype=numpy.dtype('float32')).T
if self.data_thresholds is None:
delta=(self.max-self.min)*(1-self.threshold_coverage)*0.5
min_thresh=self.min+delta
max_thresh=self.max-delta
if self.threshold_min is not None: min_thresh=self.threshold_min
if self.threshold_max is not None: max_thresh=self.threshold_max
thresh=random.uniform(min_thresh,max_thresh,(self.neurons,1))
else:
N=len(self.data_thresholds)
thresh=numpy.array([[self.data_thresholds[i%N] for i in range(self.neurons)]],dtype=numpy.dtype('float32')).T
self.initialize_neurons(sat,thresh)
if self.data_basis is None:
g=makeGenerator(self.basis_style,self.dimensions,int(random.randint(0x7FFFFFFF)))
self.basis=g.get(self.neurons).T
else:
N=len(self.data_basis)
self.basis=numpy.array([self.data_basis[i%N] for i in range(self.neurons)],dtype=numpy.dtype('float32'))
if len(self.basis.shape)==1:
self.basis.shape=(self.neurons,self.dimensions)
self.sample_generator=makeGenerator(self.sample_style,self.dimensions,int(random.randint(0x7FFFFFFF)))
if self.min!=-1 or self.max!=1:
scale=(self.max-self.min)/2.0
if scale!=1.0: self.sample_generator.scale=scale
offset=(self.max+self.min)/2.0
if scale!=0.0: self.sample_generator.offset=offset
def initialize_neurons(self,saturations,thresholds):
saturations=numpy.array(saturations,dtype=numpy.float32)
thresholds=numpy.array(thresholds,dtype=numpy.float32)
x1,y1=thresholds,numpy.zeros((self.neurons,1),dtype=numpy.float32)
x2,y2=self.max,saturations
if not self.lif:
m=(y1-y2)/(x1-x2)
b=y1-m*x1
else:
numpy.seterr(divide='ignore')
z1=numpy.where(y1<=0,1.0,1.0/(1-numpy.exp((self.t_ref-(1.0/y1))/self.t_rc)))
z2=numpy.where(y2<=0,1.0,1.0/(1-numpy.exp((self.t_ref-(1.0/y2))/self.t_rc)))
numpy.seterr(divide='warn')
m=(z1-z2)/(x1-x2)
b=z1-m*x1
self.alpha,self.Jbias=m,b
self.Jbias.shape=self.Jbias.shape[0]
self.alpha.shape=self.alpha.shape[0]
def current_to_activity(self,J):
if not self.lif:
return numpy.maximum(J,0)
else:
J=numpy.maximum(J,0)
numpy.seterr(invalid='ignore',divide='ignore')
G=self.t_ref-self.t_rc*numpy.log(1-self.J_threshold/J)
G=numpy.where(G>0.001,1/G,0)
numpy.seterr(invalid='warn',divide='warn')
return G
def array_to_current(self,array):
b=self.basis
phi_x=numpy.dot(b,array)
J=self.alpha*phi_x
return J
def arrays_to_currents(self,arrays):
b=self.basis
phi_x=numpy.dot(b,arrays)
J=self.alpha*phi_x.T
return J.T
def activity_to_array(self,activity,decoder=None):
if activity is None: activity=numpy.zeros(self.neurons)
if self.lesion_size>0:
activity=activity[:]
activity[:self.lesion_size]=0.0
if self.lesion_cells is not None:
activity[self.lesion_cells]=0.0
if decoder is None:
decoder=self.get_decoder()
array=numpy.dot(activity.T,decoder)
return array
def add_activation_noise(self,actv,noise=None):
if noise is None: noise=self.activation_noise
if noise>0:
actv=numpy.random.normal(actv,self.saturation_range[1]*noise)
actv=numpy.maximum(0,actv)
return actv
def get_decoder(self,func=None,noise=None):
if self.decoder_mode=='NxN':
decoder=self.get_decoder_NxN
elif self.decoder_mode=='NxS':
decoder=self.get_decoder_NxS
else:
raise Exception('Unknown decoder mode: %s'%self.decoder_mode)
return decoder(func=func,noise=noise)
def get_decoder_NxN(self,func=None,noise=None):
if noise is None: noise=self.activation_noise
name='decoderNxN'
name_gamma='gamma'
name_gammainv='gammainv'
name_upsilon='upsilon'
if func:
hash_info=make_hash_info(func)
n='-%s-%08x'%(func.__name__,hash(tuple(hash_info)))#hash(tuple(func_id)))
n=n.replace('<','_').replace('>','_')
name_upsilon+=n
name+=n
name+='-%4.2f'%noise
name_gammainv+='-%4.2f'%noise
if noise>0:
if self.decoder_noise_use_gamma:
name+='g'
name_gammainv+='g'
if self.decoder_noise_use_limit:
name+='l'
name_gammainv+='l'
if self.decoder_noise_use_activity:
name+='a'
name_gammainv+='a'
name_upsilon+='-%4.2fa'%noise
name_gamma+='-%4.2fa'%noise
if name in self.decoders: return self.decoders[name]
decoder=self.storage.get(name,(self.neurons,-1))
if decoder is not None:
self.decoders[name]=decoder
return decoder
warning=False
if self.neurons>self.decoder_size_warning:
warning=True
print('Warning: calculating decoder for size %d neural group. This may take a while.'%self.neurons)
upsilon=None
