-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_experiment.py
194 lines (154 loc) · 10.4 KB
/
run_experiment.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
# Author: Beren Millidge, with substantial contributions from past Msc student, Florian Bolenz
# MSc Dissertation
# Summer 2017
import numpy as np
from experiment import DataProvider, Experiment, save_results, LearningRateContainer, LearningRateScheduler
from mask import MaskContainer
from create_data import *
import pickle
SEED = [1, 28, 102] # for experiment A
# SEED = [13, 11, 1415] # for experiment B
# SEED = [622, 622, 522] # for experiments C, E
# SEED = [3, 2121, 1978] # for experiments D
# SEED = [21, 04, 2017] # for experiment F
FILENAME = 'dat_0_1_20_nodupl.p'
#MASKCONTNAME = 'interintra_4m_20iu_16hu_20ou_fullin_fullout_withrec' # for experiment A:
MASKCONTNAME = 'interintra_4m_20iu_16hu_20ou_partialin_partialout_withrec' # for experiments B, C, D, F:
# MASKCONTNAME = 'interintra_2m_20iu_16hu_20ou_partialin_partialout_withrec' # for experiments F:
INPUT_LENGTH = 5 # time steps activity circulates through hidden layer
LEARNING_RATE = 0.005 # initial learning rate
N_EPOCHS = 2000 #1000
N_NETWORKS = 100 #200
OPTIMIZER = 'adagrad'
INIT_RANGE_HID = 0.05 # upper bound for weigh initialisation in hidden layer - #this is Q
TRAIN_ON_VALID = False # set true for switching training and validation data sets
SAVE_RESULTS = True
VERBOSE = False
#data characteristics:
BATCH_SIZE_TRAIN = 100
BATCH_SIZE_VALID = 100
BATCH_SIZE_TEST = 2
INPUT_UNITS = 20
INPUT_MIN = 0
INPUT_MAX = 1
COEFFICIENTS = [0, 0.3,0.7,-0.3,-0.7]
def main(data, maskpath, savepath, var = 0):
np.random.seed(SEED)
print("Loading data ...")
if var ==0:
data = [DataProvider(path='data/%s'%FILENAME, input_length=INPUT_LENGTH, train_on_valid=TRAIN_ON_VALID)]
maskcont = MaskContainer(path="masks/%s"%maskpath)
lrcont = LearningRateContainer(LearningRateScheduler(learning_rate_init=LEARNING_RATE))
print("Start experiment ...")
exp = Experiment(maskcont=maskcont, data=data)
res_train, res_valid = exp.run(N_NETWORKS, N_EPOCHS, lrcont=lrcont, optimizer=OPTIMIZER, init_range_hid=INIT_RANGE_HID, verbose=True)
if SAVE_RESULTS:
#try:
# save_results(res_train, res_valid, data, maskcont, lrcont, OPTIMIZER, INIT_RANGE_HID, SEED, savepath)
#print "Results saved."
# except:
#print 'Results could not be saved'
save_results(res_train, res_valid, data, maskcont, lrcont, OPTIMIZER, INIT_RANGE_HID, SEED, savepath)
def create_correlated_data(coefficients):
dataproviders = []
for coeff in coefficients:
data = build_correlated_sequence(INPUT_UNITS, BATCH_SIZE_TRAIN, BATCH_SIZE_VALID, BATCH_SIZE_TEST, coeff)
print("Creating data with coefficient %s"%coeff)
pickle.dump(data,open("data/dat_%s_%s_%s_coeff_%s.p"%(str(INPUT_MIN), str(INPUT_MAX), str(INPUT_UNITS), str(coeff)), 'wb'))
print("Data saved.")
provider = DataProvider(data, input_length = INPUT_LENGTH, train_on_valid = TRAIN_ON_VALID)
dataproviders.append(provider)
return dataproviders
def create_time_correlated(coefficients):
dataproviders = []
for coeff in coefficients:
data = build_correlated_across_time(INPUT_UNITS, BATCH_SIZE_TRAIN, BATCH_SIZE_VALID, BATCH_SIZE_TEST, coeff)
print("Creating data with coefficient %s"%coeff)
#pickle.dump(data,open("data/dat_%s_%s_%s_coeff_%s.p"%(str(INPUT_MIN), str(INPUT_MAX), str(INPUT_UNITS), str(coeff)), 'wb'))
#print("Data saved.")
provider = DataProvider(data, input_length = INPUT_LENGTH, train_on_valid = TRAIN_ON_VALID)
dataproviders.append(provider)
return dataproviders
def create_sized_data(coefficients, save = False):
dataproviders = []
for coeff in coefficients:
data = build_set_sized_sequence(INPUT_UNITS, BATCH_SIZE_TRAIN, BATCH_SIZE_VALID, BATCH_SIZE_TEST, coeff)
print("Creating data with coefficient %s"%coeff)
if save:
pickle.dump(data,open("data/dat_%s_%s_%s_coeff_%s.p"%(str(INPUT_MIN), str(INPUT_MAX), str(INPUT_UNITS), str(coeff)), 'wb'))
print("Data saved.")
