-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathofestimation.py
273 lines (209 loc) · 10.7 KB
/
ofestimation.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
import numpy
import pylab
import contrib.modelfit
def run_nonlinearity_detection(activities,predicted_activities,num_bins=20,display=False ,name='piece_wise_nonlinearity.png'):
(num_act,num_neurons) = numpy.shape(activities)
a=pylab.rcParams['font.size']
os = []
if display:
pylab.rc('font', size=1)
pylab.figure(dpi=100,facecolor='w',figsize=(17,12))
for i in xrange(0,num_neurons):
min_pact = numpy.min(predicted_activities[:,i])
max_pact = numpy.max(predicted_activities[:,i])
bins = numpy.arange(0,num_bins+1,1)/(num_bins*1.0)*(max_pact-min_pact) + min_pact
bins[-1]+=0.000001
ps = numpy.zeros(num_bins)
pss = numpy.zeros(num_bins)
for j in xrange(0,num_act):
bin = numpy.nonzero(bins>=predicted_activities[j,i])[0][0]-1
ps[bin]+=1
pss[bin]+=activities[j,i]
idx = numpy.nonzero(ps==0)
ps[idx]=1.0
tf = pss/ps
tf[idx]=0.0
if display:
pylab.subplot(15,7,i+1)
#pylab.plot(bins[0:-1],ps)
#pylab.plot(bins[0:-1],pss)
pylab.plot(bins[0:-1],tf)
os.append((bins,tf))
pylab.rc('font', size=a)
if display:
contrib.modelfit.release_fig(name)
return os
def apply_output_function(activities,of):
(x,y) = numpy.shape(activities)
acts = numpy.zeros(numpy.shape(activities))
for i in xrange(0,x):
for j in xrange(0,y):
(bins,tf) = of[j]
if activities[i,j] >= numpy.max(bins):
acts[i,j] = tf[-1]
elif activities[i,j] <= numpy.min(bins):
acts[i,j] = tf[0]
else:
bin = numpy.nonzero(bins>=activities[i,j])[0][0]-1
# do linear interpolation
a = bins[bin]
b = bins[bin+1]
alpha = (activities[i,j]-a)/(b-a)
if bin!=0:
c = (tf[bin]+tf[bin-1])/2
else:
c = tf[bin]
if bin!=len(tf)-1:
d = (tf[bin]+tf[bin+1])/2
else:
d = tf[bin]
acts[i,j] = c + (d-c)* alpha
return acts
def fit_sigmoids_to_of(activities,predicted_activities,offset=True,display=True,name='piece_wise_nonlinearity.png'):
(num_in,num_ne) = numpy.shape(activities)
from scipy import optimize
rand =numbergen.UniformRandom(seed=513)
a=pylab.rcParams['font.size']
if display:
pylab.rc('font', size=1)
pylab.figure(dpi=100,facecolor='w',figsize=(17,12))
fitfunc = lambda p, x: (offset*p[2])+p[3] / (1 + numpy.exp(-p[0]*(x-p[1]))) # Target function
errfunc = lambda p,x, y: numpy.mean(numpy.power(fitfunc(p, x) - y,2)) # Distance to the target function
params=[]
for i in xrange(0,num_ne):
min_err = 10e10
best_p = 0
for j in xrange(0,1000):
p0 = [20*rand(),10*(rand()-0.5),20*(rand()-0.5),50*rand()]
(p,success,c)=optimize.fmin_tnc(errfunc,p0[:],bounds=[(0,20),(-5,5),(-10,10),(0,50)],args=(numpy.array(predicted_activities[:,i].T)[0],numpy.array(activities[:,i].T)[0]),approx_grad=True,messages=0,maxfun=1000)
err = errfunc(p,numpy.array(predicted_activities[:,i].T)[0],numpy.array(activities[:,i].T)[0])
if err < min_err:
best_p = p
params.append(best_p)
if display:
pylab.subplot(15,7,i+1)
pylab.plot(numpy.array(predicted_activities[:,i].T)[0],numpy.array(activities[:,i].T)[0],'go')
pylab.plot(numpy.array(predicted_activities[:,i].T)[0],fitfunc(best_p,numpy.array(predicted_activities[:,i].T)[0]),'bo')
if display:
contrib.modelfit.release_fig(name)
pylab.rc('font', size=a)
return params
def fit_exponential_to_of(activities,predicted_activities,offset=True,display=True):
(num_in,num_ne) = numpy.shape(activities)
from scipy import optimize
pylab.figure()
fitfunc = lambda p, x: offset*p[0] + p[1] * numpy.exp(p[2]*(x-p[3])) # Target function
errfunc = lambda p,x, y: numpy.mean(numpy.power(fitfunc(p, x) - y,2)) # Distance to the target function
params=[]
for i in xrange(0,num_ne):
p0 = [0.0,1.0,0.1,0.0] # Initial guess for the parameters
(p,success,c)=optimize.fmin_tnc(errfunc,p0[:],bounds=[(-20,20),(-10,10),(0,10),(-5,5)],args=(numpy.array(predicted_activities[:,i].T)[0],numpy.array(activities[:,i].T)[0]),approx_grad=True,messages=0)
params.append(p)
if display:
pylab.subplot(13,13,i+1)
pylab.plot(numpy.array(predicted_activities[:,i].T)[0],numpy.array(activities[:,i].T)[0],'go')
pylab.plot(numpy.array(predicted_activities[:,i].T)[0],fitfunc(p,numpy.array(predicted_activities[:,i].T)[0]),'bo')
return params
def fit_power_to_of(activities,predicted_activities,display=True):
(num_in,num_ne) = numpy.shape(activities)
from scipy import optimize
pylab.figure()
fitfunc = lambda p, x: p[0] + p[1] * numpy.power(x,p[2]) # Target function
errfunc = lambda p,x, y: numpy.mean(numpy.power(fitfunc(p, x) - y,2)) # Distance to the target function
params=[]
for i in xrange(0,num_ne):
p0 = [0.0,1.0,-0.5] # Initial guess for the parameters
(p,success,c)=optimize.fmin_tnc(errfunc,p0[:],bounds=[(-20,20),(-1,1),(-1,2)],args=(numpy.array(predicted_activities[:,i].T)[0],numpy.array(activities[:,i].T)[0]),approx_grad=True,messages=0)
params.append(p)
if display:
pylab.subplot(15,7,i+1)
pylab.plot(numpy.array(predicted_activities[:,i].T)[0],numpy.array(activities[:,i].T)[0],'go')
pylab.plot(numpy.array(predicted_activities[:,i].T)[0],fitfunc(p,numpy.array(predicted_activities[:,i].T)[0]),'bo')
return params
def fit_addition_to_of(activities,predicted_activities1,predicted_activities2):
(num_in,num_ne) = numpy.shape(activities)
from scipy import optimize
fitfunc = lambda p, x, y: p[0]*x+p[1]*y+p[2] # Target function
errfunc = lambda p,x,y,z: numpy.mean(numpy.power(fitfunc(p, x, y) - z,2)) # Distance to the target function
params=[]
for i in xrange(0,num_ne):
p0 = [1.0,1.0,0.0] # Initial guess for the parameters
(p,success,c)=optimize.fmin_tnc(errfunc,p0[:],bounds=[(-20,20),(-20,20),(-20,20)],args=(numpy.array(predicted_activities1[:,i].T)[0],numpy.array(predicted_activities2[:,i].T)[0],numpy.array(activities[:,i].T)[0]),approx_grad=True,messages=0)
params.append(p)
print params
return params
def fit_multiplication_to_of(activities,predicted_activities1,predicted_activities2):
(num_in,num_ne) = numpy.shape(activities)
from scipy import optimize
fitfunc = lambda p, x, y: p[0]*(x+p[1])*(y+p[2]) # Target function
errfunc = lambda p,x,y,z: numpy.mean(numpy.power(fitfunc(p, x, y) - z,2)) # Distance to the target function
params=[]
for i in xrange(0,num_ne):
p0 = [1.0,0.0,0.0] # Initial guess for the parameters
(p,success,c)=optimize.fmin_tnc(errfunc,p0[:],bounds=[(-20,20),(-20,20),(-20,20)],args=(numpy.array(predicted_activities1[:,i].T)[0],numpy.array(predicted_activities2[:,i].T)[0],numpy.array(activities[:,i].T)[0]),approx_grad=True,messages=0)