upsilon=self.storage.get(name_upsilon,(self.neurons,-1))
need_gammainv=False
need_gamma=False
gamma_inv=self.storage.get(name_gammainv,(self.neurons,self.neurons))
if gamma_inv is None:
need_gammainv=True
gamma=self.storage.get(name_gamma,(self.neurons,self.neurons))
if gamma is None: need_gamma=True
if need_gamma or upsilon is None:
if self.dimensions>=3 and self.use_hd:
gamma,moments=calc_gamma_moments(self,dr=0.01)
gamma*=self.sample_count
ups=self.basis*moments[1].reshape((moments[1].shape[0],1))*self.sample_count
else:
self.sample_generator.reset()
count=self.sample_count
ups=None
while count>0:
s=min(count,self.sample_step_size)
if not self.sample_generator.can_continue(self.dimensions):
s=count
if warning:
print('processing %d of %d samples (%d left)'%(s,self.sample_count,count))
samples=self.sample_generator.get(s)
count-=s
curr=self.arrays_to_currents(samples)
curr=curr.T+self.Jbias
actv=self.current_to_activity(curr.T)
if self.decoder_noise_use_activity:
actv=numpy.random.normal(actv,noise*self.saturation_range[1])
actv=numpy.maximum(0,actv)
actv=numpy.array(actv,dtype=numpy.float32)
if need_gamma:
g=numpy.dot(actv,actv.T)
if gamma is None: gamma=g
else: gamma+=g
if upsilon is None:
samples=samples.T
if func is not None:
if self.dimensions==1:
samples=samples[:,0]
samples=numpy.array([self.value_to_array(func(self.array_to_value(x))) for x in samples],dtype=numpy.float32)
u=numpy.dot(actv,samples)
if ups is None: ups=u
else: ups+=u
if need_gamma:
self.storage.set(name_gamma,gamma)
if upsilon is None:
upsilon=ups
self.storage.set(name_upsilon,upsilon)
error_flag=False
if need_gammainv:
if self.decoder_noise_use_gamma:
if noise>0:
gamma+=numpy.identity(gamma.shape[0])*(((noise*self.saturation_range[1])**2)*self.sample_count)
if warning:
print('inverting %dx%d gamma matrix'%gamma.shape)
w,v=numpy.linalg.eigh(gamma)
limit=svd_limit*max(w)
if self.decoder_noise_use_limit:
if noise>0: limit=noise*noise*max(w)
for i in range(len(w)):
if w[i]<limit: w[i]=0
else: w[i]=1.0/w[i]
gamma_inv=numpy.dot(v,numpy.multiply(w[:,numpy.core.newaxis],v.T))
self.storage.set(name_gammainv,gamma_inv)
decoder=numpy.dot(gamma_inv,upsilon)
if len(decoder.shape)==1:
decoder.shape=decoder.shape[0],1
self.decoders[name]=decoder
self.storage.set(name,decoder)
return decoder
def get_decoder_NxS(self,func=None,noise=None):
if noise is None: noise=self.activation_noise
name='decoderNxS'
name_Ainv='Ainv'
name_B='B'
if func:
hash_info=make_hash_info(func)
n='-%s-%08x'%(func.__name__,hash(tuple(hash_info)))#hash(tuple(func_id)))
n=n.replace('<','_').replace('>','_')
name_B+=n
name+=n
name+='-%4.2f'%noise
name_Ainv+='-%4.2f'%noise
if noise>0:
if self.decoder_noise_use_limit:
name+='l'
name_Ainv+'l'
if self.decoder_noise_use_activity:
name+='a'
name_Ainv+='a'
name_B+='-%4.2fa'%noise
if name in self.decoders: return self.decoders[name]
decoder=self.storage.get(name,(self.neurons,-1))
if decoder is not None:
self.decoders[name]=decoder
return decoder
warning=False
if self.neurons>self.decoder_size_warning:
warning=True
print('Warning: calculating decoder for size %d neural group. This may take a while.'%self.neurons)
B=self.storage.get(name_B,(self.sample_count,-1))
Ainv=self.storage.get(name_Ainv,(self.sample_count,self.neurons))
need_B=B is None
need_A=Ainv is None
if B is None or Ainv is None:
self.sample_generator.reset()
count=self.sample_count
A=None
while count>0:
s=min(count,self.sample_step_size)
if not self.sample_generator.can_continue(self.dimensions):
s=count
if warning:
print('processing %d of %d samples (%d left)'%(s,self.sample_count,count))
samples=self.sample_generator.get(s)
count-=s
curr=self.arrays_to_currents(samples)
curr=curr.T+self.Jbias
actv=self.current_to_activity(curr.T)
if self.decoder_noise_use_activity:
actv=numpy.random.normal(actv,noise*self.saturation_range[1])
actv=numpy.maximum(0,actv)
actv=numpy.array(actv,dtype=numpy.float32)
if need_A:
if A is None: A=actv
else:
A=numpy.hstack((A,actv))
if need_B:
samples=samples.T
if func is not None:
if self.dimensions==1:
samples=samples[:,0]
samples=numpy.array([self.value_to_array(func(self.array_to_value(x))) for x in samples],dtype=numpy.float32)
if B is None: B=samples
