forked from iitis/qcontrol_lstm_approx
-
Notifications
You must be signed in to change notification settings - Fork 0
/
lstm_as_approximation.py
295 lines (244 loc) · 11.3 KB
/
lstm_as_approximation.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
import os
os.environ["CUDA_VISIBLE_DEVICES"]="-1"
import tensorflow as tf
import numpy as np
import scipy
import scipy.io
from itertools import product as prod
from tensorflow.python.client import timeline
from get_data import *
import pathlib
from noise_models_and_integration import *
from architecture import *
from constants_of_experiments import *
def variation_acc2_local_disturb(sess,
network,
x_,
keep_prob,
saver,
noise_name,
gamma,
alpha,
controls_nb,
test_input,
test_target,
n_ts,
evo_time,
eps,
accept_err):
# restoring saved model
saver.restore(sess, "weights/dim_{}/{}/gam_{}_alfa_{}.ckpt".format(model_dim, noise_name, gamma, alpha))
# initializoing resulting tensor, first two dimensions corresponds to coordinate which will be disturbed, on the last dimension, there will be added variation of outputs
results = np.zeros((n_ts, controls_nb, len(np.array(test_input))))
print(len(test_input))
print(np.shape(results))
iter = -1
for sample_nb in range(len(np.array(test_input))):
# taking sample NCP
origin_NCP = test_input[sample_nb]
# taking target superoperator corresponding to the NCP
origin_superoperator = test_target[sample_nb]
tf_result = False
# calculating nnDCP corresponding to input NCP
pred_DCP = get_prediction(sess, network, x_, keep_prob, np.reshape(origin_NCP, [1, n_ts, controls_nb]))
# calculating superoperator from nnDCP
sup_from_pred_DCP = integrate_lind(pred_DCP[0], (alpha, gamma), n_ts, evo_time, noise_name, tf_result)
print("sanity check")
acceptable_error = fidelity_err([origin_superoperator, sup_from_pred_DCP], dim, tf_result)
print("predicted DCP", acceptable_error)
print("---------------------------------")
############################################################################################################
#if sanity test is above assumed error then the experiment is performed
if acceptable_error <= accept_err:
iter += 1
# iteration over all coordinates
for (t, c) in prod(range(n_ts), range(controls_nb)):
new_NCP = origin_NCP
if new_NCP[t, c] < (1 - eps):
new_NCP[t, c] += eps
else:
new_NCP[t, c] -= eps
sup_from_new_NCP = integrate_lind(new_NCP, (alpha, 0.), n_ts, evo_time, noise_name, tf_result)
new_DCP = get_prediction(sess, network, x_, keep_prob,
np.reshape(new_NCP, [1, n_ts, controls_nb]))
sup_form_new_DCP = integrate_lind(new_DCP[0], (alpha, gamma), n_ts, evo_time, noise_name, tf_result)
error = fidelity_err([sup_from_new_NCP, sup_form_new_DCP], dim, tf_result)
#print(error)
# if predicted nnDCP gives wrong superopertaor, then we add not variation of output, but some label
if error <= accept_err:
results[t, c, iter] = np.linalg.norm(pred_DCP - new_DCP)
else:
results[t, c, iter] = -1
print(iter)
print(np.shape(results))
return results
def experiment_loc_disturb(n_ts,
gamma,
alpha,
evo_time,
supeop_size,
controls_nb,
train_set_size,
test_set_size,
size_of_lrs,
noise_name,
model_dim,
eps,
accept_err):
###########################################
# PLACEHOLDERS
###########################################
# input placeholder
x_ = tf.placeholder(tf.float32, [None, n_ts, controls_nb])
# output placeholder
y_ = tf.placeholder(tf.complex128, [None, supeop_size, supeop_size])
# dropout placeholder
keep_prob = tf.placeholder(tf.float32)
# creating the graph
network = my_lstm(x_, controls_nb, size_of_lrs, keep_prob)
# instance for saving the model
saver = tf.train.Saver()
# loading the data
(_, _, test_input, test_target) = get_data(train_set_size, test_set_size, model_dim)
# maintaining the memory
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
# essential function which executes the experiment
result = variation_acc2_local_disturb(sess,
network,
x_,
keep_prob,
saver,
noise_name,
gamma,
alpha,
controls_nb,
test_input,
test_target,
n_ts,
evo_time,
eps,
accept_err)
sess.close()
tf.reset_default_graph()
return result
def train_and_predict(n_ts,
gamma,
alpha,
evo_time,
batch_size,
supeop_size,
controls_nb,
nb_epochs,
learning_rate,
train_set_size,
test_set_size,
size_of_lrs,
dim,
noise_name,
model_dim):
###########################################
# PLACEHOLDERS
###########################################
# input placeholder
x_ = tf.placeholder(tf.float32, [None, n_ts, controls_nb])
# output placeholder
y_ = tf.placeholder(tf.complex128, [None, supeop_size, supeop_size])
# dropout placeholder
keep_prob = tf.placeholder(tf.float32)
# creating the graph
network = my_lstm(x_,controls_nb, size_of_lrs, keep_prob)
# instance for saving the model
saver = tf.train.Saver()
# loading the data
(train_input, train_target, test_input, test_target) = get_data(train_set_size,
test_set_size,
model_dim)
# maintaining the memory
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
# training the network
acc = fit(sess,
network,
x_,
y_,
keep_prob,
train_input,
train_target,
test_input,
test_target,
nb_epochs,
batch_size,
train_set_size,
learning_rate,
gamma,
alpha,
n_ts,
evo_time,
dim,
noise_name)
# making prediction by trained model
pred = get_prediction(sess, network, x_, keep_prob, test_input)
# saving trained model
saver.save(sess, "weights/dim_{}/{}/gam_{}_alfa_{}.ckpt".format(model_dim, noise_name, gamma, alpha))
sess.close()
tf.reset_default_graph()
return (pred,acc)
# ---------------------------------------------------------------------------
if __name__ == "__main__":
# prepare dirs for the output files
pathlib.Path("weights/dim_{}/{}".format(model_dim, noise_name)).mkdir(parents=True, exist_ok=True)
# Note: change the below value if you have already trained the network
# train_model = True
if testing_effectiveness:
pathlib.Path("results/eff_fid_lstm/dim_{}".format(model_dim)).mkdir(parents=True, exist_ok=True)
# main functionality
for gamma in list_gammas:
for alpha in list_alphas:
statistic = dict()
for i in range(10):
pred, acc = train_and_predict(n_ts,
gamma,
alpha,
evo_time,
batch_size,
supeop_size,
controls_nb,
nb_epochs,
learning_rate,
train_set_size,
test_set_size,
size_of_lrs,
dim,
noise_name,
model_dim)
statistic[i] = acc
# save the results
np.savez("results/eff_fid_lstm/dim_{}/statistic_{}_gam_{}_alpha_{}".format(model_dim,
noise_name,
gamma,
alpha), statistic)
else:
eps = 10**(-eps_order)
# main functionality
data = experiment_loc_disturb(n_ts,
gamma,
alpha,
evo_time,
supeop_size,
controls_nb,
train_set_size,
test_set_size,
size_of_lrs,
noise_name,
model_dim,
eps,
accept_err)
pathlib.Path("results/NN_as_approx/dim_{}".format(model_dim)).mkdir(parents=True, exist_ok=True)
np.savez("results/NN_as_approx/dim_{}/{}_gam_{}_alpha_{}_epsilon_1e-{}".format(model_dim,
noise_name,
gamma,
alpha,
eps_order), data)