-
Notifications
You must be signed in to change notification settings - Fork 0
/
bandit_task_3arm.py
306 lines (255 loc) · 9.98 KB
/
bandit_task_3arm.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
import random
from numeric import array,norm
import sys
import nef
import nps
import hrr
import datetime
import timeview.components.core as core
import timeview.view
import os
from ca.nengo.math.impl import ConstantFunction
from ca.nengo.model.impl import FunctionInput
from ca.nengo.model import Units
# For experiment running
#directory = 'trevor/data'
directory = None
e_name = 'Normal'
num_expts = 26
# State lengths
delay_dur = 0.3
approach_dur = 0.1
reward_dur = 0.1
# Experiment parameters
state_d = 4
action_d = 2
NperD = 40
# Gate parameters
gate_N = 100
gate_intercept = (-0.15,0.2)
gate_rate = (50,150)
class BanditTask(nef.SimpleNode):
def __init__(self,name,dims,trials_per_block=40,block_rewards=[[0.21,0.63],[0.63,0.21],[0.12,0.72],[0.72,0.12]]):
# parameters
self.dims = dims
self.trials_per_block = trials_per_block
if len(block_rewards[0]) != dims:
raise Exception('block_reward dimensionality must match dims')
self.block_rewards = block_rewards
# vars and constants
self.trial_num = 0
self.delay_t = 0.0
self.approach_t = 0.0
self.reward_t = 0.0
self.reward = [0.0] * dims
self.thalamus_sum = [0.0] * dims
self.thalamus_choice = 0
self.rewarded = 0
self.reward_val = 1.0
self.gate_val = [0.9]
self.vstr_gate_val = [1.0]
self.data_log = []
# generate random state_d-d unit vector
self.ctx_val = array([random.gauss(0,1) for i in range(state_d)])
self.ctx_val /= norm(self.ctx_val)
self.state = 'delay'
nef.SimpleNode.__init__(self,name)
def get_experiment_length(self):
leeway = 0.003
return (delay_dur + approach_dur + reward_dur +
leeway) * self.trials_per_block * len(self.block_rewards)
def origin_cortex(self):
return self.ctx_val
def origin_cortex_gate(self):
return self.gate_val
def origin_vstr_gate(self):
return self.vstr_gate_val
def origin_reward(self):
return self.reward
def termination_thalamus(self,x):
t = self.t_start
if self.state == 'delay':
self.gate_val = [self.gate_val[0]+0.001]
self.vstr_gate_val = [self.vstr_gate_val[0]+0.001]
self.reward = [0.0] * self.dims
if t >= self.delay_t + delay_dur:
self.state = 'go'
elif self.state == 'go':
self.gate_val = [0.0]
self.thalamus_sum = [0.0] * self.dims
self.trial_num += 1
self.approach_t = t
self.state = 'approach'
elif self.state == 'approach':
for i in range(self.dims):
self.thalamus_sum[i] += x[i]
if t >= self.approach_t + approach_dur:
thalamus_min = min(self.thalamus_sum)
for i in range(len(self.thalamus_sum)):
if self.thalamus_sum[i] == thalamus_min:
self.thalamus_choice = i
block = (self.trial_num-1) / self.trials_per_block
if (block >= len(self.block_rewards)):
self.state = 'reward'
return
##########
## NB!! ##
##########
rand = random.random()
if rand <= self.block_rewards[block][self.thalamus_choice]:
self.rewarded = 1
self.reward = [-1.0*self.reward_val] * self.dims
self.reward[self.thalamus_choice] = self.reward_val
else:
self.rewarded = 0
self.reward = [self.reward_val] * self.dims
self.reward[self.thalamus_choice] = -1.0 * self.reward_val
# out_file structure:
# trial number, choice, rewarded, thalamus_sums
out_l = str(self.trial_num)+', '+str(self.thalamus_choice)+', '+str(self.rewarded)
for i in range(len(self.thalamus_sum)):
out_l += ', '+str(self.thalamus_sum[i])
self.data_log.append(out_l)
self.reward_t = t
self.state = 'reward'
elif self.state == 'reward':
self.vstr_gate_val = [0.0]
if t >= self.reward_t + reward_dur:
self.gate_val = [1.0]
self.delay_t = t
self.state = 'delay'
def write_data_log(self, filename):
"""Attempts to write the contents of self.data_log to
the file pointed to by the consumed string, filename.
If there is an error writing to that file,
the contents of self.data_log are printed to console instead.