else:
B=numpy.vstack((B,samples))
if need_A:
if warning:
print('inverting %dx%d activity matrix'%A.shape)
Ainv=numpy.linalg.pinv(A,rcond=noise*0.1)
self.storage.set(name_Ainv,Ainv)
if need_B:
self.storage.set(name_B,B)
decoder=numpy.dot(Ainv.T,B)
if len(decoder.shape)==1:
decoder.shape=decoder.shape[0],1
self.decoders[name]=decoder
self.storage.set(name,decoder)
return decoder
class ActivityNode(ArrayNode):
_set_activity=None
decoder_size_warning=1000
sample_step_size=500
decoder_mode='NxN'
decoder_noise_use_gamma=False
decoder_noise_use_limit=False
decoder_noise_use_activity=False
lesion_size=0
lesion_cells=None
def configure(self,neurons,lif=True,saturation_range=(200,300),
t_ref=0.001,t_rc=0.01,J_threshold=1,activation_noise=0.1,apply_noise=False,
threshold_coverage=0.9,threshold_min=None,threshold_max=None,
basis_style='Sphere',sample_style='DefaultSampling',
sample_count=None,seed=None,code=None,force_new=False,use_hd=False,
basis=None,thresholds=None,saturations=None,alphas=None,Jbiases=None):
self.neurons=neurons
self.lif=lif
self.saturation_range=saturation_range
self.t_ref=t_ref
self.t_rc=t_rc
self.J_threshold=J_threshold
self.basis_style=basis_style
self.sample_style=sample_style
self.threshold_coverage=threshold_coverage
self.threshold_min=threshold_min
self.threshold_max=threshold_max
if sample_count is None:
sample_count=self.dimensions*500
if sample_count>5000: sample_count=5000
self.sample_count=sample_count
self.activation_noise=activation_noise
self.apply_noise=apply_noise
self.decoders={}
self.use_hd=use_hd
self.data_basis=basis
self.data_thresholds=thresholds
self.data_saturations=saturations
self.data_alphas=alphas
self.data_Jbiases=Jbiases
self.storage=Storage(self,seed=seed,code=code,force_new=force_new)
self.seed=self.storage.seed
self.initialize_node()
self.mode='rate'
if self.sample_count<self.neurons:
self.decoder_mode='NxS'
self.decoder_noise_use_limit=True
else:
self.decoder_mode='NxN'
self.decoder_noise_use_gamma=True
def set(self,value,calc_output=True):
ArrayNode.set(self,value,calc_output=False)
if self.mode=='rate':
if value is None:
self._set_activity=None
else:
c=self.array_to_current(self._set_array)+self.Jbias
self._set_activity=self.current_to_activity(c)
if calc_output: self._calc_output()
def array(self):
if self._array is None:
if self.mode!='rate': return ArrayNode.array(self)
self._array=self.activity_to_array(self._output)
return self._array
def _calc_output(self):
if self.mode!='rate': return ArrayNode._calc_output(self)
if self._set_activity is not None:
out=self._set_activity
else:
out=self.current_to_activity(self.array_to_current(self.accumulator.value())+self.Jbias)
if self.apply_noise:
out=self.add_activation_noise(out)
self._output=out
def _transmit_rate_direct(self,conn,dt):
x=self.activity_to_array(self._output,decoder=self.get_decoder(conn.func))
conn.pop2.accumulator.add(conn.apply_weight(x),conn.tau,dt)
def _transmit_rate_rate(self,conn,dt):
x=self.activity_to_array(self._output,decoder=self.get_decoder(conn.func))
conn.pop2.accumulator.add(conn.apply_weight(x),conn.tau,dt)
def _transmit_direct_rate(self,conn,dt):
conn.pop2.accumulator.add(conn.apply_func_weight(self._output),conn.tau,dt)
def initialize_node(self):
random=numpy.random.RandomState(seed=self.seed)
if self.data_alphas is not None and self.data_Jbiases is not None:
self.alpha=numpy.array([self.data_alphas[i%len(self.data_alphas)] for i in range(self.neurons)],dtype=numpy.dtype('float32'))
self.Jbias=numpy.array([self.data_Jbiases[i%len(self.data_Jbiases)] for i in range(self.neurons)],dtype=numpy.dtype('float32'))
else:
if self.data_saturations is None:
sr=self.saturation_range
sat=random.uniform(sr[0],sr[1],(self.neurons,1))
else:
N=len(self.data_saturations)
sat=numpy.array([[self.data_saturations[i%N] for i in range(self.neurons)]],dtype=numpy.dtype('float32')).T
if self.data_thresholds is None:
delta=(self.max-self.min)*(1-self.threshold_coverage)*0.5
min_thresh=self.min+delta
max_thresh=self.max-delta
if self.threshold_min is not None: min_thresh=self.threshold_min
if self.threshold_max is not None: max_thresh=self.threshold_max
thresh=random.uniform(min_thresh,max_thresh,(self.neurons,1))
