def run(tag='hierarchical', ind=[], par={'N_e': 900}):
sm = StorageManager(tag, random_nr=True, print_msg=display)
sm.save_param_dict(par)
#source = FDTGrammarActivator_New(tag='grammar_act', random_blocks=True, input_density=10/par['N_e'])#.plot_char_input_statistics()#output_size=par['N_e']#15
#source = LongDelayGrammar(tag='grammar_act', output_size=par['N_e'], random_blocks=True, mode=['very simple'], input_density=10/par['N_e'])#.print_test().plot_char_input_statistics()
#print(source.get_text_score('. parrot likes trees. wolf wolf wolf..'))
source = Line_Patterns(tag='image_act',
group_possibility=1.0,
grid_width=30,
grid_height=30,
center_x=list(range(30)),
center_y=30 / 2,
degree=90,
line_length=60)
#source = Line_Patterns(tag='image_act', group_possibility=1.0, grid_width=10, grid_height=10, center_x=10 / 2, center_y=list(range(10)), degree=0, line_length=20)
#source = Line_Patterns(tag='image_act', group_possibility=1.0, grid_width=10, grid_height=10, center_x=10 / 2, center_y=10 / 2, degree=list(np.arange(0, 360, 20)), line_length=20)
#source = MNIST_Patterns(tag='image_act', group_possibility=1.0, repeat_same_label_time=1)
#source = TNAP_Image_Patches(tag='image_act', image_path='../../../Images/pexels-photo-275484.jpeg', grid_width=20, grid_height=20, dimensions=['on_center_white'], patch_norm=True)#'red', 'green', 'blue', 'gray', '255-red', '255-green', '255-blue', '255-gray',, 'rgbw', '255-rgbw''off_center_white',
SORN = Network([], [], initialize=False)
for timecale in [1]: #2
e_ng = NeuronGroup(
net=SORN,
tag='main_exc_group,ts={}'.format(timecale),
size=get_squared_dim(int(par['N_e'])),
behaviour={
1:
STDP_simple(exponent='[3.0#1]',
post_learn_value='[6.55#2]'), # 8.92 #6.55
2:
TemporalWeightCache(decay=1,
strength=1,
GLU_density=0.5,
set_weights=True),
# 3: RandomWeightFluctuation(beta=4*0.2, gamma=3*0.2),
3:
GlutamateCacheConvergeAndNormalization(
norm_value='0.1;0.6'), # ,#20.0 #(norm_value=0.001)#40
5:
InterGammaGlutamate(), #0.005
6:
IntraGammaGlutamate(),
7:
IntraGammaGABA(GABA_density=1.0,
GABA_random_factor=10.0,
GABA_Norm='[3.8#0]'), #3.4 #3.8
8:
ActivityBuffering(
activity_multiplyer=0.0,
firetreshold='0.1;+-0%'), #todo warning: relu deactivated
9:
HomeostaticMechanism(range_end='[550#3];+-30%',
inc='[0.184#4];+-10%',
dec='[0.492#5];+-10%',
pattern_chance='[0.0086#6];+-30%'
), #target_max=20.0#0.0086 #,'
10:
RandomLeakInput(random_strength='[0.5#7]'),
11:
additional(),
100:
Recorder(['n.output'], tag='exc_out_rec')
})
#i_ng['structure', 0].stretch_to_equal_size(e_ng)
SynapseGroup(
net=SORN, src=e_ng, dst=e_ng, tag='GLU,e->e'
) #.partition([10, 10], [6, 6]) #, connectivity='(np.abs(sx-dx)<10)*(np.abs(sy-dy)<10)'
SynapseGroup(
net=SORN, src=e_ng, dst=e_ng, tag='GABA,e->e'
) #.partition([10, 10], [6, 6]) #, connectivity='(np.abs(sx-dx)<10)*(np.abs(sy-dy)<10)',
#SynapseGroup(net=SORN, src=e_ng, dst=i_ng, connectivity='(np.abs(sx-dx)<10)*(np.abs(sy-dy)<10)', tag='GLU,e->i').partition([10, 10], [4, 4])
#SynapseGroup(net=SORN, src=i_ng, dst=e_ng, connectivity='(np.abs(sx-dx)<10)*(np.abs(sy-dy)<10)', tag='GABA,i->e')
#SynapseGroup(net=SORN, src=i_ng, dst=i_ng, connectivity='(np.abs(sx-dx)<10)*(np.abs(sy-dy)<10)', tag='GABA,i->i')
e_ng.add_behaviour(
4, TREN_external_input(strength=0.5, pattern_groups=[source]))
# i_ng.add_behaviour(10, SORN_external_input(strength=0.3, pattern_groups=[source]))
#if timecale == 1:
e_ng.add_behaviour(101, Recorder(['n.pattern_index'], tag='inp_rec'))
#else:
# #forward synapses
# SynapseGroup(net=SORN, src=last_e_ng, dst=e_ng, tag='GLU,e->e(+1)').partition([10, 10], [6, 6])
# SynapseGroup(net=SORN, src=last_e_ng, dst=e_ng, tag='GABA,e->i(+1)').partition([10, 10], [2, 2])
# #backward synapses
# SynapseGroup(net=SORN, src=e_ng, dst=last_e_ng, tag='GLU,e(+1)->e').partition([10, 10], [6, 6])
# SynapseGroup(net=SORN, src=e_ng, dst=last_e_ng, tag='GABA,e(+1)->i').partition([10, 10], [2, 2])
#last_e_ng = e_ng
#last_i_ng = i_ng
SORN.set_marked_variables(ind, info=(ind == []))
SORN.initialize(info=False)
############################################################################################################################################################
score = 0
#SORN.simulate_iterations(100, 100, measure_block_time=True, disable_recording=True)
import Exploration.Network_UI.Network_UI as SUI
SUI.Network_UI(SORN,
label='T',
exc_group_name='main_exc_group',
inh_group_name='main_inh_group',
storage_manager=sm).show()
score += get_evolution_score_words(
SORN,
15000,
5000,
2000,
display=False,
stdp_off=True,
same_timestep_without_feedback_loop=True,
steps_recovery=15000) #, steps_recovery=15000
#score += get_oscillation_score_hierarchical(SORN, 0, 5000)
return score
from PymoNNto.Exploration.Network_UI.Network_UI import *
from PymoNNto.NetworkBehaviour.Structure.Structure import *
#print('\r\n\r\nNumpy Version:')
#net = Network()
#ng = NeuronGroup(net=net, size=1000000, behaviour={
# 1: TestModuleNumpy(),
#})
#net.initialize(info=False)
#net.simulate_iterations(1000, batch_size=1000, measure_block_time=True)
#print('\r\n\r\nTensorflow Version:')
net = Network()
ng = NeuronGroup(net=net,
size=get_squared_dim(100000),
behaviour={
1: TestModuleTensorflow(),
})
mask = np.zeros(ng.size)
mask[1] = 1
mask[5] = 1
sub_ng = ng.subGroup(mask)
net.initialize(info=False)
#sub_ng.voltage = [5, 1]
#print(np.array(ng.voltage))
def run(
attrs={
'name': 'KWTA',
'ind': [],
'N_e': 900,
'TS': [1],
'ff': True,
'fb': True,
'plastic': 15000
}):
so = True
print_info = attrs.get('print', True)
if print_info:
print(attrs)
sm = StorageManager(attrs['name'], random_nr=True, print_msg=print_info)
sm.save_param_dict(attrs)
source = FewSentencesGrammar(tag='grammar_act',
output_size=attrs['N_e'],
random_blocks=True,
input_density=0.015)
#source = SingleWordGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, input_density=0.015)
#source = FewLongSentencesGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, input_density=0.015)
#source.plot_char_frequency_histogram(20)
SORN = Network()
e_ng = NeuronGroup(
net=SORN,
tag='PC_{},prediction_source'.format(1),
size=get_squared_dim(attrs['N_e']),
behaviour={
2:
SORN_init_neuron_vars_no_reset(timescale=1),
#3: SORN_init_afferent_synapses(transmitter='GLU', density='90%', distribution='uniform(0.5,1.0)', normalize=True),
5:
SORN_init_afferent_synapses(transmitter='GLU_cluster',
density='90%',
distribution='uniform(0.9,1.0)',
normalize=True),
#10.0: SORN_slow_syn(transmitter='GLU', strength='1.0', so=so),
10.1:
SORN_IP_WTA_apply(),
10.15:
WTA_refrac_apply(strengthfactor=0.1),
#10.2: SORN_generate_output_K_WTA_partitioned(K='0.24'),
8.3:
SORN_slow_syn_simple(transmitter='GLU_cluster', strength='1.0'),
8.4:
SORN_generate_output_K_WTA_partitioned(
K='0.04', filter_temporal_output=False),
#10.5: SORN_slow_syn_simple(transmitter='GLU_cluster', strength='1.0'),
#10.6: SORN_generate_output_K_WTA_partitioned(K='0.08', filter_temporal_output=True),
#10.7: SORN_slow_syn_simple(transmitter='GLU_cluster', strength='1.0'),
#10.8: SORN_generate_output_K_WTA_partitioned(K='0.04', filter_temporal_output=True),
15:
SORN_buffer_variables(random_temporal_output_shift=False),
18:
WTA_refrac(),
20:
SORN_IP_WTA(h_ip='lognormal_real_mean(0.04, 0.2944)',
eta_ip='0.007',
clip_min=None),
#21.1: SORN_STDP(transmitter='GLU', eta_stdp='0.00015', STDP_F={-1: 1, 1: -1}),#, 0: 1 #[0.00015#7]
21.2:
SORN_STDP(transmitter='GLU_cluster',
eta_stdp='0.00015',
STDP_F={0: 0.5}), #[0.00015#7]
#22: SORN_SN(syn_type='GLU', behaviour_norm_factor=1.0),
23:
SORN_SN(syn_type='GLU_cluster', behaviour_norm_factor=0.1)
})
SynapseGroup(net=SORN,
src=e_ng,
dst=e_ng,
tag='syn,GLU',
connectivity='(s_id!=d_id)*in_box(10)') #, partition=True)
#SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='syn,GLU_cluster', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
#SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='GLU', connectivity='(s_id!=d_id)*in_box(10)')#, partition=True)
SynapseGroup(
net=SORN,
src=e_ng,
dst=e_ng,
tag='GLU_cluster',
connectivity=
'(s_id!=d_id)*((sy<0)*(dy<0)*(sx<0)*(dx<0)+(sy>=0)*(dy>=0)*(sx>=0)*(dx>=0)+(sy<0)*(dy<0)*(sx>=0)*(dx>=0)+(sy>=0)*(dy>=0)*(sx<0)*(dx<0))'
)
'''
'(s_id!=d_id)*('+
'(s_id>=35*0)*(d_id>=35*0)*(s_id<35*1)*(d_id<35*1)+' +
'(s_id>=35*1)*(d_id>=35*1)*(s_id<35*2)*(d_id<35*2)+' +
'(s_id>=35*2)*(d_id>=35*2)*(s_id<35*3)*(d_id<35*3)+' +
'(s_id>=35*3)*(d_id>=35*3)*(s_id<35*4)*(d_id<35*4)+' +
'(s_id>=35*4)*(d_id>=35*4)*(s_id<35*5)*(d_id<35*5)+' +
'(s_id>=35*5)*(d_id>=35*5)*(s_id<35*6)*(d_id<35*6)+' +
'(s_id>=35*6)*(d_id>=35*6)*(s_id<35*7)*(d_id<35*7)+' +
'(s_id>=35*7)*(d_id>=35*7)*(s_id<35*8)*(d_id<35*8)+' +
'(s_id>=35*8)*(d_id>=35*8)*(s_id<35*9)*(d_id<35*9)+' +