provider = DataProvider(data, input_length = INPUT_LENGTH, train_on_valid = TRAIN_ON_VALID)
dataproviders.append(provider)
return dataproviders
if __name__ == '__main__':
#data = create_correlated_data(COEFFICIENTS)
#main(data,'interintra7_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra7_coeff1',1)
#main(data,'interintra6_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra6_coeff1',1)
#main(data,'interintra5_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra5_coeff1',1)
#main(data,'interintra4_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra4_coeff1',1)
#main(data,'interintra3_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra3_coeff1',1)
#main(data,'interintra2_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra2_coeff1',1)
#main(data,'interintra1_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra1_coeff1',1)
#data = create_time_correlated(COEFFICIENTS)
#main(data,'interintra7_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra7_seq1',1)
#main(data,'interintra6_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra6_seq1',1)
#main(data,'interintra5_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra5_seq1',1)
#main(data,'interintra4_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra4_seq1',1)
#main(data,'interintra3_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra3_seq1',1)
#main(data,'interintra2_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra2_seq1',1)
#main(data,'interintra1_4m_20iu_16hu_20ou_fullin_fullout_withrec', 'intra1_seq1',1)
#shuffled masks
#data = create_time_correlated(COEFFICIENTS)
#main(data,'interintra7_4m_20iu_16hu_20ou_fullin_fullout_withrec_SHUFFLE', 'intra7_seq1_shuffle',1)
#main(data,'interintra6_4m_20iu_16hu_20ou_fullin_fullout_withrec_SHUFFLE', 'intra6_seq1_shuffle',1)
#main(data,'interintra5_4m_20iu_16hu_20ou_fullin_fullout_withrec_SHUFFLE', 'intra5_seq1_shuffle',1)
#main(data,'interintra4_4m_20iu_16hu_20ou_fullin_fullout_withrec_SHUFFLE', 'intra4_seq1_shuffle',1)
#main(data,'interintra3_4m_20iu_16hu_20ou_fullin_fullout_withrec_SHUFFLE', 'intra3_seq1_shuffle',1)
#main(data,'interintra2_4m_20iu_16hu_20ou_fullin_fullout_withrec_SHUFFLE', 'intra2_seq1_shuffle',1)
#main(data,'interintra1_4m_20iu_16hu_20ou_fullin_fullout_withrec_SHUFFLE', 'intra1_seq1_shuffle',1)
#data = create_correlated_data(COEFFICIENTS)
#main(data,'interintra7_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra7_seq1_partial',1)
#main(data,'interintra6_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra6_seq1_partial',1)
#main(data,'interintra5_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra5_seq1_partial',1)
#main(data,'interintra4_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra4_seq1_partial',1)
#main(data,'interintra3_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra3_seq1_partial',1)
#main(data,'interintra2_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra2_seq1_partial',1)
#main(data,'interintra1_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra1_seq1_partial',1)
#data = create_time_correlated(COEFFICIENTS)
#main(data,'interintra7_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra7_seq_time_partial',1)
#main(data,'interintra6_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra6_seq_time_partial',1)
#main(data,'interintra5_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra5_seq_time_partial',1)
#main(data,'interintra4_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra4_seq_time_partial',1)
#main(data,'interintra3_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra3_seq_time_partial',1)
#main(data,'interintra2_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra2_seq_time_partial',1)
#main(data,'interintra1_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra1_seq_time_partial',1)
#data = create_correlated_data(COEFFICIENTS)
#main(data,'interintra7_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra7_seq1_partial_shuffle',1)
#main(data,'interintra6_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra6_seq1_partial_shuffle',1)
#main(data,'interintra5_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra5_seq1_partial_shuffle',1)
#main(data,'interintra4_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra4_seq1_partial_shuffle',1)
#main(data,'interintra3_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra3_seq1_partial_shuffle',1)
#main(data,'interintra2_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra2_seq1_partial_shuffle',1)
#main(data,'interintra1_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra1_seq1_partial_shuffle',1)
#data = create_correlated_data(COEFFICIENTS)
#main(data, 'interintra7_4m_20iu_16hu_20ou_partialin_partialout_EQUAL', 'CONNECTIONS_EQUAL', 1)
#COEFFS = [0.2,0.4,0.6,0.8,1]
#data = create_sized_data(COEFFS)
#main(data,'interintra7_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra7_sized1_partial',1)
#main(data,'interintra6_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra6_sized1_partial',1)
#main(data,'interintra5_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra5_sized1_partial',1)
#main(data,'interintra4_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra4_sized1_partial',1)
#main(data,'interintra3_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra3_sized1_partial',1)
#main(data,'interintra2_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra2_sized1_partial',1)
#main(data,'interintra1_4m_20iu_16hu_20ou_partialin_partialout_withrec', 'intra1_sized1_partial',1)
#RUN THIS TOMORROW!
#data = create_sized_data(COEFFS)
#main(data,'interintra7_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra7_sized1_partial_shuffle',1)
#main(data,'interintra6_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra6_sized1_partial_shuffle',1)
#main(data,'interintra5_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra5_sized1_partial_shuffle',1)
#main(data,'interintra4_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra4_sized1_partial_shuffle',1)
#main(data,'interintra3_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra3_sized1_partial_shuffle',1)
#main(data,'interintra2_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra2_sized1_partial_shuffle',1)
#main(data,'interintra1_4m_20iu_16hu_20ou_partialin_partialout_withrec_SHUFFLE', 'intra1_sized1_partial_shuffle',1)
#data = create_time_correlated(COEFFICIENTS)
#main(data, 'NULL_PARTIAL_MASK_TEST', 'NULL_MASK_TEST', 1)