"""
try:
f = open(filename, 'a+')
except:
print "Error opening %s" % filename
return self.print_data_log()
for line in self.data_log:
f.write("%s\n" % line)
f.close()
def print_data_log(self):
"""Prints the contents of self.data_log to the console."""
for line in self.data_log:
print line
class BanditWatch:
def __init__(self,objs):
self.objs=objs
def check(self,obj):
return obj in self.objs
def measure(self,obj):
r=[]
r.append(obj.trial_num)
r.append(obj.state)
r.append(obj.thalamus_choice)
r.append(obj.rewarded)
for sum in obj.thalamus_sum:
r.append(sum)
return r
def views(self,obj):
return [('bandit task',BanditView,dict(func=self.measure,label="Bandit Task"))]
from javax.swing.event import *
from java.awt import *
from java.awt.event import *
class BanditView(core.DataViewComponent):
def __init__(self,view,name,func,args=(),label=None):
core.DataViewComponent.__init__(self,label)
self.view=view
self.name=name
self.func=func
self.data=self.view.watcher.watch(name,func,args=args)
self.setSize(200,100)
def paintComponent(self,g):
core.DataViewComponent.paintComponent(self,g)
f_size = g.getFont().size
x_offset = 5
try:
data=self.data.get(start=self.view.current_tick,count=1)[0]
except:
return
cur_y = f_size*3
g.drawString("Trail "+str(data[0]),x_offset,cur_y)
cur_y += f_size
g.drawString("State: "+data[1],x_offset,cur_y)
cur_y += f_size
g.drawString("Thalamus sum",x_offset,cur_y)
cur_y += f_size
cur_x = x_offset
for sum in data[4:]:
g.drawString(str(round(sum*100)/100),cur_x,cur_y)
cur_x += 40
cur_y += f_size
g.drawString("Choice: "+str(data[2]),x_offset,cur_y)
cur_y += f_size
if data[3]: r_s = "Yes"
else: r_s = "No"
g.drawString("Rewarded: "+r_s,x_offset,cur_y)
def gate_weights(w):
for i in range(len(w)):
for j in range(len(w[0])):
#w[i][j] = -0.02
w[i][j] = -0.0002
return w
def rand_weights(w):
for i in range(len(w)):
for j in range(len(w[0])):
w[i][j] = random.uniform(-1e-4,1e-4)
return w
alpha = 1.0
def pred_error(x):
# for each action, prediction error is
# a [ R + g * V(S) - V(S(t-1))]
return [alpha * (x[2] - x[0]), alpha * (x[3] - x[1])]
def build_network():
net = nef.Network('BanditTask_o')
experiment = BanditTask('ExperimentRunner',action_d)
experiment.getTermination('thalamus').setDimensions(action_d)
net.add(experiment)
timeview.view.watches.append(BanditWatch([experiment]))
cortex = net.make('Cortex',NperD*state_d,state_d)
net.connect(experiment.getOrigin('cortex'),cortex)
cortex_gate = net.make('CortexGate',gate_N,1,encoders=[[1.0]],intercept=gate_intercept,max_rate=gate_rate)
net.connect(experiment.getOrigin('cortex_gate'),cortex_gate)
net.connect(cortex_gate,cortex,weight_func=gate_weights)
thalamus = net.make('Thalamus',NperD*action_d,action_d)
net.connect(thalamus,experiment.getTermination('thalamus'))
nps.basalganglia.make_basal_ganglia(net,cortex,thalamus,action_d,NperD,learn=True)
StrD1 = net.network.getNode('StrD1')
StrD2 = net.network.getNode('StrD2')
vStr = net.make('Ventral Striatum',NperD*action_d*2,action_d*2,max_rate=(100,200))
net.connect(cortex,vStr,index_post=[0,1],weight_func=rand_weights)
net.connect(experiment.getOrigin('reward'),vStr,index_post=[2,3])
net.connect(vStr,vStr,func=pred_error,index_post=range(action_d),modulatory=True)
net.connect(vStr,StrD1,func=pred_error,modulatory=True)
net.connect(vStr,StrD2,func=pred_error,modulatory=True)
l_args = {'stpd':False, 'oja':False, 'rate':1e-7}
net.learn(vStr,'Cortex','Ventral Striatum',**l_args)
net.learn_array(StrD1,'Cortex','Ventral Striatum',**l_args)
net.learn_array(StrD2,'Cortex','Ventral Striatum',**l_args)
vStr_gate = net.make('vStrGate',gate_N,1,encoders=[[1.0]],intercept=gate_intercept,max_rate=gate_rate)
net.connect(experiment.getOrigin('vstr_gate'),vStr_gate)
net.connect(vStr_gate,vStr,weight_func=gate_weights)
return net
def run_experiment():
net = build_network()
experiment = net.network.getNode('ExperimentRunner')
if directory != None:
net.network.run(0,experiment.get_experiment_length())
now = datetime.datetime.now()
f_name = os.path.join(directory, e_name+'-'+now.strftime("%Y-%m-%d_%H-%M-%S")+'.csv')
f = open(f_name, 'w')
f.write('delay_dur=%f\napproach_dur=%f\nreward_dur=%f\n' % (delay_dur,approach_dur,reward_dur))
f.close()
experiment.write_data_log(f_name)
return net
if directory is not None:
for _ in range(num_expts):
net = run_experiment()
sys.exit()
else:
net = build_network()
net.add_to_nengo()