-
Notifications
You must be signed in to change notification settings - Fork 0
/
EmulatorNetwork.py
335 lines (285 loc) · 15.1 KB
/
EmulatorNetwork.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
from brian2 import *
prefs.codegen.target = 'numpy' # use the Python fallback
import numpy as np
from brian2_loihi import *
from time import *
class PathFinding():
def __init__(self, graph_info):
# Define variables/parameters
self.net = LoihiNetwork()
self.num_bits = graph_info["num_bits"]
self.edge_tuples = graph_info["edge_tuples"]
self.num_nodes = graph_info["num_nodes"]
self.costs_edges = graph_info["costs_edges"].reshape(len(graph_info["costs_edges"]),1)
self.costs_edges = np.unpackbits(self.costs_edges, axis=1)
self.costs_edges = np.flip(self.costs_edges,axis=1)
self.num_words = np.zeros(self.num_nodes)
#dictionaries for easy access to edges and synapses
self.edges = {}
self.synapses = {}
# count number of input to each node
self.count_num_words()
# list for Spike monitors necessary for reading output
self.output_monitors = []
# make nodes
self.nodes = self.make_nodes()
# Connect path by making edges
self.connect()
def connect(self):
#list for how many ports are needed at each minimum component
ports = np.zeros(self.num_nodes,dtype = int)
#time for one cycle to be completed
t_cyc = 4*self.num_bits + 11
#external neuron for cost input, fires at beginning of each addition phase of each cycle
times = np.arange(self.num_nodes)*(t_cyc)
indices = np.zeros(len(times))
ar = np.arange(len(self.costs_edges[0]))
EXT = LoihiSpikeGeneratorGroup(1, indices, times)
self.net.add(EXT)
#make edges and connect them to nodes
k = 0
for (fr, to) in self.edge_tuples:
# make edge and append to network
edge = AddComponent(self.num_bits, fr, to, ports[to])
self.net.add(edge.objects)
#name convention for edges: edge_fr_to_port
edge_name = "edge"+"_"+str(fr)+"_"+str(to)+"_"+str(ports[to])
self.edges[edge_name] = edge
#connect node output port to edge input port and edge output port to node input port
syn_name_input = edge_name + "_" + "input"
self.synapses[syn_name_input] = LoihiSynapses(self.nodes[fr].objects[0][1],self.edges[edge_name].objects[0][0])
self.synapses[syn_name_input].connect(i=np.arange(self.num_bits), j = np.arange(self.num_bits))
self.synapses[syn_name_input].w = 4
syn_name_output = edge_name + "_" + "output"
self.synapses[syn_name_output] = LoihiSynapses(self.edges[edge_name].objects[0][0],self.nodes[to].objects[0][0])
self.synapses[syn_name_output].connect(i=np.arange(6*self.num_bits,7*self.num_bits),
j = np.arange(self.num_bits*int(ports[to]), self.num_bits*(int(ports[to])+1)))
self.synapses[syn_name_output].w = 2
#count number of ports up for next iteration
ports[to] +=1
#connect external cost neuron to edge input
syn_name_cost = edge_name + "_" + "cost"
self.synapses[syn_name_cost] = LoihiSynapses(EXT,self.edges[edge_name].objects[0][0])
self.synapses[syn_name_cost].connect(i = np.zeros(np.count_nonzero(self.costs_edges[k]),dtype=int),
j = ar[np.where(self.costs_edges[k])]+self.num_bits)
self.synapses[syn_name_cost].w = 4
k += 1
self.net.add([self.synapses[syn_name_input],self.synapses[syn_name_output],self.synapses[syn_name_cost]])
def get_node(self,node_id):
return nodes[node_id]
def count_num_words(self):
#count number of inputs to each node
self.num_words[0] = 1
for (fr,to) in self.edge_tuples:
self.num_words[to]+=1
def make_nodes(self):
#make all nodes by initializing Minimum Components with appropriate numbers of inputs
nodes = []
for i in range(self.num_nodes):
node = MinimumComponent(i, self.num_bits, int(self.num_words[i]),self.num_nodes)
nodes.append(node)
self.net.add(node.objects)
self.output_monitors.append(node.objects[0][-1])
#print number of edge to see progress during initialization
print(i)
return nodes
class AddComponent():
def __init__(self, num_bits, source_id, target_id, port_id):
#initialize important parameters
self.num_bits = num_bits
self.objects=[]
self.source_id = source_id
self.target_id = target_id
self.port_id = port_id
self.make_edge()
def make_edge(self):
"""
These are the weight matrices necessary for connecting the different types of neurons within the addition component.