else:
N=len(self.data_thresholds)
thresh=numpy.array([[self.data_thresholds[i%N] for i in range(self.neurons)]],dtype=numpy.dtype('float32')).T
self.initialize_neurons(sat,thresh)
if self.data_basis is None:
g=makeGenerator(self.basis_style,self.dimensions,int(random.randint(0x7FFFFFFF)))
self.basis=g.get(self.neurons).T
else:
N=len(self.data_basis)
self.basis=numpy.array([self.data_basis[i%N] for i in range(self.neurons)],dtype=numpy.dtype('float32'))
if len(self.basis.shape)==1:
self.basis.shape=(self.neurons,self.dimensions)
self.sample_generator=makeGenerator(self.sample_style,self.dimensions,int(random.randint(0x7FFFFFFF)))
if self.min!=-1 or self.max!=1:
scale=(self.max-self.min)/2.0
if scale!=1.0: self.sample_generator.scale=scale
offset=(self.max+self.min)/2.0
if scale!=0.0: self.sample_generator.offset=offset
def initialize_neurons(self,saturations,thresholds):
saturations=numpy.array(saturations,dtype=numpy.float32)
thresholds=numpy.array(thresholds,dtype=numpy.float32)
x1,y1=thresholds,numpy.zeros((self.neurons,1),dtype=numpy.float32)
x2,y2=self.max,saturations
if not self.lif:
m=(y1-y2)/(x1-x2)
b=y1-m*x1
else:
numpy.seterr(divide='ignore')
z1=numpy.where(y1<=0,1.0,1.0/(1-numpy.exp((self.t_ref-(1.0/y1))/self.t_rc)))
z2=numpy.where(y2<=0,1.0,1.0/(1-numpy.exp((self.t_ref-(1.0/y2))/self.t_rc)))
numpy.seterr(divide='warn')
m=(z1-z2)/(x1-x2)
b=z1-m*x1
self.alpha,self.Jbias=m,b
self.Jbias.shape=self.Jbias.shape[0]
self.alpha.shape=self.alpha.shape[0]
def current_to_activity(self,J):
if not self.lif:
return numpy.maximum(J,0)
else:
J=numpy.maximum(J,0)
numpy.seterr(invalid='ignore',divide='ignore')
G=self.t_ref-self.t_rc*numpy.log(1-self.J_threshold/J)
G=numpy.where(G>0.001,1/G,0)
numpy.seterr(invalid='warn',divide='warn')
return G
def array_to_current(self,array):
b=self.basis
phi_x=numpy.dot(b,array)
J=self.alpha*phi_x
return J
def arrays_to_currents(self,arrays):
b=self.basis
phi_x=numpy.dot(b,arrays)
J=self.alpha*phi_x.T
return J.T
def activity_to_array(self,activity,decoder=None):
if activity is None: activity=numpy.zeros(self.neurons)
if self.lesion_size>0:
activity=activity[:]
activity[:self.lesion_size]=0.0
if self.lesion_cells is not None:
activity[self.lesion_cells]=0.0
if decoder is None:
decoder=self.get_decoder()
array=numpy.dot(activity.T,decoder)
return array
def add_activation_noise(self,actv,noise=None):
if noise is None: noise=self.activation_noise
if noise>0:
actv=numpy.random.normal(actv,self.saturation_range[1]*noise)
actv=numpy.maximum(0,actv)
return actv
def get_decoder(self,func=None,noise=None):
if self.decoder_mode=='NxN':
decoder=self.get_decoder_NxN
elif self.decoder_mode=='NxS':
decoder=self.get_decoder_NxS
elif self.decoder_mode=='binary':
decoder=self.get_decoder_binary
else:
raise Exception('Unknown decoder mode: %s'%self.decoder_mode)
return decoder(func=func,noise=noise)
def get_decoder_NxN(self,func=None,noise=None):
if noise is None: noise=self.activation_noise
name='decoderNxN'
name_gamma='gamma'
name_gammainv='gammainv'
name_upsilon='upsilon'
if func:
hash_info=make_hash_info(func)
n='-%s-%08x'%(func.__name__,hash(tuple(hash_info)))#hash(tuple(func_id)))
n=n.replace('<','_').replace('>','_')
name_upsilon+=n
name+=n
name+='-%4.2f'%noise
name_gammainv+='-%4.2f'%noise
if noise>0:
if self.decoder_noise_use_gamma:
name+='g'
name_gammainv+='g'
if self.decoder_noise_use_limit:
name+='l'
name_gammainv+='l'
if self.decoder_noise_use_activity:
name+='a'
name_gammainv+='a'
name_upsilon+='-%4.2fa'%noise
name_gamma+='-%4.2fa'%noise
if name in self.decoders: return self.decoders[name]
decoder=self.storage.get(name,(self.neurons,-1))
if decoder is not None:
self.decoders[name]=decoder
return decoder
warning=False
if self.neurons>self.decoder_size_warning:
warning=True
print 'Warning: calculating decoder for size %d neural group. This may take a while.'%self.neurons
upsilon=None
upsilon=self.storage.get(name_upsilon,(self.neurons,-1))
need_gammainv=False
need_gamma=False
gamma_inv=self.storage.get(name_gammainv,(self.neurons,self.neurons))
if gamma_inv is None:
need_gammainv=True
gamma=self.storage.get(name_gamma,(self.neurons,self.neurons))
if gamma is None: need_gamma=True
