from NetworkBehaviour.Input.TREN.Lines import *
from NetworkCore.Network import *
from NetworkCore.Synapse_Group import *
from NetworkBehaviour.Structure.Structure import *
from Exploration.StorageManager.StorageManager import *
so = True
n_Neurons = 900
#source = FDTGrammarActivator_New(tag='grammar_act', random_blocks=True, input_density=0.015)#.plot_char_input_statistics()#output_size=par['N_e']#15
source = LongDelayGrammar(tag='grammar_act', random_blocks=True, mode=['simple'], input_density=0.015)
#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)
SORN = Network()
PC = NeuronGroup(net=SORN, tag='Pyramidal,prediction_source', size=get_squared_dim(int(n_Neurons)), behaviour={
2: SORN_init_neuron_vars(nit_TH='0.1;+-100%'),
3: SORN_init_afferent_synapses(transmitter='GLU', density='50%', distribution='uniform(0.1,0.11)', normalize=True, partition_compensation=True), #lognormal(0,0.6)
4: SORN_init_afferent_synapses(transmitter='GABA_Dendrite', density='30%', distribution='uniform(0.1,0.11)', normalize=True),
5: SORN_init_afferent_synapses(transmitter='GABA_Soma', density='30%', distribution='uniform(0.1,0.11)', normalize=True),
6: SORN_init_afferent_synapses(transmitter='GABA_AIS', density='30%', distribution='uniform(0.1,0.11)', normalize=True),
12: SORN_slow_syn(transmitter='GLU', strength='0.1383', so=so),
13: SORN_slow_syn(transmitter='GABA_Dendrite', strength='-0.1', so=False),
14: SORN_slow_input_collect(),
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),
19: SORN_input_collect(),
def run(tag='PV_SOM', ind=[], par={'N_e': 900, 'TS': [1]}):
print(tag, par)
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
#print(len(source.alphabet))
source = LongDelayGrammar(
tag='grammar_act',
output_size=par['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 timecale in par['TS']:
PC = NeuronGroup(
net=SORN,
tag='Pyramidal Cell {},prediction_source'.format(timecale),
size=get_squared_dim(int(par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timecale,
init_TH='0.1;+-100%'),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='[50#0]%',
distribution='uniform(0.1,0.11)',
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),
14:
SORN_slow_input_collect(),
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),
19:
SORN_input_collect(),
20:
SORN_Refractory(factor='[0.5#5];+-50%'),
21:
SORN_STDP_new(eta_stdp='[0.0015#6]',
prune_stdp=False,
excitation_punishment=0.0), #0.1#todo: test!!!
22:
SORN_SN(syn_type='GLU', clip_max=None, init_norm_factor=1.0),
23:
SORN_IP_TI(mp='n.output_new/2.0+n.output_new_temp/2.0',
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),
#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]'),
30:
SORN_finish()
})
MT_SOM = NeuronGroup(
net=SORN,
tag='Martinotti Cell {},Somatostatin'.format(timecale),
size=get_squared_dim(int(0.07 * par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timecale,
init_TH='0.1;+-0%'),
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
19:
SORN_input_collect(),
#20: SORN_Refractory(factor='0.2;0.7'),
30:
SORN_finish()
})
EXP_NOX_CELL = NeuronGroup(
net=SORN,
tag='NOX Cell {}'.format(timecale),
size=get_squared_dim(int(16)),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timecale,
init_TH='0.04',
activation_function='identity'),
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
19:
SORN_input_collect(),
#20: SORN_Refractory(factor='0.2;0.7'),
30:
SORN_finish()
})
BA_PV = NeuronGroup(
net=SORN,
tag='Basket Cell {},Parvalbumin'.format(timecale),
size=get_squared_dim(int(0.07 * par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timecale,
init_TH='0.1;+-0%'),
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
19:
SORN_input_collect(),
#20: SORN_Refractory(factor='0.2;0.7'),
30:
SORN_finish()
})
CH_PV = NeuronGroup(
net=SORN,
tag='Chandelier Cell {},Parvalbumin'.format(timecale),
size=get_squared_dim(int(0.07 * par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timecale,
init_TH='0.1;+-0%'),
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='0.4',
so=so), #1.5353
#15: SORN_fast_syn(transmitter='GABA', strength='-[0.08#17]', so=False),#0.08
19:
SORN_input_collect(),
#19: SORN_Refractory(factor='0.2;0.7'),
30:
SORN_finish()
})
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 last_PC is None:
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
