-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
166 lines (137 loc) · 6.36 KB
/
train.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
import os
import pandas as pd
import numpy as np
import datetime
import time
import re
from sklearn.metrics import classification_report, confusion_matrix
from sklearn.metrics import precision_recall_fscore_support
from sklearn.metrics import f1_score
from sklearn.metrics import make_scorer
from sklearn.metrics import accuracy_score
from keras import backend as K
from keras.models import load_model
from statistics import mode
from keras.callbacks import Callback, EarlyStopping, ModelCheckpoint
class Train():
def __init__(self,model,model_name,summary,x_cols,modeldir='./models/'):
self.x_cols=x_cols
self.model=model
self.summary=summary
self.modeldir=modeldir
self.model_name=model_name
def train(self,data,model,path,total_epochs=50,batch_size=1000,monitor_score='val_f1',monitor_mode='auto',patience=5):#TOODO Change this so that data_seqs are in the constructor
self.monitor_score=monitor_score
self.monitor_mode=monitor_mode
self.total_epochs=total_epochs
self.batch_size=batch_size
self.patience=patience
self.path=path
val=False
X_train=data['train']['xarr']
y_train=data['train']['yarr']
try:
X_val=data['val']['xarr']
y_val=data['val']['yarr']
val=True
except KeyError:
pass
filepath=self.path+"_e{epoch:02d}_"+monitor_score+"-{"+monitor_score+":.2f}.hdf5"
#filepath=self.path+"_e{epoch:02d}_acc-{val_acc:.2f}.hdf5"
checkpoint = ModelCheckpoint(os.path.join(self.modeldir,filepath),
monitor=monitor_score,
verbose=1,
save_best_only=True,
mode=monitor_mode)
earlystop = EarlyStopping(monitor=monitor_score, min_delta=0.00005, patience=patience, mode=monitor_mode)
starttime=time.time()
if val==True:
model.fit(X_train,y_train,
validation_data=(X_val,y_val),
batch_size=batch_size,
epochs=total_epochs,
verbose=1,
callbacks=[earlystop,checkpoint])
else:
model.fit(X_train,y_train,
batch_size=batch_size,
epochs=total_epochs,
verbose=1,
callbacks=[earlystop,checkpoint])
endtime=time.time()
train_time=endtime-starttime
self.train_time=train_time/60
def get_best_model(self):
model_filenames=[fn for fn in os.listdir(self.modeldir) if fn.startswith(self.path)]
model_filenames.sort()
if len(model_filenames)==1:
best_fn=model_filenames[0]
else:
remove_filenames=model_filenames[:-1]
best_fn=model_filenames[-1]
for fn in remove_filenames:
os.remove(os.path.join(self.modeldir,fn))
self.best_fn=best_fn
#TODO finish writing metrics etc to dataframe
def evaluate(self,data,y_map,tracker_fn='tracker.csv',channel_probs=False,predict_english_only=False):
print(channel_probs)
best_epochs=int(re.findall('_e(\d+)_',self.best_fn)[0])
try:
model = load_model(os.path.join(self.modeldir,self.best_fn))
except:
model = load_model(os.path.join(self.modeldir,self.best_fn),custom_objects={'f1': ut.f1})
y_decoder={v: k for k, v in y_map.items()}
X_test=data['test']['xarr']
y_test=data['test']['yarr']
y_test_labels=[np.argmax(cat) for cat in y_test]
#predict
#y_pred = model.predict(data.X_test_seq,batch_size=self.batch_size)
if channel_probs==True:
all_lang_preds=[]
for arr in X_test:
y_pred = model.predict(arr,batch_size=self.batch_size)
#y_pred_labels=[np.argmax(pred) for pred in y_pred]
all_lang_preds.append(y_pred)
#all_lang_preds.append(y_pred_labels)
y_pred_labels=[]
for i,probs in enumerate(zip(*all_lang_preds)):
y_pred_labels.append(np.argmax(np.average(np.array(probs),axis=0)))
elif predict_english_only==True:
y_pred = model.predict(X_test[0],batch_size=self.batch_size)
y_pred_labels=[np.argmax(pred) for pred in y_pred]
else:
y_pred = model.predict(X_test,batch_size=self.batch_size)
y_pred_labels=[np.argmax(pred) for pred in y_pred]
#evaluate
conf_matrix=confusion_matrix(y_test_labels, y_pred_labels)
acc=accuracy_score(y_test_labels, y_pred_labels)
w_prec,w_rec,w_f1,_=precision_recall_fscore_support(y_test_labels, y_pred_labels,
average='weighted')
mic_prec,mic_rec,mic_f1,_=precision_recall_fscore_support(y_test_labels, y_pred_labels,
average='micro')
mac_prec,mac_rec,mac_f1,_=precision_recall_fscore_support(y_test_labels, y_pred_labels,
average='macro')
results={'model_filepath':self.best_fn,
'languages':self.x_cols,
'model_name':self.model_name,
'date':'-',
'time':'-',
'train_time':self.train_time,
'best_epochs':best_epochs,
'accuracy':acc,
'batch_size':self.batch_size,
'total_epochs': self.total_epochs,
'summary':self.summary,
'precision (weighted)':w_prec,
'f1 (weighted)':w_f1,
'precision (micro)':mic_prec,
'recall (micro)':mic_rec,
'f1 (micro)':mic_f1,
'precision (macro)':mac_prec,
'recall (macro)':mac_rec,
'f1 (macro)':mac_f1,
'confusion_matrix':conf_matrix}
tracker=pd.read_csv(os.path.join(self.modeldir,tracker_fn))
tracker=tracker.append(results, ignore_index=True)
tracker.to_csv(os.path.join(self.modeldir,tracker_fn),index=False)
#TODO finish writing metrics etc to dataframe