'(s_id>=35*9)*(d_id>=35*9)*(s_id<35*10)*(d_id<35*10)+' +
'(s_id>=35*10)*(d_id>=35*10)*(s_id<35*11)*(d_id<35*11)+' +
'(s_id>=35*11)*(d_id>=35*11)*(s_id<35*12)*(d_id<35*12)+' +
'(s_id>=35*12)*(d_id>=35*12)*(s_id<35*13)*(d_id<35*13)+' +
'(s_id>=35*13)*(d_id>=35*13)*(s_id<35*14)*(d_id<35*14)+' +
'(s_id>=35*14)*(d_id>=35*14)*(s_id<35*15)*(d_id<35*15)+' +
'(s_id>=35*15)*(d_id>=35*15)*(s_id<35*16)*(d_id<35*16)+' +
'(s_id>=35*16)*(d_id>=35*16)*(s_id<35*17)*(d_id<35*17)+' +
'(s_id>=35*17)*(d_id>=35*17)*(s_id<35*18)*(d_id<35*18)+' +
'(s_id>=35*18)*(d_id>=35*18)*(s_id<35*19)*(d_id<35*19)+' +
'(s_id>=35*19)*(d_id>=35*19)*(s_id<35*20)*(d_id<35*20)+' +
'(s_id>=35*20)*(d_id>=35*20)*(s_id<35*21)*(d_id<35*21)+' +
'(s_id>=35*21)*(d_id>=35*21)*(s_id<35*22)*(d_id<35*22)+' +
'(s_id>=35*22)*(d_id>=35*22)*(s_id<35*23)*(d_id<35*23)+' +
'(s_id>=35*23)*(d_id>=35*23)*(s_id<35*24)*(d_id<35*24)+' +
'(s_id>=35*24)*(d_id>=35*24)*(s_id<35*25)*(d_id<35*25)+' +
'(s_id>=35*25)*(d_id>=35*25)*(s_id<35*26)*(d_id<35*26)+' +
'(s_id>=35*26)*(d_id>=35*26)*(s_id<35*27)*(d_id<35*27)+' +
'(s_id>=35*27)*(d_id>=35*27)*(s_id<35*28)*(d_id<35*28)+' +
'(s_id>=35*28)*(d_id>=35*28)*(s_id<35*29)*(d_id<35*29)+' +
'(s_id>=35*29)*(d_id>=35*29)*(s_id<35*30)*(d_id<35*30)+' +
'(s_id>=35*30)*(d_id>=35*30)*(s_id<35*31)*(d_id<35*31)+' +
'(s_id>=35*31)*(d_id>=35*31)*(s_id<35*32)*(d_id<35*32)+' +
'(s_id>=35*32)*(d_id>=35*32)*(s_id<35*33)*(d_id<35*33)+' +
'(s_id>=35*33)*(d_id>=35*33)*(s_id<35*34)*(d_id<35*34)+' +
'(s_id>=35*34)*(d_id>=35*34)*(s_id<35*35)*(d_id<35*35)+' +
'(s_id>=35*35)*(d_id>=35*35)*(s_id<35*36)*(d_id<35*36)+' +
'(s_id>=35*36)*(d_id>=35*36)*(s_id<35*37)*(d_id<35*37)+' +
'(s_id>=35*37)*(d_id>=35*37)*(s_id<35*38)*(d_id<35*38)+' +
'(s_id>=35*38)*(d_id>=35*38)*(s_id<35*39)*(d_id<35*39)+' +
'(s_id>=35*39)*(d_id>=35*39)*(s_id<35*40)*(d_id<35*40)' +
')')#, partition=True)#
'(s_id>=140*0)*(d_id>=140*0)*(s_id<140*1)*(d_id<140*1)+' +
'(s_id>=140*1)*(d_id>=140*1)*(s_id<140*2)*(d_id<140*2)+' +
'(s_id>=140*2)*(d_id>=140*2)*(s_id<140*3)*(d_id<140*3)+' +
'(s_id>=140*3)*(d_id>=140*3)*(s_id<140*4)*(d_id<140*4)+' +
'(s_id>=140*4)*(d_id>=140*4)*(s_id<140*5)*(d_id<140*5)+' +
'(s_id>=140*5)*(d_id>=140*5)*(s_id<140*6)*(d_id<140*6)+' +
'(s_id>=140*6)*(d_id>=140*6)*(s_id<140*7)*(d_id<140*7)+' +
'(s_id>=140*7)*(d_id>=140*7)*(s_id<140*8)*(d_id<140*8)+' +
'(s_id>=140*8)*(d_id>=140*8)*(s_id<140*9)*(d_id<140*9)+' +
'(s_id>=140*9)*(d_id>=140*9)*(s_id<140*10)*(d_id<140*10)' +'''
#SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='GLU_cluster', connectivity='(s_id!=d_id)*(sy>=20)*(dy>=20)')
e_ng.add_behaviour(9, SORN_K_random_input())
if __name__ == '__main__' and attrs.get('UI', False):
e_ng.color = get_color(0, 1)
SORN.set_marked_variables(info=print_info, storage_manager=sm)
SORN.initialize(info=False)
#for i, syn in enumerate(SORN['syn']):
# syn.enabled = np.load('Data/E{}.npy'.format(i))
# syn.W = np.load('Data/W{}.npy'.format(i))
###################################################################################################################
if __name__ == '__main__' and attrs.get('UI', False):
Network_UI(SORN,
label='SORN UI K_WTA',
storage_manager=sm,
group_display_count=1,
reduced_layout=False).show()
score = 0
plastic_steps = attrs.get('plastic', 20000)
for i in range(1):
sm = StorageManager(attrs['name'] + '[{:03d}]'.format(i + 1),
random_nr=True,
print_msg=print_info)
sm.save_param_dict(attrs)
score += train_and_generate_text(
SORN,
plastic_steps,
5000,
1000,
display=print_info,
stdp_off=True,
same_timestep_without_feedback_loop=True,
steps_recovery=1000,
storage_manager=sm)
print('score=', score)
return score
def run(tag='hierarchical', ind=[], par={'N_e': 900, 'TS': [1]}):
sm = StorageManager(tag, random_nr=True, print_msg=display)
sm.save_param_dict(par)
#source = FDTGrammarActivator_New(tag='grammar_act', random_blocks=True, input_density=15/par['N_e'])#.plot_char_input_statistics()#output_size=par['N_e']#15
#source = LongDelayGrammar(tag='grammar_act', output_size=par['N_e'], random_blocks=True, mode=['very simple'], input_density=10/par['N_e'])#.print_test().plot_char_input_statistics()
#source = Line_Patterns(tag='image_act', group_possibility=1.0, grid_width=20, grid_height=20, center_x=list(range(20)), center_y=30 / 2, degree=90, line_length=60)
#print(source.get_text_score('. parrot likes trees. wolf wolf wolf..'))
source = MNIST_Patterns(tag='image_act',
group_possibility=1.0,
repeat_same_label_time=1)
SORN = Network()
for timecale in par['TS']:
e_ng = NeuronGroup(
net=SORN,
tag='PC_Neurons_{},prediction_source'.format(timecale),
size=get_squared_dim(int(par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timecale,
init_TH='0.05;+-80%',
activation_function='binary'),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='13%',
distribution='lognormal(0,[0.89#0])',
normalize=True,
partition_compensation=True),
4:
SORN_init_afferent_synapses(
transmitter='GABA',
density='45%',
distribution='lognormal(0,[0.80222#1])',
normalize=True), #electrical synapses
12:
SORN_slow_syn(transmitter='GLU', strength='[0.14#2]'),
13:
SORN_slow_syn(transmitter='GABA', strength='-[0.14#3]'),
17:
SORN_fast_syn(transmitter='GABA', strength='-[0.03#4]'),
18:
SORN_random_act(chance=0.001),
19:
SORN_input_collect(),
#20: SORN_Refractory(strength=1, factor='0.5;+-50%'),
21:
SORN_STDP(eta_stdp='[0.0001#5]', prune_stdp=False),
22:
SORN_SN(syn_type='GLU', clip_max=None, init_norm_factor=1.0),
23:
SORN_IP_TI(h_ip='lognormal_real_mean([0.04#6], [0.2944#7])',
eta_ip='[0.0006#8];+-50%',
integration_length='[30#18];+-50%',
clip_min=0.0),
25:
SORN_NOX(mp='self.partition_sum(n)',
eta_nox='[0.0005#9];+-50%'), #0.4
26:
SORN_SC_TI(h_sc='lognormal_real_mean([0.01#10], [0.2944#11])',
eta_sc='[0.1#12];+-50%',
integration_length='1'),
#27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]'),
30:
SORN_finish()
})
SynapseGroup(net=SORN,
src=e_ng,
dst=e_ng,
tag='GLU,e->e',
connectivity='(s_id!=d_id)*in_box(10)',
partition=True) #.partition([10, 10], [6, 6])
SynapseGroup(net=SORN,
src=e_ng,
dst=e_ng,
tag='GABA,e->e',
connectivity='(s_id!=d_id)*in_box(10)',
partition=True) #.partition([10, 10], [6, 6])
e_ng.add_behaviour(
9, SORN_external_input(strength=0.3, pattern_groups=[source]))
if timecale == 1:
e_ng.add_behaviour(101, Recorder(['n.pattern_index'],
tag='inp_rec'))
else:
#forward synapses
SynapseGroup(net=SORN,
src=last_e_ng,
dst=e_ng,
tag='GLU,e->e(+1)',
connectivity='in_box(10)',
partition=True) #.partition([10, 10], [6, 6])
SynapseGroup(net=SORN,
src=last_e_ng,
dst=e_ng,
tag='GABA,e->i(+1)',
connectivity='in_box(10)',
partition=True) #.partition([10, 10], [2, 2])
#backward synapses
SynapseGroup(net=SORN,
src=e_ng,
dst=last_e_ng,
tag='GLU,e(+1)->e',
connectivity='in_box(10)',
partition=True) #.partition([10, 10], [6, 6])
SynapseGroup(net=SORN,
src=e_ng,
dst=last_e_ng,
tag='GABA,e(+1)->i',
connectivity='in_box(10)',
partition=True) #.partition([10, 10], [2, 2])
last_e_ng = e_ng
e_ng.color = (0, 0, 255, 255)
SORN.set_marked_variables(ind, info=(ind == []))
SORN.initialize(info=False)
############################################################################################################################################################
score = 0
#SORN.simulate_iterations(10000, 100, measure_block_time=True, disable_recording=True)
Network_UI(SORN,
label='no inh neurons',
storage_manager=sm,
group_display_count=1).show()
score += train_and_generate_text(
SORN,
15000,
5000,
2000,
display=False,
stdp_off=True,
same_timestep_without_feedback_loop=True,
steps_recovery=15000) #, steps_recovery=15000
#score += get_oscillation_score_hierarchical(SORN, 0, 5000)
return score
def run(attrs={'name':'PV_SOM', 'ind':[], 'N_e':900, 'TS':[1]}):
so = True
print_info = attrs.get('print', True)
if print_info:
print(attrs)
sm = StorageManager(attrs['name'], random_nr=True, print_msg=print_info)
sm.save_param_dict(attrs)
#source = FDTGrammarActivator_New(tag='grammar_act', random_blocks=True, input_density=15/par['N_e'])#.plot_char_input_statistics()#output_size=par['N_e']#15
#print(len(source.alphabet))
source = LongDelayGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, mode=['simple'], input_density=0.015)#.print_test().plot_char_input_statistics()#10/par['N_e']
#source.plot_char_input_statistics()
#print(len(source.alphabet))
#source = Line_Patterns(tag='image_act', group_possibility=1.0, grid_width=30, grid_height=30, center_x=list(range(30)), center_y=30 / 2, degree=90, line_length=60)
#source = Line_Patterns(tag='image_act', group_possibility=1.0, grid_width=30, grid_height=30, center_x=list(range(30)), center_y=30 / 2, degree=90, line_length=60)
#print(source.get_text_score('. parrot likes trees. wolf wolf wolf..'))
