-
Notifications
You must be signed in to change notification settings - Fork 0
/
executor.py
216 lines (212 loc) · 8.81 KB
/
executor.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
import logging
import keras
from consts import stop,nparray,make_net,determe_X_Y,push_obj,predict,push_i,push_str, \
plot_train,sav_model_wei,sav_model,fit_net,get_mult_class_matr,cr_callback_wi_loss_treshold_and_acc_shure, \
evalu_,load_model_wei,cr_sav_model_wei_best_callback,compile_net,get_weis,k_summary,push_fl,make_net_on_contrary
from keras.models import Sequential, model_from_json
from keras.layers import Dense
from util import matr_img
import numpy as np
from keras.callbacks import History, ModelCheckpoint
from plot_history import plot_history_
from my_keras_customs import My_subcl_model_checkpoint
"""
Замечания: для Python3 с f строками (используются в логинге)
"""
def exec(buffer:tuple,logger:logging.Logger, date:str)->None:
"""
Исполнитель байт-кода (управляющих команд)
:param buffer: буфер команд
:param logger: обьект логер
:param date: сегодняшняя дата
:return: None
"""
X_t=None
Y_t=None
len_=256
save_wei_best_callback:ModelCheckpoint=None
threshold_callback=None
model_obj: keras.models.Model = None
history:History=None
l_weis=[]
logger.info(logger.debug(f'Log started {date}'))
vm_is_running=True
ip=0
sp=-1
steck=[0]*len_
op=buffer[ip]
while vm_is_running:
#------------основные коды памяти---------------
if op==stop:
return
if op == push_i:
sp += 1
ip += 1
steck[sp] = int(buffer[ip])
elif op == push_fl:
sp += 1
ip += 1
steck[sp] = float(buffer[ip])
elif op == push_str:
sp += 1
ip += 1
steck[sp] = buffer[ip]
elif op== push_obj:
sp+=1
ip+=1
steck[sp]=buffer[ip]
#------------------------------------
elif op==determe_X_Y:
Y_t=steck[sp]
sp-=1
X_t=steck[sp]
sp-=1
elif op==nparray:
st_arg=steck[sp]
sp-=1
sp+=1
steck[sp]=np.array(st_arg)
elif op==predict:
out_nn=model_obj.predict(X_t)
print("Predict matr: ",out_nn)
logger.info(f"Predict matr: {out_nn}")
elif op==evalu_:
out_ev=model_obj.evaluate(X_t, Y_t)
print("Eval model: ",out_ev)
logger.info(f"Eval model: {out_ev}")
elif op==sav_model:
with open("model_json.json", 'w') as f:
f.write(model_obj.to_json())
print("Model saved")
logger.info('Model saved')
elif op==sav_model_wei:
model_obj.save_weights("wei.h5", overwrite=True)
print("Weights saved")
logger.info('Weights saved')
elif op==load_model_wei:
loaded_json_model=''
with open('model_json.json','r') as f:
loaded_json_model=f.read()
model_obj=model_from_json(loaded_json_model)
model_obj.load_weights('wei.h5')
print("Loaded model and weights")
logger.info("Loaded model and weights")
elif op==get_weis:
for i in range(len(model_obj.layers)):
l_weis.append(model_obj.layers[i].get_weights())
elif op==get_mult_class_matr:
pix_am=steck[sp]
sp-=1
path_=steck[sp]
sp-=1
X_t, Y_t=matr_img(path_,pix_am)
X_t=np.array(X_t)
Y_t=np.array(Y_t)
X_t.astype('float32')
Y_t.astype('float32')
X_t/=255
elif op==make_net:
l_tmp = None
acts_di:dict=None
acts_di={'s':'sigmoid','r':'relu','t':'tanh','S':'softmax'}
use_bias_ = False
ip += 1
arg = buffer[ip]