params.append(p)
print params
return params
def apply_sigmoid_output_function(activities,of,offset=True):
sig = lambda p, x: (offset*p[2]) + p[3] * 1 / (1 + numpy.exp(-p[0]*(x-p[1])))
(x,y) = numpy.shape(activities)
new_acts = numpy.zeros((x,y))
for i in xrange(0,y):
new_acts[:,i] = sig(of[i],numpy.array(activities[:,i].T)[0]).T
return new_acts
def apply_exponential_output_function(activities,of,offset=True):
sig = lambda p, x: offset*p[0] + p[1] * numpy.exp(p[2]*(x-p[3]))
(x,y) = numpy.shape(activities)
new_acts = numpy.zeros((x,y))
for i in xrange(0,y):
new_acts[:,i] = sig(of[i],numpy.array(activities[:,i].T)[0]).T
return new_acts
def apply_power_output_function(activities,of):
sig = lambda p, x: p[0] + p[1] * numpy.power(x,p[2])
(x,y) = numpy.shape(activities)
new_acts = numpy.zeros((x,y))
for i in xrange(0,y):
new_acts[:,i] = sig(of[i],numpy.array(activities[:,i].T)[0]).T
return new_acts
def apply_addition_output_function(activities1,activities2,of):
fitfunc = lambda p, x, y: p[0]*x+p[1]*y+p[2] # Target function
(x,y) = numpy.shape(activities1)
new_acts = numpy.zeros((x,y))
for i in xrange(0,y):
new_acts[:,i] = fitfunc(of[i],numpy.array(activities1[:,i].T)[0],numpy.array(activities2[:,i].T)[0]).T
return new_acts
def apply_multiplication_output_function(activities1,activities2,of):
fitfunc = lambda p, x, y: p[0]*(x+p[1])*(y+p[2]) # Target function
(x,y) = numpy.shape(activities1)
new_acts = numpy.zeros((x,y))
for i in xrange(0,y):
new_acts[:,i] = fitfunc(of[i],numpy.array(activities1[:,i].T)[0],numpy.array(activities2[:,i].T)[0]).T
return new_acts
def fit2DOF(pred_act1,pred_act2,act,num_bins=10):
bin_size1 = (numpy.max(pred_act1,axis=0) - numpy.min(pred_act1,axis=0))/num_bins
bin_size2 = (numpy.max(pred_act2,axis=0) - numpy.min(pred_act2,axis=0))/num_bins
of = numpy.zeros((numpy.shape(act)[1],num_bins,num_bins))
ofn = numpy.zeros((numpy.shape(act)[1],num_bins,num_bins))
for i in xrange(0,numpy.shape(act)[0]):
idx1 = numpy.round_((pred_act1[i,:]-numpy.min(pred_act1,axis=0)) / bin_size1)
idx2 = numpy.round_((pred_act2[i,:]-numpy.min(pred_act2,axis=0)) / bin_size2)
idx1 = idx1 -(idx1 >= num_bins)
idx2 = idx2 -(idx2 >= num_bins)
j=0
for (x,y) in zip(numpy.array(idx1).flatten().tolist(),numpy.array(idx2).flatten().tolist()):
of[j,x,y] = of[j,x,y] + act[i,j]
ofn[j,x,y] = ofn[j,x,y] + 1
j=j+1
ofn = ofn + (ofn <= 0)
return (of/ofn,bin_size1,bin_size2,numpy.min(pred_act1,axis=0),numpy.min(pred_act2,axis=0))
def apply2DOF(activities1,activities2,ofs):
(of,bin_size1,bin_size2,offset1,offset2) = ofs
new_activities = numpy.zeros(numpy.shape(activities1))
idx = numpy.arange(0,numpy.shape(of)[0],1)
for i in xrange(0,numpy.shape(activities1)[0]):
idx1 = numpy.round_((activities1[i,:]-offset1) / bin_size1)
idx2 = numpy.round_((activities2[i,:]-offset2) / bin_size2)
idx1 = idx1 - (idx1 >= numpy.shape(of)[1])
idx2 = idx2 - (idx2 >= numpy.shape(of)[1])
new_activities[i,:] = [of[i] for i in zip(idx,numpy.array(idx1).flatten().tolist(),numpy.array(idx2).flatten().tolist())]
return new_activities