SORN = Network()#[], [], initialize=False
last_PC = None
for layer, timescale in enumerate(attrs['TS']):
PC = NeuronGroup(net=SORN, tag='Pyramidal Cell {},prediction_source'.format(timescale), size=get_squared_dim(int(attrs['N_e'])), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='[50#0]%', distribution='lognormal(0,0.6)', normalize=True, partition_compensation=True), #lognormal(0,0.6)
4: SORN_init_afferent_synapses(transmitter='GABA_Dendrite', density='[30#1]%', distribution='uniform(0.1,0.11)', normalize=True),
5: SORN_init_afferent_synapses(transmitter='GABA_Soma', density='[30#2]%', distribution='uniform(0.1,0.11)', normalize=True),
6: SORN_init_afferent_synapses(transmitter='GABA_AIS', density='[30#3]%', distribution='uniform(0.1,0.11)', normalize=True),
#7: SORN_init_afferent_synapses(transmitter='GABA_NOX', density='full', distribution=None, normalize=True),
12: SORN_slow_syn(transmitter='GLU', strength='[0.1383#4]', so=so),
13: SORN_slow_syn(transmitter='GABA_Dendrite', strength='-0.1', so=False),
17: SORN_fast_syn(transmitter='GABA_Soma', strength='-0.1', so=False),
17.1: SORN_fast_syn(transmitter='GABA_AIS', strength='-0.1', so=False),
#17.2: SORN_fast_syn(transmitter='GABA_NOX', strength='-0.1', so=False),
18: SORN_generate_output(init_TH='0.1;+-100%'),
19: SORN_buffer_variables(),
20: SORN_Refractory_Digital(factor='[0.5#5];+-50%'),
#20: SORN_Refractory_Analog(factor='[0.5#5];+-50%'),
21: SORN_STDP(eta_stdp='[0.0015#6]'),#0.1#todo: test!!!, prune_stdp=False, excitation_punishment=0.0
22: SORN_SN(syn_type='GLU', clip_max=None, behaviour_norm_factor=1.0),
23: SORN_IP_TI(mp='n.output', h_ip='lognormal_real_mean([0.04#7], [0.2944#8])', eta_ip='[0.0006#9];+-50%', integration_length='[15#10];+-50%', clip_min=None),#mp='n.output_new/2.0+n.output_new_temp/2.0'
25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='0.5;+-50%', h_dh=0.0),#0.9#0.3
26: SORN_SC_TI(h_sc='lognormal_real_mean([0.015#11], [0.2944#12])', eta_sc='[0.1#13];+-50%', integration_length='1'),
#27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]'),
})
MT_SOM = NeuronGroup(net=SORN, tag='Martinotti Cell {},Somatostatin'.format(timescale), size=get_squared_dim(int(0.07 * attrs['N_e'])), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='50%', distribution='lognormal(0,0.87038)', normalize=True),
#4: SORN_init_afferent_synapses(transmitter='GABA', density='20%', distribution='lognormal(0,[0.82099#15])', normalize=True), # 40
10: SORN_slow_syn(transmitter='GLU', strength='0.6', so=so), # 1.5353
#11: SORN_slow_syn(transmitter='GABA', strength='-[0.1838#16]', so=so),
#15: SORN_fast_syn(transmitter='GABA', strength='-[0.08#17]', so=False),#0.08
18: SORN_generate_output(init_TH='0.1;+-0%'),
19: SORN_buffer_variables(),
#20: SORN_Refractory(factor='0.2;0.7'),
})
'''
EXP_NOX_CELL = NeuronGroup(net=SORN, tag='NOX Cell {}'.format(timescale), size=get_squared_dim(int(16)), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='full', distribution=None, normalize=True),
#4: SORN_init_afferent_synapses(transmitter='GABA', density='20%', distribution='lognormal(0,[0.82099#15])', normalize=True), # 40
10: SORN_slow_syn(transmitter='GLU', strength='9.0', so=so),#1.5353
#11: SORN_slow_syn(transmitter='GABA', strength='-[0.1838#16]', so=so),
#15: SORN_fast_syn(transmitter='GABA', strength='-[0.08#17]', so=False),#0.08
18: SORN_generate_output(init_TH='0.0', digital_output=False),
19: SORN_buffer_variables(),
#20: SORN_Refractory(factor='0.2;0.7'),
})
'''
BA_PV = NeuronGroup(net=SORN, tag='Basket Cell {},Parvalbumin'.format(timescale), size=get_squared_dim(int(0.07 * attrs['N_e'])), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='50%', distribution='lognormal(0,0.87038)', normalize=True),
#4: SORN_init_afferent_synapses(transmitter='GABA', density='20%', distribution='lognormal(0,[0.82099#15])', normalize=True), # 40
10: SORN_slow_syn(transmitter='GLU', strength='0.6', so=so), # 1.5353
#11: SORN_slow_syn(transmitter='GABA', strength='-[0.1838#16]', so=so),
#15: SORN_fast_syn(transmitter='GABA', strength='-[0.08#17]', so=False),#0.08
18: SORN_generate_output(init_TH='0.1;+-0%'),
19: SORN_buffer_variables(),
#20: SORN_Refractory(factor='0.2;0.7'),
})
CH_PV = NeuronGroup(net=SORN, tag='Chandelier Cell {},Parvalbumin'.format(timescale), size=get_squared_dim(int(0.07 * attrs['N_e'])), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='50%', distribution='lognormal(0,0.87038)', normalize=True),
#4: SORN_init_afferent_synapses(transmitter='GABA', density='20%', distribution='lognormal(0,[0.82099#15])', normalize=True), # 40
#11: SORN_slow_syn(transmitter='GABA', strength='-[0.1838#16]', so=so),
14: SORN_fast_syn(transmitter='GLU', strength='0.2', so=so),#1.5353
#15: SORN_fast_syn(transmitter='GABA', strength='-[0.08#17]', so=False),#0.08
18: SORN_generate_output(init_TH='0.1;+-0%'),
19: SORN_buffer_variables(),
#19: SORN_Refractory(factor='0.2;0.7'),
})
MT_SOM['structure', 0].stretch_to_equal_size(PC)
BA_PV['structure', 0].stretch_to_equal_size(PC)
CH_PV['structure', 0].stretch_to_equal_size(PC)
#EXP_NOX_CELL['structure', 0].stretch_to_equal_size(PC)
#plt.scatter(PC.x, PC.y)
#plt.scatter(MT_SOM.x, MT_SOM.y)
#plt.scatter(BA_PV.x, BA_PV.y)
#plt.scatter(CH_PV.x, CH_PV.y)
#plt.scatter(EXP_NOX_CELL.x, EXP_NOX_CELL.y)
#plt.show()
#print(np.min(EXP_NOX_CELL.x), np.max(EXP_NOX_CELL.x))
#print(np.min(EXP_NOX_CELL.y), np.max(EXP_NOX_CELL.y))
#print(np.min(PC.x), np.max(PC.x))
#print(np.min(PC.y), np.max(PC.y))
SynapseGroup(net=SORN, src=PC, dst=PC, tag='GLU', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
SynapseGroup(net=SORN, src=PC, dst=MT_SOM, tag='GLU', connectivity='in_box(2)', partition=True)
SynapseGroup(net=SORN, src=PC, dst=BA_PV, tag='GLU', connectivity='in_box(2)', partition=True)
SynapseGroup(net=SORN, src=PC, dst=CH_PV, tag='GLU', connectivity='in_box(2)', partition=True)
#SynapseGroup(net=SORN, src=PC, dst=EXP_NOX_CELL, tag='GLU', connectivity='in_box(3.75)', partition=True)
SynapseGroup(net=SORN, src=MT_SOM, dst=PC, tag='GABA_Dendrite', connectivity='in_box(10)', partition=True)
SynapseGroup(net=SORN, src=BA_PV, dst=PC, tag='GABA_Soma', connectivity='in_box(10)', partition=True)
SynapseGroup(net=SORN, src=CH_PV, dst=PC, tag='GABA_AIS', connectivity='in_box(10)', partition=True)
#SynapseGroup(net=SORN, src=EXP_NOX_CELL, dst=PC, tag='GABA_NOX', connectivity='in_box(3.75)', partition=True)
#SynapseGroup(net=SORN, src=SOM, dst=SOM, tag='GABA_D,SOM->SOM', connectivity='(s_id!=d_id)*(np.abs(sx-dx)<=10)*(np.abs(sy-dy)<=10)', partition=True)
#SynapseGroup(net=SORN, src=SOM, dst=PV, tag='GABA_D,SOM->PV', connectivity='(s_id!=d_id)*(np.abs(sx-dx)<=10)*(np.abs(sy-dy)<=10)', partition=True)
##SynapseGroup(net=SORN, src=PV, dst=SOM, tag='GABA_P,PV->SOM')#not realistic?
#SynapseGroup(net=SORN, src=PV, dst=PV, tag='GABA_P,PV->PV', connectivity='(s_id!=d_id)*(np.abs(sx-dx)<=10)*(np.abs(sy-dy)<=10)', partition=True)
if layer==0:
PC.add_behaviour(9, SORN_external_input(strength=1.0, pattern_groups=[source]))
#MT_SOM.add_behaviour(9, SORN_external_input(strength=1.0, pattern_groups=[source]))
#BA_PV.add_behaviour(9, SORN_external_input(strength=1.0, pattern_groups=[source]))
#CH_PV.add_behaviour(9, SORN_external_input(strength=1.0, pattern_groups=[source]))
#else:
# #forward synapses
# SynapseGroup(net=SORN, src=last_PC, dst=PC, tag='GLU,PC->PC(+1)', connectivity='(s_id!=d_id)*(np.abs(sx-dx)<=10)*(np.abs(sy-dy)<=10)', partition=True)#.partition([10, 10], [partition, partition])
# SynapseGroup(net=SORN, src=last_PC, dst=SOM, tag='GABA,PC->SOM(+1)', connectivity='(s_id!=d_id)*(np.abs(sx-dx)<=10)*(np.abs(sy-dy)<=10)', partition=True)#.partition([5, 5], [2, 2])
# #backward synapses
# SynapseGroup(net=SORN, src=PC, dst=last_PC, tag='GLU,PC(+1)->PC', connectivity='(s_id!=d_id)*(np.abs(sx-dx)<=10)*(np.abs(sy-dy)<=10)', partition=True)#.partition([10, 10], [partition, partition])
# SynapseGroup(net=SORN, src=PC, dst=last_SOM, tag='GABA,PC(+1)->SOM', connectivity='(s_id!=d_id)*(np.abs(sx-dx)<=10)*(np.abs(sy-dy)<=10)', partition=True)#.partition([5, 5], [2, 2])
last_PC = PC
#last_MT_SOM = MT_SOM
#last_BA_PV = BA_PV
#last_CH_PV = CH_PV
#last_EXP_NOX_CELL = EXP_NOX_CELL
if __name__ == '__main__':
PC.color = get_color(0, timescale)
MT_SOM.color = get_color(1, timescale)
BA_PV.color = get_color(2, timescale)
CH_PV.color = get_color(3, timescale)
#EXP_NOX_CELL.color = get_color(4, timescale)
#EXP_NOX_CELL.display_min_max_act = (0, 0.3)
SORN.set_marked_variables(attrs['ind'], info=print_info, storage_manager=sm)
SORN.initialize(info=False)
###################################################################################################################
#if __name__ == '__main__':
# Network_UI(SORN, label='SORN UI PC PV SOM', storage_manager=sm, group_display_count=4).show()#'GLU', 'GABA_Dendrite', 'GABA_Soma', 'GABA_AIS'
score = 0
score += train_and_generate_text(SORN, 20000, 5000, 3000, display=print_info, stdp_off=True, same_timestep_without_feedback_loop=True, steps_recovery=1000, storage_manager=sm)#, steps_recovery=15000
#score += get_oscillation_score_hierarchical(SORN, 0, 5000)
return score
def run(
attrs={
'name': 'maze',
'ind': [],
'N_e': 900,
'TS': [1],
'ff': True,
'fb': True,
'plastic': 15000
}):
so = True
print_info = attrs.get('print', True)
if print_info:
print(attrs)
sm = StorageManager(attrs, random_nr=True, print_msg=print_info)
sm.save_param_dict(attrs)
#source = LongDelayGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, mode=['simple'], input_density=0.01)
maze = Maze(level='default')
SORN = Network() #behaviour={maze.get_network_behaviour()}
SORN.maze = maze
for layer, timescale in enumerate(attrs['TS']):
e_location = NeuronGroup(
net=SORN,
tag='E_location_{}'.format(timescale),
size=maze.get_location_neuron_dimension(),
behaviour={
2:
SORN_init_neuron_vars(timescale=timescale),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='full',
distribution='uniform(0.1,0.11)',
normalize=True,
partition_compensation=True
), # 0.89 uniform(0.1,0.11)uniform(0,[0.95#0])
4:
SORN_init_afferent_synapses(
transmitter='GABA',
density='full',
distribution='uniform(0.1,0.11)',
normalize=True
), # 0.80222 uniform(0.1,0.11) lognormal(0,[0.4#1])
9:
maze.get_location_neuron_behaviour(), ###############
12:
SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
13:
SORN_slow_syn(transmitter='GABA',
strength='-[0.1698#3]',
so=False),
17:
SORN_fast_syn(transmitter='GABA',
strength='-[0.1#4]',
so=False), # 0.11045
18:
SORN_generate_output(init_TH='0.1;+-100%'),
19:
SORN_buffer_variables(),
20:
SORN_Refractory_Analog(factor='0.5;+-50%'),
21:
SORN_STDP(eta_stdp='[0.0015#5]',