type_m, denses, inps, acts, use_bi, kern_init = arg
if type_m == 'S':
model_obj = Sequential()
for i in range(len(denses)):
if denses[i] == 'D':
splt_bi = use_bi[i].split('_')
if splt_bi[-1] == '1':
use_bias_ = True
elif splt_bi[-1] == '0':
use_bias_ = False
# my_kern_reg=regularizers.l1(0.0001)
my_kern_reg=None
if i == 0:
l_tmp = Dense(inps[i + 1], input_dim=inps[0], activation=acts_di.get(acts[i]), use_bias=use_bias_,
trainable=True, kernel_initializer=kern_init, kernel_regularizer=my_kern_reg)
else:
l_tmp = Dense(inps[i + 1],input_dim=inps[i],activation=acts_di.get(acts[i]), use_bias=use_bias_,
trainable=True, kernel_initializer=kern_init, kernel_regularizer=my_kern_reg)
model_obj.add(l_tmp)
elif op==make_net_on_contrary:
l_tmp = None
acts_di: dict = None
acts_di = {'s': 'sigmoid', 'r': 'relu', 't': 'tanh', 'S': 'softmax'}
use_bias_ = False
ip += 1
arg = buffer[ip]
type_m, denses, inps, acts, use_bi, kern_init = arg
if type_m == 'S':
model_obj = Sequential()
for i in range(len(denses)-1,-1,-1):
if denses[i] == 'D':
splt_bi = use_bi[i].split('_')
if splt_bi[-1] == '1':
use_bias_ = True
elif splt_bi[-1] == '0':
use_bias_ = False
if i==len(denses)-1:
l_tmp = Dense(inps[i],input_dim=inps[i+1], activation=acts_di.get(acts[i]),
use_bias=use_bias_,
kernel_initializer=kern_init)
l_tmp.build((None, inps[i+1]))
else:
l_tmp = Dense(inps[i],activation=acts_di.get(acts[i]), use_bias=use_bias_,
kernel_initializer=kern_init)
l_tmp.build((None, inps[i+1]))
if use_bias_:
# Only if we have biases
wei_t=l_weis[i]
ke,bi=wei_t
bi_n=np.zeros(inps[i])+bi[0]
l_tmp.set_weights([ke.T, bi_n])
else:
raise RuntimeError("Without biases on-contrary net not implemented")
model_obj.add(l_tmp)
print("On-contrary net created")
logger.info("On-contrary net created")
elif op==k_summary:
model_obj.summary()
elif op==plot_train:
ip+=1
arg=buffer[ip]
plot_history_('./graphic/train_graphic.png', history, arg, logger)
elif op==compile_net:
ip+=1
arg=buffer[ip]
opt, loss_obj, metrics=arg
model_obj.compile(optimizer=opt, loss=loss_obj, metrics=metrics)
elif op==fit_net: # 1-ep 2-bach_size 3-validation_split 4-shuffle 5-callbacks
ip+=1
arg=buffer[ip]
ep,ba_size,val_spl,shuffle,callbacks=arg
if save_wei_best_callback:
callbacks.append(save_wei_best_callback)
if threshold_callback:
callbacks.append(threshold_callback)
history=model_obj.fit(X_t, Y_t, epochs=ep, batch_size=ba_size, validation_split=val_spl, shuffle=shuffle, callbacks=callbacks)
elif op == cr_sav_model_wei_best_callback:
wei_file = 'wei.h5'
monitor='<uninitialize>'
save_best_only=True
ip+=1
arg=buffer[ip]
monitor=arg
save_wei_best_callback=ModelCheckpoint(wei_file, monitor,save_best_only=True, period=1, verbose=1, save_weights_only=True)
elif op == cr_callback_wi_loss_treshold_and_acc_shure:
wei_file = 'wei.h5'
ip+=1
arg=buffer[ip]
loss_threshold,acc_shureness=arg
threshold_callback=My_subcl_model_checkpoint(loss_threshold, acc_shureness, wei_file, logger)
else:
raise RuntimeError("Unknown bytecode -> %d."%op)
ip+=1
try:
op=buffer[ip]
except IndexError:
raise RuntimeError('It seems somewhere'
' skipped argument of bytecode.')