-
Notifications
You must be signed in to change notification settings - Fork 144
/
receptive.py
executable file
·237 lines (167 loc) · 9.18 KB
/
receptive.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
import numpy as np
import sklearn
import pandas as pd
import sys
from classifiers.inception import Classifier_INCEPTION
import subprocess
import tensorflow as tf
from keras.backend.tensorflow_backend import set_session
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.2
set_session(tf.Session(config=config))
from utils.utils import transform_labels
from utils.utils import create_synthetic_dataset
from utils.utils import create_directory
from utils.utils import check_if_file_exits
from utils.utils import generate_array_of_colors
import matplotlib
import matplotlib.pyplot as plt
root_dir = '/b/home/uha/hfawaz-datas/temp-dl-tsc/'
root_output_directory = root_dir + 'receptive-field/exp/'
root_output_directory_df = root_dir + 'receptive-field/'
create_directory(root_output_directory_df)
BATCH_SIZE = 128
NB_EPOCHS = 500
pattern_lens = [[0.1]]
pattern_poss = [[0.1, 0.65]]
# uncomment the following to experiment with the number of classes in a dataset
# pattern_poss = [
# [0.1, 0.65],
# [0.1, 0.3, 0.65],
# [0.1, 0.3, 0.5, 0.65],
# [0.1, 0.3, 0.5, 0.7, 0.9],
# [0.1, 0.2, 0.3, 0.4, 0.5, 0.65],
# [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7],
# [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8],
# [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]
# ]
kernel_sizes = np.arange(start=6, stop=15, step=2 * 4) # default 41
depths = np.arange(start=1, stop=10, step=1) # default 6
filters_s = [1 * (2 ** i) for i in range(8)] # default 32
ts_lens = [16 * (2 ** i) for i in range(7)] # default 256
ts_ns = [6 * (2 ** i) for i in range(8)] # default 128
use_residuals = [True, False] # default true
use_bottlenecks = [True, False] # default true
def convert_to_float(s):
s = s.split('_')[0]
return float(s)
subprocess.call(['./receptive_field_remove_non_completed.sh'],
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
if len(sys.argv) == 1:
matplotlib.use('agg')
out_df = root_output_directory_df + 'df_res_sub_0.csv'
idx_out = 1
while check_if_file_exits(out_df):
out_df = root_output_directory_df + 'df_res_sub_' + str(idx_out) + '.csv'
idx_out += 1
columns = ['pattern_len', 'pattern_pos', 'ts_len', 'ts_n', 'nb_classes',
'filters', 'kernel_size', 'depth', 'use_residual', 'use_bottleneck', 'accuracy']
df_results = pd.DataFrame(index=[], columns=columns)
df_results.to_csv(out_df)
curr_idx = 0
for pattern_len_ in pattern_lens:
for pattern_pos in pattern_poss:
pattern_len = pattern_len_
nb_classes = len(pattern_len) * len(pattern_pos)
for ts_len in ts_lens:
for ts_n in ts_ns:
x_train, y_train, x_test, y_test = create_synthetic_dataset(pattern_len=pattern_len,
pattern_pos=pattern_pos,
ts_len=ts_len, ts_n=ts_n)
# make the min to zero of labels
y_train, y_test = transform_labels(y_train, y_test)
# save orignal y because later we will use binary
y_true = y_test.astype(np.int64)
y_true_train = y_train.astype(np.int64)
# transform the labels from integers to one hot vectors
enc = sklearn.preprocessing.OneHotEncoder()
enc.fit(np.concatenate((y_train, y_test), axis=0).reshape(-1, 1))
y_train = enc.transform(y_train.reshape(-1, 1)).toarray()
y_test = enc.transform(y_test.reshape(-1, 1)).toarray()
if len(x_train.shape) == 2:
# if uni-variate add a dimension to make it multivariate with one dimension
x_train = x_train.reshape((x_train.shape[0], x_train.shape[1], 1))
x_test = x_test.reshape((x_test.shape[0], x_test.shape[1], 1))
input_shape = x_train.shape[1:]
for kernel_size in kernel_sizes:
for depth in depths:
for filters in filters_s:
for use_residual in use_residuals:
for use_bottleneck in use_bottlenecks:
output_directory = root_output_directory + 'pattern_len_' + ''.join(
[str(e) + '_' for e in pattern_len]) + '/'
output_directory = output_directory + 'pattern_pos_' + ''.join(
[str(e) + '_' for e in pattern_pos]) + '/'
output_directory = output_directory + 'ts_len_' + str(ts_len) + '/'
output_directory = output_directory + 'ts_n_' + str(ts_n) + '/'
output_directory = output_directory + 'nb_classes_' + str(nb_classes) + '/'
output_directory = output_directory + 'filters_' + str(filters) + '/'
output_directory = output_directory + 'kernel_size_' + str(kernel_size) + '/'
output_directory = output_directory + 'depth_' + str(depth) + '/'
output_directory = output_directory + 'use_residual_' + str(use_residual) + '/'
output_directory = output_directory + 'use_bottleneck_' + str(
use_bottleneck) + '/'
test_dir = create_directory(output_directory)
if test_dir is None:
# job already done / is being done
continue
# create the classifier
clf = Classifier_INCEPTION(output_directory, input_shape, nb_classes,
nb_filters=filters, use_residual=use_residual,
use_bottleneck=use_bottleneck, depth=depth,
kernel_size=int(kernel_size), verbose=False,
batch_size=BATCH_SIZE, nb_epochs=NB_EPOCHS)
# train
df_metrics = clf.fit(x_train, y_train, x_test, y_test, y_true)
acc = df_metrics['accuracy'].values[0]
df_results.loc[curr_idx] = [''.join([str(e) + '_' for e in pattern_len]),
''.join([str(e) + '_' for e in pattern_pos]),
ts_len, ts_n,
nb_classes,
filters, kernel_size, depth, use_residual,
use_bottleneck,
acc]
curr_idx += 1
df_results.to_csv(out_df)
print('DONE')
elif sys.argv[1] == 'plot_results':
matplotlib.use('pdf')
plot_3d = False
# read results
df_results = pd.read_csv(root_output_directory_df + 'df_res_sub_0.csv', index_col=0)
out_df = root_output_directory_df + 'df_res_sub_1.csv'
out_df_idx = 2
while check_if_file_exits(out_df):
df_results_temp = pd.read_csv(out_df, index_col=0)
df_results = pd.concat([df_results, df_results_temp], sort=True)
out_df = root_output_directory_df + 'df_res_sub_' + str(out_df_idx) + '.csv'
out_df_idx += 1
df_results.reset_index(inplace=True)
split_ons = ['use_bottleneck', 'use_residual', 'ts_n',
'filters', 'nb_classes']
split_values = [True, True, 128, 32, 2]
x_label = 'ts_len'
y_label = 'RF'
z_label = 'accuracy'
# add the receptive field value
df_results['RF'] = (df_results['kernel_size'] - 1) * df_results['depth'] + 1
curr_df = df_results
for i in range(len(split_values)):
split_on = split_ons[i]
split_value = split_values[i]
curr_df = curr_df.loc[df_results[split_on] == split_value]
x_labels = pd.unique(curr_df[x_label])
colors = generate_array_of_colors(len(x_labels))
fig = plt.figure()
i = 0
for xx in x_labels:
curr_df_p = curr_df.loc[curr_df[x_label] == xx].sort_values([y_label])
ts = pd.Series(index=curr_df_p[y_label], data=curr_df_p[z_label].values)
ts = ts.rolling(window=3, min_periods=1).mean()
ts.plot(color=colors[i], label=x_label + str(xx))
i += 1
plt.xlabel(y_label)
plt.ylabel(z_label)
plt.legend(loc='best')
plt.savefig(root_output_directory_df + 'out.pdf')
plt.close()