weight_attr='W_temp',
STDP_F={
-5: 0.2,
-4: 0.4,
-3: 0.6,
-2: 0.8,
-1: 1
}),
22:
SORN_temporal_synapses(syn_type='GLU',
behaviour_norm_factor=1.0),
#22: SORN_SN(syn_type='GLU', clip_max=None, behaviour_norm_factor=1.0),
23:
SORN_IP_TI(h_ip='lognormal_real_mean([0.004#6], [0.2944#7])',
eta_ip='[0.0006#8];+-50%',
integration_length='[15#18];+-50%',
clip_min=None),
25:
SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.5#9];+-50%'),
26:
SORN_SC_TI(h_sc='lognormal_real_mean([0.015#10], [0.2944#11])',
eta_sc='[0.1#12];+-50%',
integration_length='1'), # 60;+-50% #0.05
27:
SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]'),
#29: SORN_dopamine(),
})
i_location = NeuronGroup(
net=SORN,
tag='I_location{}'.format(timescale),
size=maze.get_inhibitory_location_neuron_dimension(),
behaviour={
2:
SORN_init_neuron_vars(timescale=timescale),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='50%',
distribution='uniform(0.1,0.11)',
normalize=True), #lognormal(0,[0.87038#14])
4:
SORN_init_afferent_synapses(
transmitter='GABA',
density='20%',
distribution='uniform(0.1,0.11)',
normalize=True), #lognormal(0,[0.82099#15])
11:
SORN_slow_syn(transmitter='GABA',
strength='-[0.1838#16]',
so=so),
14:
SORN_fast_syn(transmitter='GLU', strength='[1.5#16]',
so=so), # 1.5353
15:
SORN_fast_syn(transmitter='GABA',
strength='-[0.08#17]',
so=False), # 0.08
18:
SORN_generate_output(init_TH='0.1;+-0%'),
19:
SORN_buffer_variables(),
20:
SORN_Refractory_Analog(factor='0.2;0.7'),
})
'''
e_vision = NeuronGroup(net=SORN, tag='E_vision_{}'.format(timescale), size=maze.get_vision_neuron_dimension(), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='full', distribution='lognormal(0,[0.95#0])', normalize=True, partition_compensation=True), # 0.89 uniform(0.1,0.11)
4: SORN_init_afferent_synapses(transmitter='GABA', density='45%', distribution='lognormal(0,[0.4#1])', normalize=True), # 0.80222 uniform(0.1,0.11)
9: maze.get_vision_neuron_behaviour(),
12: SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
13: SORN_slow_syn(transmitter='GABA', strength='-[0.1698#3]', so=False),
17: SORN_fast_syn(transmitter='GABA', strength='-[0.1#4]', so=False), # 0.11045
18: SORN_generate_output(init_TH='0.01'),
19: SORN_buffer_variables(),
#20: SORN_Refractory(factor='0.5;+-50%'),
#21: SORN_STDP(eta_stdp='[0.00015#5]'),
#22: SORN_SN(syn_type='GLU', clip_max=None, behaviour_norm_factor=5.0),
#23: SORN_IP_TI(h_ip='0.04', eta_ip='0.001', integration_length='[15#18];+-50%', clip_min=None),
#25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.5#9];+-50%'),
#26: SORN_SC_TI(h_sc='lognormal_real_mean([0.015#10], [0.2944#11])', eta_sc='[0.1#12];+-50%',integration_length='1'), # 60;+-50% #0.05
#27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]')
})
'''
e_action = NeuronGroup(
net=SORN,
tag='E_action_{}'.format(timescale),
size=maze.get_action_neuron_dimension(),
behaviour={
2:
SORN_init_neuron_vars(timescale=timescale),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='full',
distribution='uniform(0.1,0.11)',
normalize=True,
partition_compensation=True), # 0.89 lognormal(0,[0.5#0])
12:
SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
13:
SORN_slow_syn(transmitter='GABA',
strength='-[0.1698#3]',
so=False),
17:
SORN_fast_syn(transmitter='GABA',
strength='-[0.1#4]',
so=False), # 0.11045
18:
SORN_generate_output(init_TH='0.01'),
19:
SORN_buffer_variables(),
20:
SORN_Refractory_Analog(factor='0.7;+-50%'),
21:
SORN_STDP(eta_stdp='[0.0015#5]',
STDP_F={
-1: 1,
1: 0
},
weight_attr='W_temp'), #bigger #todo!!!!
#22: SORN_SN(syn_type='GLU', clip_max=None, behaviour_norm_factor=5.0),
22:
SORN_temporal_synapses(syn_type='GLU',
behaviour_norm_factor=5.0),
23:
SORN_IP_TI(h_ip='0.04',
eta_ip='0.001',
integration_length='[15#18];+-50%',
clip_min=None),
#25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.5#9];+-50%'),
#26: SORN_SC_TI(h_sc='lognormal_real_mean([0.015#10], [0.2944#11])', eta_sc='[0.1#12];+-50%',integration_length='1'), # 60;+-50% #0.05
#27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]')
29:
SORN_dopamine(),
30:
maze.get_action_neuron_behaviour()
})
'''
e_reward = NeuronGroup(net=SORN, tag='E_reward_{}'.format(timescale), size=maze.get_reward_neuron_dimension(), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='full', distribution='lognormal(0,[0.95#0])', normalize=True, partition_compensation=True),
9: maze.get_reward_neuron_behaviour(),
12: SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
18: SORN_generate_output(init_TH='0.01'),
19: SORN_buffer_variables(),
#20: SORN_Refractory(factor='0.5;+-50%'),
21: SORN_STDP(eta_stdp='[0.00015#5]'),
22: SORN_SN(syn_type='GLU', clip_max=None, behaviour_norm_factor=0.1),
#23: SORN_IP_TI(h_ip='0.04', eta_ip='0.001', integration_length='[15#18];+-50%', clip_min=None),
#25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.5#9];+-50%'),
#26: SORN_SC_TI(h_sc='lognormal_real_mean([0.015#10], [0.2944#11])', eta_sc='[0.1#12];+-50%',integration_length='1'), # 60;+-50% #0.05
#27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]')
})
'''
e_punishment = NeuronGroup(
net=SORN,
tag='E_punishment_{}'.format(timescale),
size=maze.get_punishment_neuron_dimension(),
behaviour={
2:
SORN_init_neuron_vars(timescale=timescale),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='full',
distribution='lognormal(0,[0.95#0])',
normalize=True,
partition_compensation=True),
9:
maze.get_punishment_neuron_behaviour(),
12:
SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
18:
SORN_generate_output(init_TH='0.01'),
19:
SORN_buffer_variables(),
#20: SORN_Refractory(factor='0.5;+-50%'),
21:
SORN_STDP(eta_stdp='[0.00015#5]'), #todo!!!!
22:
SORN_SN(syn_type='GLU',
clip_max=None,
behaviour_norm_factor=0.1),
#23: SORN_IP_TI(h_ip='0.04', eta_ip='0.001', integration_length='[15#18];+-50%', clip_min=None),
#25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.5#9];+-50%'),
#26: SORN_SC_TI(h_sc='lognormal_real_mean([0.015#10], [0.2944#11])', eta_sc='[0.1#12];+-50%',integration_length='1'), # 60;+-50% #0.05
#27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]')
})
i_location['structure', 0].stretch_to_equal_size(e_location)
SynapseGroup(net=SORN,
src=e_location,
dst=e_location,
tag='GLU,GLU_same',
connectivity='(s_id!=d_id)*in_box(10)',
partition=True)
SynapseGroup(net=SORN,
src=e_location,
dst=i_location,
tag='GLU',
connectivity='in_box(10)',
partition=True)
SynapseGroup(net=SORN,
src=i_location,
dst=e_location,
tag='GABA',
connectivity='in_box(10)',
partition=True)
SynapseGroup(net=SORN,
src=i_location,
dst=i_location,
tag='GABA',
connectivity='(s_id!=d_id)*in_box(10)',
partition=True)
#SynapseGroup(net=SORN, src=e_vision, dst=e_location, tag='GLU', partition=True)
SynapseGroup(net=SORN,
src=e_location,
dst=e_action,
tag='GLU',
partition=True)
#SynapseGroup(net=SORN, src=e_location, dst=e_reward, tag='GLU', partition=True)
#SynapseGroup(net=SORN, src=e_reward, dst=e_location, tag='DOP+', partition=True)
#SynapseGroup(net=SORN, src=e_reward, dst=e_action, tag='DOP+', partition=True)
SynapseGroup(net=SORN,
src=e_location,
dst=e_punishment,
tag='GLU',
partition=True)
SynapseGroup(net=SORN,
src=e_punishment,
dst=e_location,
tag='DOP-',
partition=True)
SynapseGroup(net=SORN,
src=e_punishment,
dst=e_action,
tag='DOP-',
partition=True)
if __name__ == '__main__' and attrs.get('UI', False):
e_location.color = (0, 0, 255, 255)
i_location.color = (255, 0, 0, 255)
e_action.color = (255, 255, 0, 255)
#e_vision.color = (0, 255, 255, 255)
#e_reward.color = (100, 255, 100, 255)
e_punishment.color = (255, 100, 100, 255)
SORN.set_marked_variables(attrs['ind'],
info=print_info,
storage_manager=sm)
SORN.initialize(info=False)
###################################################################################################################
if __name__ == '__main__' and attrs.get('UI', False):
#default_modules.insert(0, maze_tab())
#Network_UI(SORN, label='SORN UI default setup', storage_manager=sm, group_display_count=4, reduced_layout=True).show()
Network_UI(SORN,
modules=get_default_UI_modules() + [maze_tab()],
label='SORN UI K_WTA',
storage_manager=sm,
group_display_count=1,
reduced_layout=False).show()
return 0
print('iteration2')
return
class my_beh3(Behaviour):
def set_variables(self, syn):
print('set3')
return
def new_iteration(self, syn):
print('iteration3')
return
net = Network(behaviour={})#6: my_beh1(tag='tag1')
ng = NeuronGroup(net=net, size=100, behaviour={
1: ClockModule(step='10*ms'),
2: Variable(eq='v=n.get_random_neuron_vec()*0.1*mV'),
3: Variable(eq='tau=10*ms+n.get_random_neuron_vec()*200*ms'),
5: EquationModule(eq='dv/dt=(0*mV-v)/tau'),
7: neuron_event(condition='n.v<0.01', eq='n.v_new=1'),
#7: my_beh2(tag='tag2'),
100: Recorder(['n.v', 'n.t'], tag='my_rec')
})
syn = SynapseGroup(net=net, src=ng, dst=ng, behaviour={
def run(attrs={'name': 'KWTA', 'ind': [], 'N_e': 900, 'TS': [1], 'ff': True, 'fb': True, 'plastic': 15000}):
so = True
print_info = attrs.get('print', True)
if print_info:
print(attrs)
sm = StorageManager(attrs['name'], random_nr=True, print_msg=print_info)
sm.save_param_dict(attrs)
source = FewSentencesGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, input_density=0.015)
#source = SingleWordGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, input_density=0.015)
#source = FewLongSentencesGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, input_density=0.015)
#source.plot_char_frequency_histogram(20)
SORN = Network()
for layer, timescale in enumerate(attrs['TS']):
e_ng = NeuronGroup(net=SORN, tag='PC_{},prediction_source'.format(timescale), size=get_squared_dim(int(attrs['N_e'] / timescale)), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='[90#0]%', distribution='lognormal(0,0.95)', normalize=True),#20%#lognormal(0,[0.95#1]) #[13#0]% #, partition_compensation=True , partition_compensation=True
4: SORN_init_afferent_synapses(transmitter='GABA', density='[30#1]%', distribution='uniform(0.0,1.0)', normalize=True),
12: SORN_slow_syn(transmitter='GLU', strength='1.0', so=so),
12.1: SORN_WTA_iSTDP(eta_iSTDP='[0.00015#2]'),
12.2: SORN_SN(syn_type='GABA'),
13.0: SORN_IP_WTA(h_ip='lognormal_real_mean(0.04, 0.2944)', eta_ip='[0.006#3]', clip_min=None),
13.4: SORN_generate_output_K_WTA_partitioned(K='[0.12#4]'),
13.5: SORN_WTA_fast_syn(transmitter='GABA', strength='-[0.5#5]', so=so),#[0.1383#2]SORN_fast_syn
18: SORN_generate_output_K_WTA_partitioned(K='[0.04#6]', filter_temporal_output=True),
19: SORN_buffer_variables(random_temporal_output_shift=False),
21: SORN_STDP(eta_stdp='[0.00015#7]'),
22: SORN_SN(syn_type='GLU')
})
#smaller number of active neuons for more competition?