There are n_bits each of COST,MIN,CARRY, AND, XOR, AND2 and SUM-neurons each, with one of each type being necessary for
each layer and n_bits layers making up the addition component.
Delay: 0
TO
[ COST MIN CARRY AND XOR AND2 SUM
COST [0 0 R 2 0 0 0 ]
MIN [0 0 R 2 0 0 0 ]
CARRY [0 0 R 0 0 2 0 ]
FROM AND [0 0 0 0 -8 0 0 ]
XOR [0 0 0 0 0 2 0 ]
AND2 [0 0 0 0 0 0 -8 ]
SUM [0 0 0 0 0 0 0 ]
R is a matrix with 1s along the first off-diagonal because the COST, MIN and CARRY neurons of one layer need to be
connected to the CARRY neurons of the next layer rather than the same layer
Weight matrix for synapses with a delay of 2 timesteps
Delay: 2
[ COST MIN CARRY AND XOR AND2 SUM
COST [0 0 0 0 4 0 0 ]
MIN [0 0 0 0 4 0 0 ]
CARRY [0 0 0 0 0 0 0 ]
FROM AND [0 0 0 0 0 0 0 ]
XOR [0 0 0 0 0 0 0 ]
AND2 [0 0 0 0 0 0 0 ]
SUM [0 0 0 0 0 0 0 ]
]
Weight matrix for synapses with a delay of 3 timesteps
Delay: 3
[ COST MIN CARRY AND XOR AND2 SUM
COST [0 0 0 0 0 0 0 ]
MIN [0 0 0 0 0 0 0 ]
CARRY [0 0 0 0 0 0 4 ]
FROM AND [0 0 0 0 0 0 0 ]
XOR [0 0 0 0 0 0 4 ]
AND2 [0 0 0 0 0 0 0 ]
SUM [0 0 0 0 0 0 0 ]
]
"""
#edge name for unique synapse names, format "edge"_source_target_port
edge_name = "edge" + str(self.source_id) + "_" + str(self.target_id) + "_" + str(self.port_id)
#initialize the ADD-neurons
ADD = LoihiNeuronGroup(7*self.num_bits, threshold_v_mant=2, decay_v=1024
, decay_I=4096)
#define weight matrices
base_del0 = np.array([
[0,0,0,2,0,0,0],
[0,0,0,2,0,0,0],
[0,0,0,0,0,2,0],
[0,0,0,0,-8,0,0],
[0,0,0,0,0,2,0],
[0,0,0,0,0,0,-8],
[0,0,0,0,0,0,0]],dtype=int)
base_del1 = np.array([
[0,0,0,0,4,0,0],
[0,0,0,0,4,0,0],
[0,0,0,0,0,0,0],
[0,0,0,0,0,0,0],
[0,0,0,0,0,0,0],
[0,0,0,0,0,0,0],
[0,0,0,0,0,0,0]],dtype=int)
base_del2 = np.array([
[0,0,0,0,0,0,0],
[0,0,0,0,0,0,0],
[0,0,0,0,0,0,4],
[0,0,0,0,0,0,0],
[0,0,0,0,0,0,4],
[0,0,0,0,0,0,0],
[0,0,0,0,0,0,0]],dtype=int)
I = np.eye(self.num_bits,dtype=int)
weight_matrix_del0 = np.kron(base_del0,I)
weight_matrix_del1 = np.kron(base_del1,I)
for i in range(self.num_bits-1):
weight_matrix_del0[i][2*self.num_bits+i+1] = 2
weight_matrix_del0[self.num_bits+i][2*self.num_bits+i+1] = 2
weight_matrix_del0[2*self.num_bits+i][2*self.num_bits+i+1] = 2
weight_matrix_del2 = np.kron(base_del2,I)
#connections without any delay
DEL0 = LoihiSynapses(ADD, ADD, name=edge_name + 'DEL_0',sign_mode=synapse_sign_mode.MIXED)
(sources,targets) = np.where(weight_matrix_del0)
DEL0.connect(i=sources, j=targets)
DEL0.w = weight_matrix_del0[sources, targets]
#connections with a delay of 2 timesteps
DEL1 = LoihiSynapses(ADD,ADD, delay = 2, name=edge_name + 'DEL1')
(sources,targets) = np.where(weight_matrix_del1)