PC.color = get_color(0, timecale)
MT_SOM.color = get_color(1, timecale)
BA_PV.color = get_color(2, timecale)
CH_PV.color = get_color(3, timecale)
EXP_NOX_CELL.color = get_color(4, timecale)
#EXP_NOX_CELL.display_min_max_act = (0, 0.3)
SORN.set_marked_variables(ind, info=(ind == []), storage_manager=sm)
SORN.initialize(info=False)
###################################################################################################################
score = 0
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 += train_and_generate_text(
SORN,
30000,
5000,
2000,
display=True,
stdp_off=True,
same_timestep_without_feedback_loop=False,
steps_recovery=0,
storage_manager=sm) #, steps_recovery=15000
#score += get_oscillation_score_hierarchical(SORN, 0, 5000)
return score
from NetworkBehaviour.Structure.Structure import *
from NetworkBehaviour.Logic.Example_Simple_Network.Simple_Network_behaviour import *
import Exploration.UI.Network_UI.Network_UI as NUI
from Exploration.UI.Network_UI.Tabs.activity_tab import *
from Exploration.UI.Network_UI.Tabs.sidebar_activity_module import *
from Exploration.UI.Network_UI.Tabs.spiketrain_tab import *
from Exploration.UI.Network_UI.Tabs.fourier_tab import *
from Exploration.UI.Network_UI.Tabs.weight_tab import *
from Exploration.UI.Network_UI.Tabs.sidebar_fast_forward_module import *
from Exploration.UI.Network_UI.Tabs.sidebar_save_load_module import *
from Exploration.UI.Network_UI.Tabs.stability_tab import *
from Exploration.UI.Network_UI.Tabs.hist_tab import *
number_of_neurons = 900
Easy_Network = Network()
Easy_Neurons = NeuronGroup(net=Easy_Network,
tag='neurons',
size=get_squared_dim(number_of_neurons),
behaviour={
1:
Easy_neuron_initialize(syn_density=0.05),
2:
Easy_neuron_collect_input(noise_density=0.1,
noise_strength=0.11),
3:
Easy_neuron_generate_output(threshold=0.1),
4:
Easy_neuron_Refractory(decay='0.8;0.99')
})
def run(tag='hierarchical', ind=[], par={'N_e': 900, 'TS': [1]}):
#print(tag, par)
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
#print(len(source.alphabet))
source = LongDelayGrammar(
tag='grammar_act',
output_size=par['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)
#print(source.get_text_score('. parrot likes trees. wolf wolf wolf..'))
SORN = Network()
last_e_ng = None
for timecale in par['TS']:
e_ng = NeuronGroup(
net=SORN,
tag='PC_{},prediction_source'.format(timecale),
size=get_squared_dim(int(par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timecale,
init_TH='0.1;+-100%'),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='13%',
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)
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_input_collect(),
19:
SORN_Refractory(factor='0.5;+-50%'),
21:
SORN_STDP(eta_stdp='[0.00015#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='[15#18];+-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]'),
30:
SORN_finish(),
})
i_ng = NeuronGroup(
net=SORN,
tag='Inter_{}'.format(timecale),
size=get_squared_dim(int(0.2 * par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timecale,
init_TH='0.1;+-0%'),
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#16]',
so=so), #1.5353
15:
SORN_fast_syn(transmitter='GABA',
strength='-[0.08#17]',
so=False), #0.08
18:
SORN_input_collect(),
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),
30:
SORN_finish()
})
i_ng['structure', 0].stretch_to_equal_size(e_ng)
SynapseGroup(
net=SORN,
src=e_ng,
dst=e_ng,
tag='GLU',
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)
#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]))
if last_e_ng is not None:
#forward synapses
SynapseGroup(
net=SORN,
src=last_e_ng,
dst=e_ng,
tag='GLU',
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])
#backward synapses
SynapseGroup(
net=SORN,
src=e_ng,
dst=last_e_ng,
tag='GLU',
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])
last_e_ng = e_ng
last_i_ng = i_ng
e_ng.color = get_color(0, timecale)
i_ng.color = get_color(1, timecale)
SORN.set_marked_variables(ind, info=(ind == []), storage_manager=sm)
SORN.initialize(info=False)
###################################################################################################################
score = 0
#Network_UI(SORN, label='SORN UI default setup', storage_manager=sm, group_display_count=2, reduced_layout=True).show()