# SynapseGroup(net=SORN, src=retina, dst=e_ng, tag='GLU,GLU_same', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='GLU,GLU_same', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='GABA,GABA_same', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
#if layer > 0:
# if attrs.get('ff', True): # forward synapses
# SynapseGroup(net=SORN, src=last_e_ng, dst=e_ng, tag='GLU,GLU_ff', connectivity='in_box(10)', partition=False)
# if attrs.get('fb', False): # backward synapses
# SynapseGroup(net=SORN, src=e_ng, dst=last_e_ng, tag='GLU,GLU_fb', connectivity='in_box(10)', partition=False)
e_ng.add_behaviour(9, SORN_external_input(strength=1.0, pattern_groups=[source]))
#last_e_ng = e_ng
if __name__ == '__main__' and attrs.get('UI', False):
e_ng.color = get_color(0, timescale)
SORN.set_marked_variables(attrs['ind'], info=print_info, storage_manager=sm)
SORN.initialize(info=False)
#for s in e_ng.afferent_synapses['GLU']:
# s.W = s.W-np.mean(s.W)
###################################################################################################################
if __name__ == '__main__' and attrs.get('UI', False):
Network_UI(SORN, label='SORN UI K_WTA', storage_manager=sm, group_display_count=1, reduced_layout=False).show()
score = 0
plastic_steps = attrs.get('plastic', 20000)
for i in range(1):
sm = StorageManager(attrs['name'] + '[{:03d}]'.format(i + 1), random_nr=True, print_msg=print_info)
sm.save_param_dict(attrs)
score += train_and_generate_text(SORN, plastic_steps, 5000, 1000, display=print_info, stdp_off=True, same_timestep_without_feedback_loop=True, steps_recovery=1000, storage_manager=sm)
print('score=', score)
return score
def run(attrs={'name': 'KWTA', 'ind': [], 'N_e': 900, 'plastic': 15000}):
print_info = attrs.get('print', True)
if print_info:
print(attrs)
sm = StorageManager(attrs['name'], random_nr=True, print_msg=print_info)
sm.save_param_dict(attrs)
source = FewSentencesGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, input_density=0.015, frequency_adjustment=True)#21
SORN = Network()
e_ng = NeuronGroup(net=SORN, tag='PC_{},prediction_source'.format(1), size=get_squared_dim(attrs['N_e']), behaviour={
2: SORN_init_neuron_varsTF(timescale=1),
3: SORN_init_afferent_synapsesTF(transmitter='GLU', density='90%', distribution='uniform(0.1,1.0)', normalize=True),#20%#lognormal(0,[0.95#1]) #[13#0]% #, partition_compensation=True , partition_compensation=True #lognormal(0,0.95)
#4: SORN_init_afferent_synapses(transmitter='GABA', density='[30#1]%', distribution='uniform(0.0,1.0)', normalize=True),
5: SORN_init_afferent_synapsesTF(transmitter='GLU_cluster', density='90%', distribution='uniform(0.1,1.0)', normalize=True),
10.0: SORN_slow_syn_simpleTF(transmitter='GLU', strength='1.0'),
10.1: SORN_IP_WTA_apply_TF(),
10.15: WTA_refrac_applyTF(strengthfactor='[0.1#0]'),#0.1 #attrs['refrac']
10.2: SORN_generate_output_K_WTA_partitionedTF_Experimental(partition_size=7, K='[0.02#1]'),
10.3: SORN_slow_syn_simpleTF(transmitter='GLU_cluster', strength='1.0'),
10.4: SORN_generate_output_K_WTA_partitionedTF_Experimental(partition_size=7, K='[0.02#1]', filter_temporal_output=False),
10.5: SORN_slow_syn_simpleTF(transmitter='GLU_cluster', strength='1.0'),
10.6: SORN_generate_output_K_WTA_partitionedTF_Experimental(partition_size=7, K='[0.02#1]', filter_temporal_output=False),
10.7: SORN_slow_syn_simpleTF(transmitter='GLU_cluster', strength='1.0'),
10.8: SORN_generate_output_K_WTA_partitionedTF_Experimental(partition_size=7, K='[0.02#1]', filter_temporal_output=False),
#15: SORN_buffer_variables(random_temporal_output_shift=False),
18: WTA_refracTF(decayfactor=0.5),
20: SORN_IP_WTA_TF(h_ip='lognormal_real_mean([0.02#1], [0.2944#2])', eta_ip='[0.007#3]', target_clip_min=None, target_clip_max=None), #-1.0 #1.0 #0.007
21.1: STDP_TF(syn_type='GLU', eta_stdp='[0.00015#4]'),#, 0: 1 #[0.00015#7] #0.5, 0: 3.0
21.2: STDP_TF_simu(syn_type='GLU_cluster', eta_stdp='[0.00015#5]'), #[0.00015#7]
22: SORN_SN_TF(syn_type='GLU', norm_factor=1.0),
23: SORN_SN_TF(syn_type='GLU_cluster', norm_factor='[0.3#6]')#0.1
})
# SynapseGroup(net=SORN, src=retina, dst=e_ng, tag='GLU,GLU_same', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
receptive_field = 18
#receptive_field = int(18*math.sqrt(attrs['N_e']/1400))
print(receptive_field)
SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='GLU,syn', connectivity='(s_id!=d_id)*in_box({})'.format(receptive_field))#, partition=True)#14
SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='GLU_cluster,syn', connectivity='(s_id!=d_id)*in_box({})'.format(receptive_field))#, partition=True)
#SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='GABA,GABA_same', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
e_ng.add_behaviour(9, SORN_external_inputTF(strength=1.0, pattern_groups=[source]))
if __name__ == '__main__' and attrs.get('UI', False):
e_ng.color = get_color(0, 1)
SORN.set_marked_variables(attrs['ind'], info=print_info, storage_manager=sm)
SORN.initialize(info=False)
###################################################################################################################
if __name__ == '__main__' and attrs.get('UI', False):
my_modules = get_default_UI_modules(['output', 'exhaustion_value', 'weight_norm_factor'])+get_my_default_UI_modules()
my_modules[0] = UI_sidebar_activity_module(1, add_color_dict={'output': (255, 255, 255), 'Input_Mask': (-100, -100, -100)})
#my_modules[1] = multi_group_plot_tab(['output', 'exhaustion_value', 'weight_norm_factor']) # , 'nox', 'refractory_counter'
my_modules[8] = single_group_plot_tab({'activity': (0, 0, 0), 'excitation': (0, 0, 255), 'inhibition': (255, 0, 0), 'input_act': (255, 0, 255),'exhaustion_value': (0, 255, 0)})
Network_UI(SORN, modules=my_modules, label='SORN UI K_WTA', storage_manager=sm, group_display_count=1, reduced_layout=False).show()
'''
if __name__ == '__main__' and attrs.get('UI', False):
my_modules = [
UI_sidebar_activity_module(add_color_dict={'output': (255, 255, 255), 'Input_Mask': (-100, -100, -100)}),
weight_tab(weight_attrs=['W']),
sidebar_fast_forward_module(),
sidebar_save_load_module(),
multi_group_plot_tab(['output'], tensorflow=True),
# fourier_tab(parameter='voltage'),
info_tab(),
]
Network_UI(SORN, modules=my_modules, label='Tensorflow Test', storage_manager=None, group_display_count=1).show()
#my_modules = get_default_UI_modules()
#my_modules[1] = multi_group_plot_tab(['output', 'exhaustion_value', 'weight_norm_factor']) # , 'nox', 'refractory_counter'
#my_modules[18] = single_group_plot_tab({'activity': (0, 0, 0), 'excitation': (0, 0, 255), 'inhibition': (255, 0, 0), 'input_act': (255, 0, 255),'exhaustion_value': (0, 255, 0)})
#Network_UI(SORN, modules=my_modules, label='SORN UI K_WTA', storage_manager=sm, group_display_count=1, reduced_layout=False).show()
'''
score = 0
plastic_steps = attrs.get('plastic', 20000)
#for i in range(1):
# #sm = StorageManager(attrs['name'] + '[{:03d}]'.format(i + 1), random_nr=True, print_msg=print_info)
# #sm.save_param_dict(attrs)
# sm=None
# score += train_and_generate_text(SORN, plastic_steps, 5000, 1000, display=print_info, stdp_off=True, same_timestep_without_feedback_loop=True, steps_recovery=1000, storage_manager=sm, return_key='right_sentences_square_score')
score = 0
#print(train_and_generate_text(SORN, plastic_steps, 5000, 1000, display=print_info, stdp_off=True, same_timestep_without_feedback_loop=True, steps_recovery=5000, storage_manager=sm, return_key='right_sentences_square_score'))
#print('Output:')
#print(predict_text_max_source_act(SORN, plastic_steps, 5000, 5000, display=True, stdp_off=True))#plastic_steps, 5000, 1000
#print('score=', score)
#print(SORN.simulate_iteration(True))
return get_max_text_score(SORN, plastic_steps, 5000, 5000, display=True, stdp_off=True, return_key='right_word_square_score')
def run(tag, ind=[], par={'N_e':[1800], 'TS':[1]}):
name = 'DrumBeats_'+'_'+tag+str(par['N_e'])+'N_e_'+str(par['TS'])
sm = StorageManager(main_folder_name=tag, folder_name=name, random_nr=False, print_msg=display)
sm.save_param_dict(par)
source = DrumBeatActivator(tag='drum_act', which_tracks=[100], filter_silence=False, THR_similar_tracks=0.5, input_density=0.015, offtoken=True, ontoken=True, include_inverse_alphabet=False)#, include_inverse_alphabet= True)#output_size=par['N_e']
inh_thr = ['0.1;+-0%','0.1;+-0%', '0.1;+-0%']
exc_thr = ['0.1;+-100%', '0.1;+-100%', '0.1;+-100%']
SORN = Network()
for i, lag in enumerate(par['TS']):#
e_ng = NeuronGroup(net=SORN, tag='exc_cell_{},prediction_source,text_input_group'.format(lag), size=get_squared_dim(int(par['N_e'][i])), behaviour={
2: SORN_init_neuron_vars(iteration_lag=lag, init_TH=exc_thr[i]),
3: SORN_init_afferent_synapses(transmitter='GLU', density='13%', distribution='lognormal(0,[0.89#0])', normalize=True, partition_compensation=True),
4: SORN_init_afferent_synapses(transmitter='GABA', density='50%', distribution='lognormal(0,[0.80222#1])', normalize=True),
12: SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
13: SORN_slow_syn(transmitter='GABA', strength='-[0.1698#3]', so=so),
17: SORN_fast_syn(transmitter='GABA', strength='-[0.11045#4]', so=so),
18: SORN_input_collect(),
19: SORN_Refractory(factor='0.5;+-50%'),
21: SORN_STDP(eta_stdp='[0.0001#5]', prune_stdp=False),
22: SORN_SN(syn_type='GLU', clip_max=None, init_norm_factor=1.0),
23: SORN_IP_TI(h_ip='lognormal_real_mean([0.04#6], [0.2944#7])', eta_ip='[0.0006#8];+-50%', integration_length='[30#18];+-50%', clip_min=None),#, gap_percent=10 #30;+-50% #0.0003
#24: SORN_NOX(mp='np.mean(n.output_new)', eta_nox='[0.3#9];+-50%', behaviour_enabled=False), #0.4
25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.3#9];+-50%'), # 0.4
26: SORN_SC_TI(h_sc='lognormal_real_mean([0.01#10], [0.2944#11])', eta_sc='[0.1#12];+-50%', integration_length='1'), #60;+-50% #0.05
27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]'),
30: SORN_finish(),
#99: SynapseRecorder(['[np.sum(s.slow_add)]'], tag='ex_glu_syn_rec'),
#100: Recorder(['n.output'], tag='exc_out_rec')
})
i_ng = NeuronGroup(net=SORN, tag='inh_cell_{}'.format(lag), size=get_squared_dim(int(0.2 * (par['N_e'][i]))), behaviour={
2: SORN_init_neuron_vars(iteration_lag=lag, init_TH=inh_thr[i]),
3: SORN_init_afferent_synapses(transmitter='GLU', density='45%', distribution='lognormal(0,[0.87038#14])', normalize=True), # 450
4: SORN_init_afferent_synapses(transmitter='GABA', density='20%', distribution='lognormal(0,[0.82099#15])', normalize=True), # 40
#11: SORN_slow_syn(transmitter='GABA', strength='-[0.1838#16]', so=so),
14: SORN_fast_syn(transmitter='GLU', strength='[1.5#16]', so=so),#1.5353
15: SORN_fast_syn(transmitter='GABA', strength='-[0.08#17]', so=so),#0.08
18: SORN_input_collect(),
19: SORN_Refractory(factor='0.1;0.4'),
#23: SORN_IP_TI(h_ip='lognormal_real_mean([0.08#6], [0.2944#7])', eta_ip='[0.0003#8];+-50%', integration_length='30;+-50%', clip_min=None),
30: SORN_finish(),
#100: Recorder(['n.output'], tag='inh_out_rec')
})
i_ng['structure', 0].stretch_to_equal_size(e_ng)
SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='GLU,ee', connectivity='(s_id!=d_id)*in_box(10)', partition=True)#.partition([10, 10], [4, 4])
SynapseGroup(net=SORN, src=e_ng, dst=i_ng, tag='GLU,ie', connectivity='(s_id!=d_id)*in_box(10)', partition=True)#.partition([5, 5], [2, 2])
SynapseGroup(net=SORN, src=i_ng, dst=e_ng, tag='GABA,ei', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
SynapseGroup(net=SORN, src=i_ng, dst=i_ng, tag='GABA,ii', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
if lag == 1:
i_ng.add_behaviour(10, SORN_external_input(strength=1.0, pattern_groups=[source]))
e_ng.add_behaviour(10, SORN_external_input(strength=1.0, pattern_groups=[source]))
#e_ng.add_behaviour(101, Recorder(['n.pattern_index'], tag='inp_rec'))
else:
#forward synapses
SynapseGroup(net=SORN, src=last_e_ng, dst=e_ng, tag='GLU,eeff')#.partition([10, 10], [4, 4])
SynapseGroup(net=SORN, src=last_e_ng, dst=i_ng, tag='GABA,ieff')#.partition([5, 5], [2, 2])
#backward synapses
SynapseGroup(net=SORN, src=e_ng, dst=last_e_ng, tag='GLU,eebw')#.partition([10, 10], [4, 4])
SynapseGroup(net=SORN, src=e_ng, dst=last_i_ng, tag='GABA,iebw')#.partition([5, 5], [2, 2])
last_e_ng = e_ng # excitatory neuron group
last_i_ng = i_ng # inhibitory neuron group
e_ng.color = (0,0,255,255)
i_ng.color = (255,0,255,255)
e_ng.pattern_index=0
i_ng.pattern_index=0
SORN.set_marked_variables(ind, info=(ind == []))
SORN.initialize(info=False)
############################################################################################################################################################
score_spont = score_predict_train = score_predict_test = 0
import Exploration.Network_UI.Network_UI as SUI
SUI.Network_UI(SORN, label='SORN UI default setup', storage_manager=sm, group_display_count=2, reduced_layout=False).show()
#SORN = run_plastic_phase(SORN, steps_plastic=10000, display=True, storage_manager=sm)#1000000
#print(set(np.nonzero(np.array(SORN['drum_act', 0].seen))[1]))
readout, X_train, Y_train, X_test, Y_test = train_readout(SORN, steps_train=10000, steps_test=100, source = SORN['drum_act', 0], display=True, stdp_off=True, storage_manager=sm)#100000
score_predict_train = get_score_predict_next_step(SORN, SORN['drum_act', 0],readout, X_train, Y_train, display=True, stdp_off=True, storage_manager=sm)
score_predict_test = get_score_predict_next_step(SORN, SORN['drum_act', 0],readout, X_test, Y_test, display=True, stdp_off=True, storage_manager=sm)
score_spont = get_score_spontaneous_music(SORN, SORN['drum_act', 0], readout, split_tracks=False, steps_spont=2000, display=True, stdp_off=True,
same_timestep_without_feedback_loop=False, steps_recovery=0, create_MIDI=True, storage_manager=sm)#, steps_recovery=15000
sm.save_obj('score_train', score_predict_train)
sm.save_obj('score_test', score_predict_test)
sm.save_obj('score_spontaneous', score_spont)
#plot_frequencies_poly(score_spont, path = sm.absolute_path+'frequencies', title='{} Ne, {} lag'.format(par['N_e'], par['TS']))
return score_predict_train, score_predict_test, score_spont
def run(attrs={'name':'hierarchical', 'ind':[], 'N_e':900, 'TS':[1], 'ff':True, 'fb':True, 'plastic':15000}):
so = True
print_info = attrs.get('print', True)
if print_info:
print(attrs)
#sm = None
sm = StorageManager(attrs['name'], random_nr=True, print_msg=print_info)
sm.save_param_dict(attrs)
#grammar mode: what level17_short simple
#source = FDTGrammarActivator_New(tag='grammar_act', random_blocks=True, input_density=0.015)#15/par['N_e']#.plot_char_input_statistics()#output_size=par['N_e']#15
#print(len(source.alphabet))
#source = LongDelayGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, mode=['level17_short'], input_density=0.015)#.print_test()#.plot_char_input_statistics()#10/par['N_e']
#source = SingleWordGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, input_density=0.015)
source = FewSentencesGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, input_density=0.015)
#print(source.print_test())
#print([t for t in source.get_all_grammar_transitions() if len(t)==2])
#source.plot_char_input_statistics()
#print(len(source.alphabet))
#img_source = Line_Patterns(tag='image_act', group_possibility=1.0, grid_width=20, grid_height=20, center_x=list(range(20)), center_y=20 / 2, degree=90, line_length=60, select_mode='random')
#print(source.get_text_score('. parrot likes trees. wolf wolf wolf..'))
SORN = Network()
#retina = NeuronGroup(net=SORN, tag='retina', size=get_squared_dim(20*20), behaviour={
# 2: SORN_init_neuron_vars(timescale=1),
#9: SORN_external_input(strength=1.0, pattern_groups=[img_source]),
#18: SORN_generate_output(init_TH='0.1;+-100%'),
#19: SORN_buffer_variables(random_temporal_output_shift=False),
#20: SORN_Refractory_Analog(factor='0.5;+-50%'),
#})
for layer, timescale in enumerate(attrs['TS']):
e_ng = NeuronGroup(net=SORN, tag='PC_{},prediction_source'.format(timescale), size=get_squared_dim(int(attrs['N_e']/timescale)), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='full', distribution='lognormal(0,[0.95#0])', normalize=True, partition_compensation=True),#0.89 uniform(0.1,0.11)#13%
4: SORN_init_afferent_synapses(transmitter='GABA', density='full', distribution='lognormal(0,[0.4#1])', normalize=True),#0.80222 uniform(0.1,0.11)#45%
12: SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
13: SORN_slow_syn(transmitter='GABA', strength='-[0.1698#3]', so=False),
17: SORN_fast_syn(transmitter='GABA', strength='-[0.1#4]', so=False),#0.11045
18: SORN_generate_output(init_TH='0.1;+-100%'),
19: SORN_buffer_variables(random_temporal_output_shift=False),
#20: SORN_Refractory_Digital(factor='0.5;+-50%', threshold=0.1),
20: SORN_Refractory_Analog(factor='0.5;+-50%'),
21: SORN_STDP(eta_stdp='[0.00015#5]'),#, STDP_F={-4:-0.01,-3:0.01,-2:0.1,-1:0.5,0:0.2,1:-0.3,2:-0.1,3:-0.05}, plot=True),#{-2:0.1,-1:0.5,0:0.2,1:-0.3,2:-0.1}
22: SORN_SN(syn_type='GLU', clip_max=None, behaviour_norm_factor=1.0),
#22.1: SORN_SN(syn_type='GLU_same', clip_max=None, behaviour_norm_factor=0.6),
#22.2: SORN_SN(syn_type='GLU_ff', clip_max=None, behaviour_norm_factor=0.25),
#22.3: SORN_SN(syn_type='GLU_fb', clip_max=None, behaviour_norm_factor=0.25),7z
23: SORN_IP_TI(h_ip='lognormal_real_mean([0.04#6], [0.2944#7])', eta_ip='[0.0006#8];+-50%', integration_length='[15#19];+-50%', clip_min=None),#30 #, gap_percent=10 #30;+-50% #0.0003 #np.mean(n.output_new)
25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.5#9];+-50%'),
26: SORN_SC_TI(h_sc='lognormal_real_mean([0.015#10], [0.2944#11])', eta_sc='[0.1#12];+-50%', integration_length='1'),#60;+-50% #0.05
27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]')
})
i_ng = NeuronGroup(net=SORN, tag='Inter_{}'.format(timescale), size=get_squared_dim(int(0.2 * attrs['N_e']/timescale)), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='50%', distribution='lognormal(0,[0.87038#14])', normalize=True), # 450
4: SORN_init_afferent_synapses(transmitter='GABA', density='20%', distribution='lognormal(0,[0.82099#15])', normalize=True), # 40
11: SORN_slow_syn(transmitter='GABA', strength='-[0.1838#16]', so=so),
14: SORN_fast_syn(transmitter='GLU', strength='[1.5#17]', so=so),#1.5353
15: SORN_fast_syn(transmitter='GABA', strength='-[0.08#18]', so=False),#0.08
18: SORN_generate_output(init_TH='0.1;+-0%'),
19: SORN_buffer_variables(),
#20: SORN_Refractory_Digital(factor='0.2;0.7', threshold=0.1),
20: SORN_Refractory_Analog(factor='0.2;0.7'),
#23: SORN_IP_TI(h_ip='lognormal_real_mean([0.08#6], [0.2944#7])', eta_ip='[0.0003#8];+-50%', integration_length='30;+-50%', clip_min=None),
})
i_ng['structure', 0].stretch_to_equal_size(e_ng)
#SynapseGroup(net=SORN, src=retina, dst=e_ng, tag='GLU,GLU_same', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
SynapseGroup(net=SORN, src=e_ng, dst=e_ng, tag='GLU,GLU_same', connectivity='(s_id!=d_id)*in_box(10)', partition=True)#.partition([10, 10], [partition, partition])
SynapseGroup(net=SORN, src=e_ng, dst=i_ng, tag='GLU', connectivity='in_box(10)', partition=True)#.partition([5, 5], [2, 2])
SynapseGroup(net=SORN, src=i_ng, dst=e_ng, tag='GABA', connectivity='in_box(10)', partition=True)
SynapseGroup(net=SORN, src=i_ng, dst=i_ng, tag='GABA', connectivity='(s_id!=d_id)*in_box(10)', partition=True)
if layer > 0:
if attrs.get('ff', True):#forward synapses
SynapseGroup(net=SORN, src=last_e_ng, dst=e_ng, tag='GLU,GLU_ff', connectivity='in_box(10)', partition=True)#.partition([10, 10], [partition, partition])
#SynapseGroup(net=SORN, src=last_e_ng, dst=i_ng, tag='GABA', connectivity='in_box(10)', partition=True)#.partition([5, 5], [2, 2])
if attrs.get('fb', False):#backward synapses
SynapseGroup(net=SORN, src=e_ng, dst=last_e_ng, tag='GLU,GLU_fb', connectivity='in_box(10)', partition=True)#.partition([10, 10], [partition, partition])
#SynapseGroup(net=SORN, src=e_ng, dst=last_i_ng, tag='GABA', connectivity='in_box(10)', partition=True)#.partition([5, 5], [2, 2])
#else:
# i_ng.add_behaviour(10, SORN_external_input(strength=1.0, pattern_groups=[source]))
e_ng.add_behaviour(9, SORN_external_input(strength=1.0, pattern_groups=[source]))
last_e_ng = e_ng
last_i_ng = i_ng
if __name__ == '__main__' and attrs.get('UI', False):
e_ng.color = get_color(0, timescale)
i_ng.color = get_color(1, timescale)
SORN.set_marked_variables(attrs['ind'], info=print_info, storage_manager=sm)
SORN.initialize(info=False)
###################################################################################################################
if __name__ == '__main__' and attrs.get('UI', False):
my_modules = get_default_UI_modules() + get_my_default_UI_modules()
Network_UI(SORN, modules=my_modules, label='SORN UI default setup', storage_manager=sm, group_display_count=2, reduced_layout=False).show()
score = 0
plastic_steps = attrs.get('plastic', 20000)
for i in range(1):
sm = StorageManager(attrs['name']+'[{:03d}]'.format(i+1), random_nr=True, print_msg=print_info)
sm.save_param_dict(attrs)
score += train_and_generate_text(SORN, plastic_steps, 5000, 1000, display=print_info, stdp_off=True, same_timestep_without_feedback_loop=True, steps_recovery=1000, storage_manager=sm)#5000, 3000 stdp_off=True
#score += get_oscillation_score_hierarchical(SORN, 0, 5000)
#print(predict_text_max_source_act(SORN, plastic_steps, 1000, 2000))
#t = max_source_act_text(SORN, 2000)
#print(t)
return score
def run_and_get_score(ind=[]):
input_ng = NeuronGroup(
size=4,
behaviour={
5: NeuronManualActivator(write_to='glu_inter_gamma_activity'),
7: ActivityBuffering(store_input=False,
min_buffersize=6) #original:7
}).add_tag('input')
output_ng = NeuronGroup(
size=4,
behaviour={
1:
DopamineProcessing(source_weight_id=1, target_weight_id=0),
2:
TemporalWeightCache(decay=1,
strength=1,
GLU_density=1.0,
set_weights=True,
GLU_equal_factor=1,
GLU_random_factor=0),
3:
RandomWeightFluctuation2(decay='[0.5#0]',
strength='[1.0#1]',
fluctuation='[0.001#2]',
density='[0.05#3]'),
4:
GlutamateCacheConvergeAndNormalization(
norm_value='0.05'), #todo output all same bug!!!!!!