DEL1.connect(i=sources, j=targets)
DEL1.w = weight_matrix_del1[sources, targets]
#connections with a delay of 3 timesteps
DEL2 = LoihiSynapses(ADD,ADD,delay=3, name=edge_name + 'DEL2')
(sources,targets) = np.where(weight_matrix_del2)
DEL2.connect(i=sources, j=targets)
DEL2.w = weight_matrix_del2[sources, targets]
self.objects.append([ADD,DEL0,DEL1,DEL2])
class MinimumComponent():
"""
Define a Minimum Component with num_words input words, each having num_bits bits"""
def __init__(self, node_id, num_bits, num_words,num_nodes):
self.node_id = node_id
self.num_bits = num_bits
self.num_words = num_words
self.num_nodes = num_nodes
self.objects = []
self.make_node()
def make_node(self):
#parameters: L = number of bits, d = number of words, i if True or j if False
# returns arrays for connecting IN to ZER neurons correctly, returns i (presynaptic) range for i = True,
# and j (postsynaptic) range for False
def InZer(i):
if(i):
ar= []
for i in range(self.num_bits)[::-1]:
ar = np.concatenate((ar, np.arange(self.num_words)*self.num_bits+i), axis=None )
return ar.astype(np.int64)
else:
return np.arange(self.num_bits*self.num_words)
#returns array for postsynaptic range (j) for connecting IN and COP neurons
def InCop():
i = self.num_bits-1
ar = []
for j in arange(self.num_bits*self.num_words):
ar.append(i)
if i%self.num_bits == 0:
i +=2*self.num_bits
i-=1
return ar
#time for one cycle to be completed
t_cyc = 4*self.num_bits + 11
#define external neuron for cost input, fires at beginning of each minimum phase of each cycle
times = np.arange(self.num_nodes)*(t_cyc) + 8
indices = np.zeros(len(times))
EXT2 = LoihiSpikeGeneratorGroup(1, indices, times)
#decays for different num_bits
decays = [0,2700,1500,1050,800,600,500,410,350,300,260,225,200,175]
decay = decays[self.num_bits]
IN = LoihiNeuronGroup(self.num_bits*self.num_words,threshold_v_mant=2, decay_v=decay, decay_I=4096)
#Neurons CAN- Candidates for Maximum after each bit, from most significant to least
CAN = LoihiNeuronGroup(self.num_words*(self.num_bits+1),threshold_v_mant=2, decay_v=decay, decay_I=4096)
#Layer Neurons Or
OR = LoihiNeuronGroup(self.num_bits,threshold_v_mant=2, decay_v=decay, decay_I=4096)
#Layer Neurons INH- Inhibitory Neurons for Candidate Neurons
INH = LoihiNeuronGroup(self.num_words*self.num_bits,threshold_v_mant=2, decay_v=decay, decay_I=4096)
#ZER Neurons, check whether input bit is 0 and number is candidate
ZER = LoihiNeuronGroup(self.num_words*self.num_bits,threshold_v_mant=2, decay_v=300, decay_I=4096)
#Copy Layer for output
COP = LoihiNeuronGroup(self.num_words*self.num_bits,threshold_v_mant=2, decay_v=decay, decay_I=4096)