if need_gamma or upsilon is None:
if self.dimensions>=3 and self.use_hd:
gamma,moments=calc_gamma_moments(self,dr=0.01)
gamma*=self.sample_count
ups=self.basis*moments[1].reshape((moments[1].shape[0],1))*self.sample_count
else:
self.sample_generator.reset()
count=self.sample_count
ups=None
while count>0:
s=min(count,self.sample_step_size)
if not self.sample_generator.can_continue(self.dimensions):
s=count
if warning:
print 'processing %d of %d samples (%d left)'%(s,self.sample_count,count)
samples=self.sample_generator.get(s)
count-=s
curr=self.arrays_to_currents(samples)
curr=curr.T+self.Jbias
actv=self.current_to_activity(curr.T)
if self.decoder_noise_use_activity:
actv=numpy.random.normal(actv,noise*self.saturation_range[1])
actv=numpy.maximum(0,actv)
actv=numpy.array(actv,dtype=numpy.float32)
if need_gamma:
g=numpy.dot(actv,actv.T)
if gamma is None: gamma=g
else: gamma+=g
if upsilon is None:
samples=samples.T
if func is not None:
if self.dimensions==1:
samples=samples[:,0]
samples=numpy.array([self.value_to_array(func(self.array_to_value(x))) for x in samples],dtype=numpy.float32)
u=numpy.dot(actv,samples)
if ups is None: ups=u
else: ups+=u
if need_gamma:
self.storage.set(name_gamma,gamma)
if upsilon is None:
upsilon=ups
self.storage.set(name_upsilon,upsilon)
error_flag=False
if need_gammainv:
if self.decoder_noise_use_gamma:
if noise>0:
gamma+=numpy.identity(gamma.shape[0])*(((noise*self.saturation_range[1])**2)*self.sample_count)
if warning:
print 'inverting %dx%d gamma matrix'%gamma.shape
w,v=numpy.linalg.eigh(gamma)
limit=svd_limit*max(w)
if self.decoder_noise_use_limit:
if noise>0: limit=noise*noise*max(w)
for i in range(len(w)):
if w[i]<limit: w[i]=0
else: w[i]=1.0/w[i]
gamma_inv=numpy.dot(v,numpy.multiply(w[:,numpy.core.newaxis],v.T))
self.storage.set(name_gammainv,gamma_inv)
decoder=numpy.dot(gamma_inv,upsilon)
if len(decoder.shape)==1:
decoder.shape=decoder.shape[0],1
self.decoders[name]=decoder
self.storage.set(name,decoder)
return decoder
def get_decoder_NxS(self,func=None,noise=None):
if noise is None: noise=self.activation_noise
name='decoderNxS'
name_Ainv='Ainv'
name_B='B'
if func:
hash_info=make_hash_info(func)
n='-%s-%08x'%(func.__name__,hash(tuple(hash_info)))#hash(tuple(func_id)))
n=n.replace('<','_').replace('>','_')
name_B+=n
name+=n
name+='-%4.2f'%noise
name_Ainv+='-%4.2f'%noise
if noise>0:
if self.decoder_noise_use_limit:
name+='l'
name_Ainv+'l'
if self.decoder_noise_use_activity:
name+='a'
name_Ainv+='a'
name_B+='-%4.2fa'%noise
if name in self.decoders: return self.decoders[name]
decoder=self.storage.get(name,(self.neurons,-1))
if decoder is not None:
self.decoders[name]=decoder
return decoder
warning=False
if self.neurons>self.decoder_size_warning:
warning=True
print 'Warning: calculating decoder for size %d neural group. This may take a while.'%self.neurons
B=self.storage.get(name_B,(self.sample_count,-1))
Ainv=self.storage.get(name_Ainv,(self.sample_count,self.neurons))
need_B=B is None
need_A=Ainv is None
if B is None or Ainv is None:
self.sample_generator.reset()
count=self.sample_count
A=None
while count>0:
s=min(count,self.sample_step_size)
if not self.sample_generator.can_continue(self.dimensions):
s=count
if warning:
print 'processing %d of %d samples (%d left)'%(s,self.sample_count,count)
samples=self.sample_generator.get(s)
count-=s
curr=self.arrays_to_currents(samples)
curr=curr.T+self.Jbias
actv=self.current_to_activity(curr.T)
if self.decoder_noise_use_activity:
actv=numpy.random.normal(actv,noise*self.saturation_range[1])
actv=numpy.maximum(0,actv)
actv=numpy.array(actv,dtype=numpy.float32)
if need_A:
if A is None: A=actv
else:
A=numpy.hstack((A,actv))
if need_B:
samples=samples.T
if func is not None:
if self.dimensions==1:
samples=samples[:,0]
samples=numpy.array([self.value_to_array(func(self.array_to_value(x))) for x in samples],dtype=numpy.float32)
if B is None: B=samples
else:
B=numpy.vstack((B,samples))
if need_A:
if warning:
print 'inverting %dx%d activity matrix'%A.shape
Ainv=numpy.linalg.pinv(A,rcond=noise*0.1)
self.storage.set(name_Ainv,Ainv)
if need_B:
self.storage.set(name_B,B)
decoder=numpy.dot(Ainv.T,B)
if len(decoder.shape)==1:
decoder.shape=decoder.shape[0],1