5:
InterGammaGlutamate(GLU_density=1.0), #0.005
8:
ActivityBuffering(activity_multiplyer=0.0, firetreshold=0.1),
# 7: RandomWeightFluctuation(beta=4*0.2, gamma=3*0.2),
#8: HomeostaticMechanism(range_end='[550#3],+-30%', inc='[1.84#4],+-10%', dec='[4.92#5],+-10%',pattern_chance='[0.0086#6],+-30%'),
})
syns = SynapseGroup(src=input_ng, dst=output_ng).add_tag('GLU')
SORN_Global = Network([input_ng, output_ng], [syns], initialize=False)
SORN_Global.set_marked_variables(ind)
SORN_Global.initialize()
#syns.GLU_Synapse_Caches[0] = np.array([[1,0,0,0],[0,1,0,0],[0,0,1,0],[0,0,0,1]])
reward_strength = 1
last_reward = 0
last_dist = 0
sense_dist = 5
step_count = 100
reset_count = 1000
center = np.array([5.0, 5.0])
spawn_dist = 0.1
spawn_distances = []
inc = 0.01
dec = 0.01
for runs in range(reset_count):
#pos = np.array([9.0, 1.0])
#pos = np.array(np.random.rand(2)*10)
pos = pol2cart(
spawn_dist,
(2 * np.pi) * np.random.rand()) + center #/reset_count*runs
predicted_pos = np.array(pos.copy())
path = [pos.copy(), pos.copy()]
steps = [[0, 0], [0, 0]]
rewards = [0, 0]
colors = [(0, 0, 0), (0, 0, 0)]
#print(np.sum(syns.GLU_Synapse_Caches[0]),np.sum(syns.GLU_Synapse_Caches[1]))
#print(syns.GLU_Synapses)
for i in range(step_count):
step = get_next_step(path[-1], center, sense_dist, input_ng,
SORN_Global, output_ng)
pos += step
reward = get_reward(center, pos, predicted_pos, reward_strength)
#output_ng.dopamine_level = reward
SORN_Global.learning_off()
#future_step = get_next_step(np.clip(pos, 0, 10), input_ng, SORN_Global, output_ng)
predicted_pos = np.clip(pos, 0, 10) + step
SORN_Global.learning_on()
#print(out)
#print([td, bd, ld, rd])
path.append(np.clip(pos, 0, 10))
steps.append(step)
rewards.append(reward)
colors.append((i / step_count, runs / reset_count, 0.0)) #reward
if np.linalg.norm(path[-1] - center, 2) < 0.1:
spawn_dist = np.clip(spawn_dist + inc, 0.1, 5) #make it harder
output_ng.dopamine_level = 1
if ind is []:
print(spawn_dist, 'reward', step)
break
if output_ng.dopamine_level == 0:
spawn_dist = np.clip(spawn_dist - dec, 0.1, 5) #make it eayer
if ind is []:
print(spawn_dist, 'goal not found', step)
SORN_Global.simulate_iteration()
output_ng[RandomWeightFluctuation2].clear_weights(output_ng)
spawn_distances.append(spawn_dist)
path = np.array(path)
plt.scatter(
path[:, 0], path[:, 1],
color=colors) #, color=colors)#, color=dop) #, edgecolor='black'
if ind is []:
plt.scatter(center[0], center[1], color='blue')
plt.ylim([0, 10])
plt.xlim([0, 10])
plt.show()
plt.plot(spawn_distances)
plt.show()
score = np.sum(spawn_distances)
return score
def run(tag,
ind=[],
par={
'N_e': 2000,
'TS': [1],
'ts_plastic': 1000,
'ts_train': 1000,
'ts_test': 100
}):
assert tag in ['mono', 'poly']
name = 'MelodicTrack' + '_' + tag + str(par['N_e']) + 'N_e_' + str(
par['ts_plastic']) + '_' + str(par['ts_train']) + str(par['TS'])
sm = StorageManager(main_folder_name=tag,
folder_name=name,
random_nr=False,
print_msg=display)
sm.save_param_dict(par)
if tag == 'mono':
source = MonophonicActivator(
tag='music_act',
input_density=0.01,
which_alphabet='train',
beat_resolution=8) #output_size=par['N_e']
else:
source = PolyphonicActivator(tag='music_act',
input_density=0.01,
which_alphabet='train',
beat_resolution=4,
instrument=73)
SORN = Network()
for lag in par['TS']: #
e_ng = NeuronGroup(
net=SORN,
tag='exc_cell_{},prediction_source,text_input_group'.format(lag),
size=get_squared_dim(int(par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(timescale=lag, init_TH='0.1;+-100%'),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='13%',
distribution='lognormal(0,[0.89#0])',
normalize=True,
partition_compensation=True),
4:
SORN_init_afferent_synapses(
transmitter='GABA',
density='50%',
distribution='lognormal(0,[0.80222#1])',
normalize=True),
12:
SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
13:
SORN_slow_syn(transmitter='GABA',
strength='-[0.1698#3]',
so=so),
17:
SORN_fast_syn(transmitter='GABA',
strength='-[0.11045#4]',
so=so),
18:
SORN_generate_output_and_bufer(),
19:
SORN_Refractory(factor='0.5;+-50%'),
21:
SORN_STDP(eta_stdp='[0.0001#5]', STDP_F={
-1: 1,
1: -1
}),
22:
SORN_SN(syn_type='GLU', clip_max=None, init_norm_factor=1.0),
23:
SORN_IP_TI(h_ip='lognormal_real_mean([0.04#6], [0.2944#7])',
eta_ip='[0.0006#8];+-50%',
integration_length='[30#18];+-50%',
clip_min=None), #, gap_percent=10 #30;+-50% #0.0003
#24: SORN_NOX(mp='np.mean(n.output_new)', eta_nox='[0.3#9];+-50%', behaviour_enabled=False), #0.4
25:
SORN_NOX(mp='self.partition_sum(n)',
eta_nox='[0.3#9];+-50%'), # 0.4
26:
SORN_SC_TI(h_sc='lognormal_real_mean([0.01#10], [0.2944#11])',
eta_sc='[0.1#12];+-50%',
integration_length='1'), #60;+-50% #0.05
27:
SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]'),
#99: SynapseRecorder(['[np.sum(s.slow_add)]'], tag='ex_glu_syn_rec'),
#100: Recorder(['n.output'], tag='exc_out_rec')
})
i_ng = NeuronGroup(
net=SORN,
tag='inh_cell_{}'.format(lag),
size=get_squared_dim(int(0.2 * (par['N_e']))),
behaviour={
2:
SORN_init_neuron_vars(timescale=lag, init_TH='0.1;+-0%'),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='45%',
distribution='lognormal(0,[0.87038#14])',
normalize=True), # 450
4:
SORN_init_afferent_synapses(
transmitter='GABA',
density='20%',
distribution='lognormal(0,[0.82099#15])',
normalize=True), # 40
#11: SORN_slow_syn(transmitter='GABA', strength='-[0.1838#16]', so=so),
14:
SORN_fast_syn(transmitter='GLU', strength='[1.5#16]',
so=so), #1.5353
15:
SORN_fast_syn(transmitter='GABA', strength='-[0.08#17]',
so=so), #0.08
18:
SORN_generate_output_and_bufer(),
19:
SORN_Refractory(factor='0.2;0.7'),
#23: SORN_IP_TI(h_ip='lognormal_real_mean([0.08#6], [0.2944#7])', eta_ip='[0.0003#8];+-50%', integration_length='30;+-50%', clip_min=None),
#100: Recorder(['n.output'], tag='inh_out_rec')
})
i_ng['structure', 0].stretch_to_equal_size(e_ng)
SynapseGroup(net=SORN,
src=e_ng,
dst=e_ng,
tag='GLU,ee',
connectivity='(s_id!=d_id)*in_box(10)',
partition=True) #.partition([10, 10], [4, 4])
SynapseGroup(net=SORN,
src=e_ng,
dst=i_ng,
tag='GLU,ie',
connectivity='(s_id!=d_id)*in_box(10)',
partition=True) #.partition([5, 5], [2, 2])
SynapseGroup(net=SORN,
src=i_ng,
dst=e_ng,
tag='GABA,ei',
connectivity='(s_id!=d_id)*in_box(10)',
partition=True)
SynapseGroup(net=SORN,
src=i_ng,
dst=i_ng,
tag='GABA,ii',
connectivity='(s_id!=d_id)*in_box(10)',
partition=True)
if lag == 1:
i_ng.add_behaviour(
10, SORN_external_input(strength=1.0, pattern_groups=[source]))
e_ng.add_behaviour(
10, SORN_external_input(strength=1.0, pattern_groups=[source]))
#e_ng.add_behaviour(101, Recorder(['n.pattern_index'], tag='inp_rec'))
else:
#forward synapses
SynapseGroup(net=SORN, src=last_e_ng, dst=e_ng,
tag='GLU,eeff') #.partition([10, 10], [4, 4])
SynapseGroup(net=SORN, src=last_e_ng, dst=i_ng,
tag='GABA,ieff') #.partition([5, 5], [2, 2])
#backward synapses
SynapseGroup(net=SORN, src=e_ng, dst=last_e_ng,
tag='GLU,eebw') #.partition([10, 10], [4, 4])
SynapseGroup(net=SORN, src=e_ng, dst=last_i_ng,
tag='GABA,iebw') #.partition([5, 5], [2, 2])
last_e_ng = e_ng # excitatory neuron group
last_i_ng = i_ng # inhibitory neuron group
e_ng.color = (0, 0, 255, 255)
i_ng.color = (255, 0, 255, 255)
e_ng.pattern_index = 0
i_ng.pattern_index = 0
SORN.set_marked_variables(ind, info=(ind == []))
SORN.initialize(info=False)
############################################################################################################################################################
score_spont = score_predict_train = score_predict_test = 0
import Exploration.Network_UI.Network_UI as SUI
SUI.Network_UI(SORN,
label='SORN UI default setup',
storage_manager=sm,
group_display_count=2,
reduced_layout=False).show()
SORN = run_plastic_phase(SORN,
steps_plastic=1000,
display=True,
storage_manager=sm)
readout, X_train, Y_train, X_test, Y_test = train_readout(
SORN,
source=SORN['music_act', 0],
steps_train=1000,
steps_test=500,
display=True,
stdp_off=True,
storage_manager=sm)
score_predict_train = get_score_predict_next_step(SORN,
SORN['music_act', 0],
readout,
X_train,
Y_train,
display=True,
stdp_off=True,
storage_manager=sm)
score_predict_test = get_score_predict_next_step(SORN,
SORN['music_act', 0],
readout,
X_test,
Y_test,
display=True,
stdp_off=True,
storage_manager=sm)
score_spont = get_score_spontaneous_music(
SORN,
SORN['music_act', 0],
readout,
steps_spont=1000,
display=True,
stdp_off=True,
same_timestep_without_feedback_loop=False,
steps_recovery=0,
create_MIDI=True,
storage_manager=sm) #, steps_recovery=15000
sm.save_obj('score_train', score_predict_train)
sm.save_obj('score_test', score_predict_test)
sm.save_obj('score_spontaneous', score_spont)
if tag == 'mono':
plot_specific_performance(score_predict_train,
path=sm.absolute_path + 'train_',