score += train_and_generate_text(
SORN,
15000,
5000,
2000,
display=True,
stdp_off=True,
same_timestep_without_feedback_loop=True,
steps_recovery=2000,
storage_manager=sm) #, steps_recovery=15000
#score += get_oscillation_score_hierarchical(SORN, 0, 5000)
return 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: NeuronRecorder(['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: NeuronRecorder(['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, NeuronRecorder(['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.UI.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=1000000,
display=True,
storage_manager=sm)
readout, X_train, Y_train, X_test, Y_test = train_readout(
SORN,
steps_train=100000,
steps_test=100,
source=SORN['drum_act', 0],
display=True,
stdp_off=True,
storage_manager=sm)
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(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']
SORN = Network()
last_PC = None
for i, timescale in enumerate(par['TS']): #
PC = NeuronGroup(
net=SORN,
tag='Pyramidal Cell {},prediction_source'.format(timescale),
size=get_squared_dim(int(par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timescale,
init_TH='0.1;+-100%'),
3:
SORN_init_afferent_synapses(
transmitter='GLU',
density='[50#0]%',
distribution='uniform(0.1,0.11)',
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),
14:
SORN_slow_input_collect(),
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),
19:
SORN_input_collect(),
20:
SORN_Refractory(factor='[0.5#5];+-50%'),
21:
SORN_STDP_new(eta_stdp='[0.0015#6]',
prune_stdp=False,
excitation_punishment=0.0), #0.1#todo: test!!!
22:
SORN_SN(syn_type='GLU', clip_max=None, init_norm_factor=1.0),
23:
SORN_IP_TI(mp='n.output_new/2.0+n.output_new_temp/2.0',
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),
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]'),
30:
SORN_finish()
})
MT_SOM = NeuronGroup(
net=SORN,
tag='Martinotti Cell {},Somatostatin'.format(timescale),
size=get_squared_dim(int(0.07 * par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timescale,
init_TH='0.1;+-0%'),
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
19:
SORN_input_collect(),
#20: SORN_Refractory(factor='0.2;0.7'),
30:
SORN_finish()
})
'''
EXP_NOX_CELL = NeuronGroup(net=SORN, tag='NOX Cell {}'.format(timescale), size=get_squared_dim(int(16)), behaviour={
2: SORN_init_neuron_vars(iteration_lag=timescale, init_TH='0.04', activation_function='identity'),
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
19: SORN_input_collect(),
#20: SORN_Refractory(factor='0.2;0.7'),
30: SORN_finish()
})
'''
BA_PV = NeuronGroup(
net=SORN,
tag='Basket Cell {},Parvalbumin'.format(timescale),
size=get_squared_dim(int(0.07 * par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timescale,
init_TH='0.1;+-0%'),
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
19:
SORN_input_collect(),
#20: SORN_Refractory(factor='0.2;0.7'),
30:
SORN_finish()
})
CH_PV = NeuronGroup(
net=SORN,
tag='Chandelier Cell {},Parvalbumin'.format(timescale),
size=get_squared_dim(int(0.07 * par['N_e'])),
behaviour={
2:
SORN_init_neuron_vars(iteration_lag=timescale,
init_TH='0.1;+-0%'),
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='0.4',
so=so), #1.5353
#15: SORN_fast_syn(transmitter='GABA', strength='-[0.08#17]', so=False),#0.08
19:
SORN_input_collect(),
#19: SORN_Refractory(factor='0.2;0.7'),
30:
SORN_finish()
})
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)
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 last_PC is None:
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
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(ind, info=(ind == []), storage_manager=sm)
SORN.initialize(info=False)
############################################################################################################################################################
score_spont = score_predict_train = score_predict_test = 0
Network_UI(SORN,
label='SORN UI PC PV SOM',
storage_manager=sm,
group_display_count=4,
reduced_layout=False).show()
SORN = run_plastic_phase(SORN,
steps_plastic=1000000,
display=True,
storage_manager=sm)
#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=100000,
steps_test=100,
source=SORN['drum_act', 0],
display=True,
stdp_off=True,
storage_manager=sm)
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