self.decoders[name]=decoder
self.storage.set(name,decoder)
return decoder
def get_decoder_binary(self,func=None,noise=None):
if noise is None: noise=self.activation_noise
name='decoderBinary'
name_A='A'
name_B='B'
if func:
hash_info=make_hash_info(func)
n='-%s-%08x'%(func.__name__,hash(tuple(hash_info)))#hash(tuple(func_id)))
n=n.replace('<','_').replace('>','_')
name_B+=n
name+=n
name+='-%4.2f'%noise
name_A+='-%4.2f'%noise
if noise>0:
if self.decoder_noise_use_activity:
name+='a'
name_A+='a'
name_B+='-%4.2fa'%noise
if name in self.decoders: return self.decoders[name]
decoder=self.storage.get(name,(self.neurons,-1))
if decoder is not None:
self.decoders[name]=decoder
return decoder
warning=False
if self.neurons>self.decoder_size_warning:
warning=True
print 'Warning: calculating decoder for size %d neural group. This may take a while.'%self.neurons
B=self.storage.get(name_B,(self.sample_count,-1))
A=self.storage.get(name_A,(self.sample_count,self.neurons))
need_B=B is None
need_A=A is None
if B is None or A is None:
self.sample_generator.reset()
count=self.sample_count
A=None
while count>0:
s=min(count,self.sample_step_size)
if not self.sample_generator.can_continue(self.dimensions):
s=count
if warning:
print 'processing %d of %d samples (%d left)'%(s,self.sample_count,count)
samples=self.sample_generator.get(s)
count-=s
curr=self.arrays_to_currents(samples)
curr=curr.T+self.Jbias
actv=self.current_to_activity(curr.T)
if self.decoder_noise_use_activity:
actv=numpy.random.normal(actv,noise*self.saturation_range[1])
actv=numpy.maximum(0,actv)
actv=numpy.array(actv,dtype=numpy.float32)
if need_A:
if A is None: A=actv
else:
A=numpy.hstack((A,actv))
if need_B:
samples=samples.T
if func is not None:
if self.dimensions==1:
samples=samples[:,0]
samples=numpy.array([self.value_to_array(func(self.array_to_value(x))) for x in samples],dtype=numpy.float32)
if B is None: B=samples
else:
B=numpy.vstack((B,samples))
if need_A:
self.storage.set(name_A,A)
if need_B:
self.storage.set(name_B,B)
options=[0.0,1.0,-1.0]#,2.0,-2.0,3.0,-3.0]
import random
dec=[]
for index in range(B.shape[1]):
d=None
e=None
B2=B[:,index]
for a in range(10):
decoder=numpy.array([random.choice(options) for i in range(len(A))])
if len(decoder.shape)==1:
decoder.shape=decoder.shape[0],1
scale=numpy.linalg.norm(B2)/numpy.linalg.norm(numpy.dot(A.T,decoder))
decoder*=scale
error=numpy.linalg.norm(numpy.dot(A.T,decoder)-B2)
for i in range(500):
#print i,error
j=random.randrange(len(actv))
val=random.choice(options)*scale
if val!=decoder[j][0]:
oldval=decoder[j][0]
decoder[j][0]=val
error2=numpy.linalg.norm(numpy.dot(A.T,decoder)-B2)
if error2>=error:
decoder[j][0]=oldval
else:
error=error2
scale2=numpy.linalg.norm(B2)/numpy.linalg.norm(numpy.dot(A.T,decoder))
scale*=scale2
decoder*=scale2
if e is None or error<e:
d=decoder
e=error
dec.append(d)
decoder=numpy.array(dec).T
decoder.shape=self.basis.shape
#import scipy.io as sio
#sio.savemat('example%08x.mat'%self.seed,dict(A=A,B=B,dec=decoder))
self.decoders[name]=decoder
self.storage.set(name,decoder)
return decoder
class ActivityNode(ArrayNode):
_set_activity = None
decoder_size_warning = 1000
decoder_size_limit = 5000
sample_step_size = 500
def configure(self,
neurons,
lif=True,
saturation_range=(200, 300),
t_ref=0.001,
t_rc=0.01,
J_threshold=1,
activation_noise=0.1,
apply_noise=True,
threshold_coverage=0.9,
threshold_min=None,
threshold_max=None,
basis_style='Sphere',
sample_style='DefaultSampling',
sample_count=None,
seed=None,
code=None,
force_new=False,
use_hd=False,
basis=None,
thresholds=None,
saturations=None):
self.neurons = neurons
self.lif = lif
self.saturation_range = saturation_range
self.t_ref = t_ref
self.t_rc = t_rc
self.J_threshold = J_threshold
self.basis_style = basis_style
self.sample_style = sample_style
self.threshold_coverage = threshold_coverage
self.threshold_min = threshold_min
self.threshold_max = threshold_max
if sample_count is None: sample_count = self.dimensions * 500
self.sample_count = sample_count
self.activation_noise = activation_noise
self.apply_noise = apply_noise
self.decoders = {}
self.use_hd = use_hd
self.data_basis = basis
self.data_thresholds = thresholds