title='predict training set, {} exc'.format(
par['N_e']))
plot_specific_performance(score_predict_test,
path=sm.absolute_path + 'test_',
title='predict test set, {} exc'.format(
par['N_e']))
plot_frequencies(score_spont,
path=sm.absolute_path + 'frequencies',
title='frequencies spontaneous output, {} exc'.format(
par['N_e']))
if tag == 'poly':
plot_frequencies_poly(
score_spont,
path=sm.absolute_path + 'frequencies_notes',
title='note frequencies spontaneous output, {} exc'.format(
par['N_e']))
plot_frequencies_chords_poly(
score_spont,
path=sm.absolute_path + 'frequencies_chords',
title='chord frequencies spontaneous output, {} exc'.format(
par['N_e']))
return score_predict_train, score_predict_test, score_spont
def run(
attrs={
'name': 'maze',
'ind': [],
'N_e': 900,
'TS': [1],
'ff': True,
'fb': True,
'plastic': 15000
}):
so = True
print_info = attrs.get('print', True)
if print_info:
print(attrs)
sm = StorageManager(attrs, random_nr=True, print_msg=print_info)
sm.save_param_dict(attrs)
#source = LongDelayGrammar(tag='grammar_act', output_size=attrs['N_e'], random_blocks=True, mode=['simple'], input_density=0.01)
maze = Maze(level='default', same_color=False)
SORN = Network() #behaviour={maze.get_network_behaviour()}
SORN.maze = maze
e_location = NeuronGroup(
net=SORN,
tag='PC_{},prediction_source'.format(1),
size=maze.get_location_neuron_dimension(),
behaviour={
2:
SORN_init_neuron_vars(timescale=1),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='90%',
distribution='uniform(0.1,1.0)',
normalize=True
), #20%#lognormal(0,[0.95#1]) #[13#0]% #, partition_compensation=True , partition_compensation=True #lognormal(0,0.95)
#4: SORN_init_afferent_synapses(transmitter='GABA', density='[30#1]%', distribution='uniform(0.0,1.0)', normalize=True),
5:
SORN_init_afferent_synapses(transmitter='GLU_cluster',
density='90%',
distribution='uniform(0.1,1.0)',
normalize=True),
8:
maze.get_vision_neuron_behaviour(),
9:
maze.get_location_neuron_behaviour(),
#10.0: SORN_slow_syn
10.0:
SORN_slow_syn_simple(transmitter='GLU', strength='1.0',
so=so), #todo: SORN_slow_syn_simple??????
10.1:
SORN_IP_WTA_apply(),
#10.15: WTA_refrac_apply(strengthfactor='[0.1#0]'),#0.1 #attrs['refrac']
10.2:
SORN_generate_output_K_WTA_partitioned(partition_size=7,
K='[0.02#1]'),
10.3:
SORN_slow_syn_simple(transmitter='GLU_cluster', strength='1.0'),
10.4:
SORN_generate_output_K_WTA_partitioned(
partition_size=7, K='[0.02#1]', filter_temporal_output=False),
10.5:
SORN_slow_syn_simple(transmitter='GLU_cluster', strength='1.0'),
10.6:
SORN_generate_output_K_WTA_partitioned(
partition_size=7, K='[0.02#1]', filter_temporal_output=False),
10.7:
SORN_slow_syn_simple(transmitter='GLU_cluster', strength='1.0'),
10.8:
SORN_generate_output_K_WTA_partitioned(
partition_size=7, K='[0.02#1]', filter_temporal_output=False),
# 12.1: SORN_WTA_iSTDP(eta_iSTDP='[0.00015#2]'),
# 12.2: SORN_SN(syn_type='GABA'),
#13.4: SORN_generate_output_K_WTA_partitioned(K='[0.12#4]'),
#13.5: SORN_WTA_fast_syn(transmitter='GABA', strength='-[0.5#5]', so=so),#[0.1383#2]SORN_fast_syn
#14: WTA_refrac(),
#, filter_temporal_output=True
15:
SORN_buffer_variables(random_temporal_output_shift=False),
18:
WTA_refrac(decayfactor=0.5),
20:
SORN_IP_WTA(h_ip='lognormal_real_mean([0.02#1], [0.2944#2])',
eta_ip='[0.007#3]',
target_clip_min=None,
target_clip_max=None), #-1.0 #1.0 #0.007
21.1:
SORN_STDP(transmitter='GLU',
eta_stdp='[0.00015#4]',
STDP_F={
-1: 0.2,
1: -1
}), #, 0: 1 #[0.00015#7] #0.5, 0: 3.0
21.2:
SORN_STDP(transmitter='GLU_cluster',
eta_stdp='[0.00015#5]',
STDP_F={0: 2.0}), #[0.00015#7]
22:
SORN_SN(syn_type='GLU', behaviour_norm_factor=1.0),
23:
SORN_SN(syn_type='GLU_cluster',
behaviour_norm_factor='[0.3#6]'), #0.1
})
'''
e_action = NeuronGroup(net=SORN, tag='E_action_{}'.format(timescale), size=maze.get_action_neuron_dimension(), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='full', distribution='uniform(0.1,0.11)', normalize=True, partition_compensation=True), # 0.89 lognormal(0,[0.5#0])
12: SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
13: SORN_slow_syn(transmitter='GABA', strength='-[0.1698#3]', so=False),
17: SORN_fast_syn(transmitter='GABA', strength='-[0.1#4]', so=False), # 0.11045
18: SORN_generate_output(init_TH='0.01'),
19: SORN_buffer_variables(),
20: SORN_Refractory_Analog(factor='0.7;+-50%'),
21: SORN_STDP(eta_stdp='[0.0015#5]', STDP_F={-1: 1, 1:0}, weight_attr='W_temp'),#bigger #todo!!!!
#22: SORN_SN(syn_type='GLU', clip_max=None, behaviour_norm_factor=5.0),
22: SORN_temporal_synapses(syn_type='GLU', behaviour_norm_factor=5.0),
23: SORN_IP_TI(h_ip='0.04', eta_ip='0.001', integration_length='[15#18];+-50%', clip_min=None),
#25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.5#9];+-50%'),
#26: SORN_SC_TI(h_sc='lognormal_real_mean([0.015#10], [0.2944#11])', eta_sc='[0.1#12];+-50%',integration_length='1'), # 60;+-50% #0.05
#27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]')
29: SORN_dopamine(),
30: maze.get_action_neuron_behaviour()
})
29: SORN_dopamine(),
30: maze.get_action_neuron_behaviour()
'''
'''
e_reward = NeuronGroup(net=SORN, tag='E_reward_{}'.format(timescale), size=maze.get_reward_neuron_dimension(), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='full', distribution='lognormal(0,[0.95#0])', normalize=True, partition_compensation=True),
9: maze.get_reward_neuron_behaviour(),
12: SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
18: SORN_generate_output(init_TH='0.01'),
19: SORN_buffer_variables(),
#20: SORN_Refractory(factor='0.5;+-50%'),
21: SORN_STDP(eta_stdp='[0.00015#5]'),
22: SORN_SN(syn_type='GLU', clip_max=None, behaviour_norm_factor=0.1),
#23: SORN_IP_TI(h_ip='0.04', eta_ip='0.001', integration_length='[15#18];+-50%', clip_min=None),
#25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.5#9];+-50%'),
#26: SORN_SC_TI(h_sc='lognormal_real_mean([0.015#10], [0.2944#11])', eta_sc='[0.1#12];+-50%',integration_length='1'), # 60;+-50% #0.05
#27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]')
})
'''
'''
e_punishment = NeuronGroup(net=SORN, tag='E_punishment_{}'.format(timescale), size=maze.get_punishment_neuron_dimension(), behaviour={
2: SORN_init_neuron_vars(timescale=timescale),
3: SORN_init_afferent_synapses(transmitter='GLU', density='full', distribution='lognormal(0,[0.95#0])', normalize=True, partition_compensation=True),
9: maze.get_punishment_neuron_behaviour(),
12: SORN_slow_syn(transmitter='GLU', strength='[0.1383#2]', so=so),
18: SORN_generate_output(init_TH='0.01'),
19: SORN_buffer_variables(),
#20: SORN_Refractory(factor='0.5;+-50%'),
21: SORN_STDP(eta_stdp='[0.00015#5]'), #todo!!!!
22: SORN_SN(syn_type='GLU', clip_max=None, behaviour_norm_factor=0.1),
#23: SORN_IP_TI(h_ip='0.04', eta_ip='0.001', integration_length='[15#18];+-50%', clip_min=None),
#25: SORN_NOX(mp='self.partition_sum(n)', eta_nox='[0.5#9];+-50%'),
#26: SORN_SC_TI(h_sc='lognormal_real_mean([0.015#10], [0.2944#11])', eta_sc='[0.1#12];+-50%',integration_length='1'), # 60;+-50% #0.05
#27: SORN_iSTDP(h_ip='same(SCTI, th)', eta_istdp='[0.0001#13]')
})
'''
SynapseGroup(net=SORN, src=e_location, dst=e_location, tag='GLU,GLU_same')
SynapseGroup(net=SORN,
src=e_location,
dst=e_location,
tag='GLU_cluster,syn')
#SynapseGroup(net=SORN, src=e_vision, dst=e_location, tag='GLU', partition=True)
#SynapseGroup(net=SORN, src=e_location, dst=e_action, tag='GLU', partition=True)
#SynapseGroup(net=SORN, src=e_location, dst=e_reward, tag='GLU', partition=True)
#SynapseGroup(net=SORN, src=e_reward, dst=e_location, tag='DOP+', partition=True)
#SynapseGroup(net=SORN, src=e_reward, dst=e_action, tag='DOP+', partition=True)
#SynapseGroup(net=SORN, src=e_location, dst=e_punishment, tag='GLU', partition=True)
#SynapseGroup(net=SORN, src=e_punishment, dst=e_location, tag='DOP-', partition=True)
#SynapseGroup(net=SORN, src=e_punishment, dst=e_action, tag='DOP-', partition=True)
if __name__ == '__main__' and attrs.get('UI', False):
e_location.color = (0, 0, 255, 255)
#e_action.color = (255, 255, 0, 255)
#e_vision.color = (0, 255, 255, 255)
#e_reward.color = (100, 255, 100, 255)
#e_punishment.color = (255, 100, 100, 255)
SORN.set_marked_variables(attrs['ind'],
info=print_info,
storage_manager=sm)
SORN.initialize(info=False)
###################################################################################################################
if __name__ == '__main__' and attrs.get('UI', False):
#default_modules.insert(0, maze_tab())
#Network_UI(SORN, label='SORN UI default setup', storage_manager=sm, group_display_count=4, reduced_layout=True).show()
Network_UI(
SORN,
modules=get_default_UI_modules(neuron_parameters=['activity']) +
[maze_tab()],
label='SORN UI K_WTA',
storage_manager=sm,
group_display_count=1,
reduced_layout=False).show()
return 0
