-
Notifications
You must be signed in to change notification settings - Fork 0
/
07-training.py
165 lines (149 loc) · 6.52 KB
/
07-training.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
import time, pickle, matplotlib.pyplot as plt
from sklearn.linear_model import Perceptron
from sklearn.linear_model import SGDClassifier
from sklearn.linear_model import PassiveAggressiveClassifier
from sklearn.externals import joblib
start_all = time.time()
def get_seconds(start):
end = time.time()
elapsed = str(round(end - start))
return elapsed
def log(line):
f = open('./training.log', 'a')
f.write(line + '\n')
f.close()
print(line)
### Treinamento
batch_size = 1000
n_pairs = 516000
train_set_size = 510000
n_batches = int(train_set_size / batch_size)
classes = [0, 1]
classifiers = [
("clfP", Perceptron(penalty = None)),
("clfPL1", Perceptron(penalty = 'l1')),
("clfSGDCL1", SGDClassifier(learning_rate = 'constant', penalty = 'l1', loss = 'hinge', eta0 = 1)),
("clfSGDO", SGDClassifier(learning_rate = 'optimal', penalty = 'l2', loss = 'hinge')), # SGD()
("clfSGDOL1", SGDClassifier(learning_rate = 'optimal', penalty = 'l1', loss = 'hinge')),
("clfSGDOLOG", SGDClassifier(learning_rate = 'optimal', penalty = 'l2', loss = 'log', eta0 = 1)),
("clfPAH", PassiveAggressiveClassifier(loss = 'hinge')),
("clfPASH", PassiveAggressiveClassifier(loss = 'squared_hinge'))
# Removidos:
#("clfPL2", Perceptron(penalty = 'l2')),
#("clfSGDC", SGDClassifier(learning_rate = 'constant', penalty = 'l2', loss = 'hinge', eta0 = 1)),
#("clfSGDCLOG", SGDClassifier(learning_rate = 'constant', penalty = 'l2', loss = 'log', eta0 = 1)),
]
log('n_pairs: ' + str(n_pairs) + ' train_set_size: ' + str(train_set_size) + ' n_batches: ' + str(n_batches) + ' batch_size: ' + str(batch_size))
for j in range(n_batches):
start_batch = time.time()
X = pickle.load(open('.\Batches\X.' + str(j), 'rb'))
y = pickle.load(open('.\Batches\y.' + str(j), 'rb'))
for name, clf in classifiers:
clf.partial_fit(X, y, classes)
joblib.dump(clf, '.\Classificadores\\' + name + '\\' + str(j) + '.clf', compress = 3)
log('[' + get_seconds(start_all) + ' | ' + str(round(100 * (j + 1) / n_batches)) + '%] - Batch ' + str(j + 1) + ' processada em ' + get_seconds(start_batch) + ' segundos')
log(str(n_batches) + ' batches processadas em ' + get_seconds(start_all) + ' segundos')
### Testes
log('Iniciando os testes.')