#Output Neurons, copy minimum
OUT = LoihiNeuronGroup(self.num_bits,threshold_v_mant=2, decay_v=decay, decay_I=4096)
#setup layer: candidate neurons for first bit of each word should be active
EXTCAN = LoihiSynapses(EXT2,CAN)
EXTCAN.connect(i=np.zeros(self.num_words).astype(int),j=np.arange(self.num_words))
EXTCAN.w = 4
#external to input for synchronization
EXTIN = LoihiSynapses(EXT2, IN)
EXTIN.connect(i=np.zeros(self.num_bits*self.num_words).astype(int),j=np.arange(self.num_bits*self.num_words))
EXTIN.w = 2
#main layers: check whether inputs still candidates after next significant bits
#for explanation of connections and weights, see thesis
INZER= LoihiSynapses(IN,ZER,sign_mode=synapse_sign_mode.INHIBITORY)
INZER.connect(i=InZer(True), j =InZer(False))
INZER.w = -16
CANZER = LoihiSynapses(CAN,ZER)
CANZER.connect(i=np.arange(self.num_words*self.num_bits),j=np.arange(self.num_words*self.num_bits))
CANZER.w = 4
ZERINH = LoihiSynapses(ZER,INH,sign_mode=synapse_sign_mode.INHIBITORY)
ZERINH.connect(i=np.arange(self.num_words*self.num_bits),j=np.arange(self.num_words*self.num_bits))
ZERINH.w = -8
ZEROR = LoihiSynapses(ZER,OR)
ZEROR.connect(i=np.arange(self.num_bits*self.num_words), j =np.repeat(np.arange(self.num_bits),self.num_words))
ZEROR.w = 4
ORINH = LoihiSynapses(OR,INH)
ORINH.connect(i=np.repeat(np.arange(self.num_bits),self.num_words),j=np.arange(self.num_words*self.num_bits))
ORINH.w = 4
INHCAN = LoihiSynapses(INH,CAN,sign_mode=synapse_sign_mode.INHIBITORY)
INHCAN.connect(i=np.arange(self.num_words*self.num_bits),j=np.arange(self.num_words,self.num_words*self.num_bits+self.num_words))
INHCAN.w = -8
CANCAN = LoihiSynapses(CAN,CAN,delay= 3)
CANCAN.connect(i=np.arange(self.num_words*self.num_bits),j=np.arange(self.num_words,self.num_words*self.num_bits+self.num_words))
CANCAN.w = 4
INCOP = LoihiSynapses(IN,COP,delay= 3*self.num_bits)
INCOP.connect(i=np.arange(self.num_words*self.num_bits),j=InCop())
INCOP.w = 2
CANCOP = LoihiSynapses(CAN,COP)
CANCOP.connect(i=np.repeat(np.arange(self.num_bits*self.num_words, self.num_bits*self.num_words+self.num_words),self.num_bits), j = np.arange(self.num_words*self.num_bits))
CANCOP.w = 2
COPOUT = LoihiSynapses(COP,OUT)
COPOUT.connect(i=np.arange(self.num_bits*self.num_words),j=np.tile(np.arange(self.num_bits)[::-1],self.num_words))
COPOUT.w = 4
OUT_MON = LoihiSpikeMonitor(OUT, name = "out_monitor_node_" + str(self.node_id))
self.objects.append([IN, OUT, EXT2, CAN, OR, INH, ZER, COP, EXTCAN, EXTIN, INZER, CANZER, ZERINH, ZEROR, ORINH, INHCAN, CANCAN, INCOP,
CANCOP, COPOUT, OUT_MON])