-
Notifications
You must be signed in to change notification settings - Fork 5
/
test_recurrent.py
153 lines (140 loc) · 6.31 KB
/
test_recurrent.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
# -*- coding: utf-8 -*-
"""
This is python 3 code
main script for training/classifying
"""
if not '__file__' in vars(): __file__= u'C:/Users/Simon/dropbox/Uni/Masterthesis/AutoSleepScorer/main.py'
import os
import gc; gc.collect()
import matplotlib
matplotlib.use('Agg')
import numpy as np
import keras
import telegram_send
import tools
import scipy
import models
from copy import deepcopy
import pickle
import keras_utils
from keras_utils import cv
import sleeploader
import matplotlib; matplotlib.rcParams['figure.figsize'] = (10, 3)
np.random.seed(42)
if __name__ == "__main__":
try:
with open('count') as f:
counter = int(f.read())
except IOError:
print('No previous experiment?')
counter = 0
with open('count', 'w') as f:
f.write(str(counter+1))
#%%
if os.name == 'posix':
datadir = '.'
else:
# datadir = 'c:\\sleep\\data\\'
# datadir = 'd:\\sleep\\vinc\\'
# datadir = 'c:\\sleep\\emsa\\'
datadir = 'c:\\sleep\\cshs50\\'
# datadir = 'c:\\sleep\\edfx\\'
#
def load_data(tsinalis=False):
sleep = sleeploader.SleepDataset(datadir)
gc.collect()
# sleep.load()
# sleep.save_object()
sleep.load_object()
data, target, groups = sleep.get_all_data(groups=True)
data = tools.normalize(data)
target[target==4] = 3
target[target==5] = 4
target[target==8] = 0
target = keras.utils.to_categorical(target)
return deepcopy(data), deepcopy(target), deepcopy(groups)
data,target,groups = load_data()
datadir = 'c:\\sleep\\edfx\\'
# data2,target2,groups2 = load_data()
trans_tuple = [data,target,groups]
#%%
# print('Extracting features')
# target = np.load('target.npy')
# groups = np.load('groups.npy')
feats_eeg = np.load('feats_eeg.npy')# tools.feat_eeg(data[:,:,0])
feats_emg = np.load('feats_emg.npy')#tools.feat_emg(data[:,:,1])
feats_eog = np.load('feats_eog.npy')#tools.feat_eog(data[:,:,2])
feats_all = np.hstack([feats_eeg, feats_emg, feats_eog ])
feats_all = scipy.stats.stats.zscore(feats_all)
# #
if 'data' in vars():
if np.sum(np.isnan(data)) or np.sum(np.isnan(data)):print('Warning! NaNs detected')
#%%
results = dict()
comment = 'rnn_test'
print(comment)
print("starting at")
#%% model comparison
#r = dict()
#r['pure_rnn_5'] = cv(feats5, target5, groups5, models.pure_rnn, name='5',epochs=epochs, folds=10,batch_size=batch_size, counter=counter, plot=True, stop_after=35)
#r['pure_rnnx3_5'] = cv(feats5, target5, groups5, models.pure_rnn_3, name='5',epochs=epochs, folds=10,batch_size=batch_size, counter=counter, plot=True, stop_after=35)
#r['pure_rrn_do_5'] = cv(feats5, target5, groups5, models.pure_rnn_do, name='5',epochs=epochs, folds=10,batch_size=batch_size, counter=counter, plot=True, stop_after=35)
#r['ann_rrn_5'] = cv(feats5, target5, groups5, models.ann_rnn, name='5',epochs=epochs, folds=10,batch_size=batch_size, counter=counter, plot=True, stop_after=35)
#with open('results_recurrent_architectures.pkl', 'wb') as f:
# pickle.dump(r, f)
#%% seqlen
#r = dict()
##r['pure_rrn_do_1'] = cv(feats1, target1, groups1, models.pure_rnn_do, name='1',epochs=epochs, folds=10,batch_size=batch_size, counter=counter, plot=True, stop_after=35)
#for i in [1,2,3,4,5,6,7,8,9,10,15]:
# feats_seq, target_seq, group_seq = tools.to_sequences(feats, target, groups, seqlen = i, tolist=False)
# r['pure_rrn_do_' + str(i)] = cv(feats_seq, target_seq, group_seq, models.pure_rnn_do_stateful,
# name=str(i),epochs=epochs, folds=10, batch_size=batch_size,
# counter=counter, plot=True, stop_after=35)
##
#with open('results_recurrent_seqlen1-15.pkl', 'wb') as f:
# pickle.dump(r, f)
#%%
# %% seqlen = 6, 5-fold
# batch_size = 512
# feats_seq, target_seq, group_seq = tools.to_sequences(feats_all, target, groups=groups, seqlen = 6, tolist=False)
# r = keras_utils.cv(feats_seq, target_seq, group_seq, models.pure_rnn_do, epochs=250, folds=5, batch_size=batch_size, name='RNN',
# counter=counter, plot=True, stop_after=15, balanced=False)
# results.update(r)
# with open('results_recurrent', 'wb') as f:
# pickle.dump(results, f)
###
##
#%% seqlen = 6, 5-fold mit past
#r = dict()
#batch_size = 1440
#feats_seq, target_seq, group_seq = tools.to_sequences(feats, target, groups, seqlen = 6, tolist=False)
#r['pure_rrn_do_6_weighted-'] = cv(feats_seq, target_seq, group_seq, models.pure_rnn_do,
# name=str(6),epochs=300, folds=10, batch_size=batch_size,
# counter=counter, plot=True, stop_after=15, weighted=True)
#r['pure_rrn_do_6_not_weighted-'] = cv(feats_seq, target_seq, group_seq, models.pure_rnn_do,
# name=str(6),epochs=300, folds=10, batch_size=batch_size,
# counter=counter, plot=True, stop_after=15, weighted=False)
##
#with open('results_recurrent_seqlen-6-w5.pkl', 'wb') as f:
# pickle.dump(r, f)
#%%
#s
batch_size = 256
epochs = 250
name = 'LSTM moreL2'
###
rnn = {'model':models.pure_rnn_do, 'layers': ['fc1'], 'seqlen':6,
'epochs': 250, 'batch_size': 512, 'stop_after':15, 'balanced':False}
print(rnn)
data = data[:12000]
target = target[:12000]
groups = groups[:12000]
# model = 'C:\\Users\\Simon\\dropbox\\Uni\\Masterthesis\\AutoSleepScorer\\weights\\balanced'
model = models.cnn3adam_filter_morel2
r = keras_utils.cv (data, target, groups, model, rnn=rnn,trans_tuple=trans_tuple, name=name,
epochs=epochs, folds=5, batch_size=batch_size, counter=counter,
plot=True, stop_after=15, balanced=False, cropsize=2800)
results.update(r)
with open('results_recurrejsuttestingnt_morel2.pkl', 'wb') as f:
pickle.dump(results, f)
telegram_send.send(parse_mode='Markdown',messages=['DONE {} {}\n```\n{}\n```\n'.format(os.path.basename(__file__),name, tools.print_string(results))])