n_batches_test = int(n_pairs / batch_size) - n_batches
results = {}
# Inicializando os resultados em 0
for clf_name, clf in classifiers:
results[clf_name] = []
for i in range(n_batches):
results[clf_name].append({
'tp': 0,
'tn': 0,
'fp': 0,
'fn': 0,
'i': i
})
for j in range(n_batches, int(n_pairs / batch_size)):
start_batch = time.time()
Xt = pickle.load(open('.\Batches\X.' + str(j), 'rb'))
yt = pickle.load(open('.\Batches\y.'+ str(j), 'rb'))
# Para cada classificador
for clf_name, clf in classifiers:
path = '.\Classificadores\\' + clf_name + '\\'
# Para cada versão do classificador
for i in range(n_batches):
clf = joblib.load(path + str(i) + '.clf')
yp = clf.predict(Xt)
tp = results[clf_name][i]['tp']
tn = results[clf_name][i]['tn']
fp = results[clf_name][i]['fp']
fn = results[clf_name][i]['fn']
for k in range(len(yt)):
if yt[k] == yp[k]:
if yt[k] == 0:
tn = tn + 1
else:
tp = tp + 1
else:
if yt[k] == 0:
fp = fp + 1
else:
fn = fn + 1
results[clf_name][i]['tp'] = tp
results[clf_name][i]['tn'] = tn
results[clf_name][i]['fp'] = fp
results[clf_name][i]['fn'] = fn
if(tp + fp != 0):
results[clf_name][i]['precision'] = tp/(tp+fp)
else:
results[clf_name][i]['precision'] = -1
if(tp+fn != 0):
results[clf_name][i]['recall'] = tp/(tp+fn)
else:
results[clf_name][i]['recall'] = -1
results[clf_name][i]['mean_accuracy'] = (tp+tn)/(tp+tn+fp+fn)
# Salva os resultados em um arquivo
save_path = '.\Classificadores\\'
pickle.dump(results, open(save_path + 'results.pkl', 'wb'))
completion = str( round( 100 * (j - n_batches + 1) / (int(n_pairs / batch_size) - n_batches) ) ) + '%'
log('[' + completion + ' - ' + get_seconds(start_all) + ' ' + get_seconds(start_batch) + '] - Batch ' + str(j + 1) + ' processada')
log('Classificadores finais:')
for clf_name, clf in classifiers:
log(clf_name + ' - ' + str(results[clf_name][-1]))
classifiers = [
('clfP', 'Perceptron', 'r-'),
('clfPL1', 'Perceptron L1', '#d83e91'),
('clfSGDCL1', 'SGD L1 hinge Const', 'k--'),
('clfSGDO', 'SGD L2 hinge Opti', 'g-'),
('clfSGDOL1', 'SGD L1 hinge Opti', 'g--'),
('clfSGDOLOG', 'SGD L2 log Opti', 'y--'),
('clfPAH', 'PA hinge', 'b--'),
('clfPASH', 'PA squared_hinge', '#958bf3')
]
# Plota os resultados em um gráfico
example_axis = [x*2000 for x in range(1, n_batches + 1)]
plt.figure(figsize=(30, 21), dpi=80)
plt.title('Acurácia média em função do número de exemplos')
plt.xlabel('Exemplos')
plt.ylabel('Acurácia')
plt.grid(True)
for clf_name, clf_desc, line_color in classifiers:
plt.plot(example_axis, [y['mean_accuracy'] for y in results[clf_name]], line_color, label = clf_desc)
plt.axis([example_axis[0], example_axis[-1], 0, 1])
plt.legend(loc='center left', bbox_to_anchor=(1, 0.5))
plt.savefig('resultadosAcurácia.png', bbox_inches='tight')
plt.figure(figsize=(30, 21), dpi=80)
plt.title('Revocação em função do número de exemplos')
plt.xlabel('Exemplos')
plt.ylabel('Revocação')
plt.grid(True)
for clf_name, clf_desc, line_color in classifiers:
plt.plot(example_axis, [y['recall'] for y in results[clf_name]], line_color, label = clf_desc)
plt.axis([example_axis[0], example_axis[-1], 0, 1])
plt.legend(loc='center left', bbox_to_anchor=(1, 0.5))
plt.savefig('resultadosRevocação.png', bbox_inches='tight')
plt.figure(figsize=(30, 21), dpi=80)
plt.title('Precisão em função do número de exemplos')
plt.xlabel('Exemplos')
plt.ylabel('Precisão')
plt.grid(True)
for clf_name, clf_desc, line_color in classifiers:
plt.plot(example_axis, [y['precision'] for y in results[clf_name]], line_color, label = clf_desc)
plt.axis([example_axis[0], example_axis[-1], 0, 1])
plt.legend(loc='center left', bbox_to_anchor=(1, 0.5))
plt.savefig('resultadosRevocação.png', bbox_inches='tight')