self.data_saturations = saturations
self.storage = Storage(self, seed=seed, code=code, force_new=force_new)
self.seed = self.storage.seed
self.initialize_node()
self.mode = 'rate'
def set(self, value, calc_output=True):
ArrayNode.set(self, value, calc_output=False)
if self.mode == 'rate':
c = self.array_to_current(self._set_array) + self.Jbias
self._set_activity = self.current_to_activity(c)
if calc_output: self._calc_output()
def array(self):
if self._array is None:
if self.mode != 'rate': return ArrayNode.array(self)
self._array = self.activity_to_array(self._output)
return self._array
def _calc_output(self):
if self.mode != 'rate': return ArrayNode._calc_output(self)
if self._set_activity is not None:
out = self._set_activity
elif self._input is not None:
out = self.current_to_activity(self._input)
else:
out = numpy.zeros(self.neurons)
if self.apply_noise:
out = self.add_activation_noise(out)
self._output = out
def _clear_inputs(self):
if self.mode != 'rate': return ArrayNode._clear_inputs(self)
if self._input is None: self._input = numpy.zeros(self.neurons)
self._input[:] = self.Jbias
def _transmit_rate_direct(self, conn):
x = self.activity_to_array(self._output,
decoder=self.get_decoder(conn.func))
conn.pop2._input += conn.apply_weight(x)
def _transmit_rate_rate(self, conn):
x = self.activity_to_array(self._output,
decoder=self.get_decoder(conn.func))
x = conn.apply_weight(x)
conn.pop2._input += conn.pop2.array_to_current(x)
def _transmit_direct_rate(self, conn):
conn.pop2._input += conn.pop2.array_to_current(
conn.apply_func_weight(self._output))
def initialize_node(self):
random = numpy.random.RandomState(seed=self.seed)
if self.data_saturations is None:
sr = self.saturation_range
sat = random.uniform(sr[0], sr[1], (self.neurons, 1))
else:
N = len(self.data_saturations)
sat = numpy.array(
[[self.data_saturations[i % N] for i in range(self.neurons)]],
dtype=numpy.dtype('float32')).T
if self.data_thresholds is None:
delta = (self.max - self.min) * (1 - self.threshold_coverage) * 0.5
min_thresh = self.min + delta
max_thresh = self.max - delta
if self.threshold_min is not None: min_thresh = self.threshold_min
if self.threshold_max is not None: max_thresh = self.threshold_max
thresh = random.uniform(min_thresh, max_thresh, (self.neurons, 1))
else:
N = len(self.data_thresholds)
thresh = numpy.array(
[[self.data_thresholds[i % N] for i in range(self.neurons)]],
dtype=numpy.dtype('float32')).T
self.initialize_neurons(sat, thresh)
if self.data_basis is None:
g = makeGenerator(self.basis_style, self.dimensions,
int(random.randint(0x7FFFFFFF)))
self.basis = g.get(self.neurons).T
else:
N = len(self.data_basis)
self.basis = numpy.array(
[self.data_basis[i % N] for i in range(self.neurons)],
dtype=numpy.dtype('float32'))
if len(self.basis.shape) == 1:
self.basis.shape = (self.neurons, self.dimensions)
self.sample_generator = makeGenerator(self.sample_style,
self.dimensions,
int(random.randint(0x7FFFFFFF)))
if self.min != -1 or self.max != 1:
scale = (self.max - self.min) / 2.0
if scale != 1.0: self.sample_generator.scale = scale
offset = (self.max + self.min) / 2.0
if scale != 0.0: self.sample_generator.offset = offset
def initialize_neurons(self, saturations, thresholds):
saturations = numpy.array(saturations, dtype=numpy.float32)
thresholds = numpy.array(thresholds, dtype=numpy.float32)
x1, y1 = thresholds, numpy.zeros((self.neurons, 1),
dtype=numpy.float32)
x2, y2 = self.max, saturations
if not self.lif:
m = (y1 - y2) / (x1 - x2)
b = y1 - m * x1
else:
numpy.seterr(divide='ignore')
z1 = numpy.where(
y1 <= 0, 1.0, 1.0 / (1 - numpy.exp(
(self.t_ref - (1.0 / y1)) / self.t_rc)))
z2 = numpy.where(
y2 <= 0, 1.0, 1.0 / (1 - numpy.exp(
(self.t_ref - (1.0 / y2)) / self.t_rc)))
numpy.seterr(divide='warn')
m = (z1 - z2) / (x1 - x2)
b = z1 - m * x1
self.alpha, self.Jbias = m, b
self.Jbias.shape = self.Jbias.shape[0]
self.alpha.shape = self.alpha.shape[0]
def current_to_activity(self, J):
if not self.lif:
return numpy.maximum(J, 0)
else:
J = numpy.maximum(J, 0)
numpy.seterr(invalid='ignore', divide='ignore')
G = self.t_ref - self.t_rc * numpy.log(1 - self.J_threshold / J)
G = numpy.where(G > 0.001, 1 / G, 0)
numpy.seterr(invalid='warn', divide='warn')
return G
def array_to_current(self, array):
b = self.basis
phi_x = numpy.dot(b, array)
J = self.alpha * phi_x
return J
def arrays_to_currents(self, arrays):
b = self.basis
phi_x = numpy.dot(b, arrays)
J = self.alpha * phi_x.T
return J.T
def activity_to_array(self, activity, decoder=None):
if activity is None: activity = numpy.zeros(self.neurons)
if decoder is None:
decoder = self.get_decoder()
array = numpy.dot(activity.T, decoder)
return array
def add_activation_noise(self, actv, noise=None):
if noise is None: noise = self.activation_noise
if noise > 0:
actv = numpy.random.normal(actv, self.saturation_range[1] * noise)
actv = numpy.maximum(0, actv)
return actv
def get_decoder(self, func=None, noise=None):
if noise is None: noise = self.activation_noise
name = ''
name_without_noise = ''
if func:
hash_info = make_hash_info(func)
name = '-%s-%08x' % (func.__name__, hash(tuple(hash_info))
) #hash(tuple(func_id)))
name = name.replace('<', '_').replace('>', '_')
#print 'name',name
name_without_noise = name
name += '-%4.2f' % noise
if name in self.decoders: return self.decoders[name]
decoder = self.storage.get('decoder' + name, (self.neurons, -1))
if decoder is not None:
self.decoders[name] = decoder
return decoder
warning = False
if self.neurons > self.decoder_size_warning:
warning = True
print 'Warning: calculating decoder for size %d neural group.' % self.neurons
if self.neurons > self.decoder_size_limit:
print ' Gamma matrices will be created, but they are too large for one machine to invert.'
else:
print ' This may take a few minutes.'
noise *= self.saturation_range[1]
upsilon = None
upsilon = self.storage.get('upsilon' + name_without_noise,
(self.neurons, -1))
gamma_inv = None
need_gamma = True
gamma_s = self.storage.get('gamma.s', self.neurons)
if gamma_s is not None:
gamma_v = self.storage.get('gamma.v', (self.neurons, self.neurons))
if svd_limit is not None:
gamma_s = gamma_s[gamma_s > gamma_s[0] * svd_limit]
gamma_v = gamma_v[:len(gamma_s), :]
s_inv = numpy.diag(1.0 /
(gamma_s + noise * noise * self.sample_count))
gamma_inv = numpy.dot(numpy.dot(gamma_v.T, s_inv), gamma_v)
need_gamma = False
else:
gamma = self.storage.get('gamma', (self.neurons, self.neurons))
if gamma is not None: need_gamma = False
if need_gamma or upsilon is None:
if self.dimensions >= 3 and self.use_hd:
gamma, moments = calc_gamma_moments(self, dr=0.01)
gamma *= self.sample_count
ups = self.basis * moments[1].reshape(
(moments[1].shape[0], 1)) * self.sample_count
else:
self.sample_generator.reset()
count = self.sample_count
ups = None
while count > 0:
s = min(count, self.sample_step_size)
if not self.sample_generator.can_continue(self.dimensions):
s = count
if warning:
print 'processing %d of %d samples (%d left)' % (
s, self.sample_count, count)
samples = self.sample_generator.get(s)
count -= s
curr = self.arrays_to_currents(samples)
curr = curr.T + self.Jbias
actv = self.current_to_activity(curr.T)
#actv=self.array_to_activity(samples)
if need_gamma:
g = numpy.dot(actv, actv.T)
if gamma is None: gamma = g
else: gamma += g
if upsilon is None:
samples = samples.T
if func is not None:
if self.dimensions == 1:
samples = samples[:, 0]
samples = numpy.array([
self.value_to_array(
func(self.array_to_value(x)))
for x in samples
],
dtype=numpy.float32)
u = numpy.dot(actv, samples)
if ups is None: ups = u
else: ups += u
if need_gamma:
self.storage.set('gamma', gamma)
if upsilon is None:
upsilon = ups
self.storage.set('upsilon' + name_without_noise, upsilon)
error_flag = False
if gamma_inv is None:
if gamma.shape[0] > self.decoder_size_limit:
print 'Need other program to invert ' + self.storage.path(
'gamma')
gamma_inv = numpy.zeros(gamma.shape, dtype=numpy.float32)
error_flag = True
else:
if noise > 0:
gamma += numpy.identity(
gamma.shape[0]) * (noise * noise) * self.sample_count
if gamma.shape[0] > self.decoder_size_warning:
print 'inverting %dx%d gamma matrix' % gamma.shape
gamma_inv = numpy.linalg.pinv(gamma)
decoder = numpy.dot(gamma_inv, upsilon)
if len(decoder.shape) == 1:
decoder.shape = decoder.shape[0], 1
self.decoders[name] = decoder
if not error_flag:
self.storage.set('decoder' + name, decoder)
return decoder