/
multilabel_classification.py
151 lines (123 loc) · 5.21 KB
/
multilabel_classification.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
import matplotlib
matplotlib.use('Agg')
from preprocessor import preprocessor
from keras_models import piczak_CNN_multi
from keras.callbacks import TensorBoard, EarlyStopping
from sklearn import metrics
from keras.models import load_model
import numpy as np
import pandas as pd
import utils
import sklearn.metrics
from sklearn.metrics import f1_score, hamming_loss, zero_one_loss
import matplotlib.pyplot as plt
from pandas_confusion import ConfusionMatrix
import keras.backend as K
import tensorflow as tf
classes = ['ac', 'ch', 'cp', 'db', 'd', 'ei', 'gs', 'j', 's', 'sm',
'ac+ch', 'ac+cp', 'ac+db',
'ac+d', 'ac+ei', 'ac+gs',
'ac+j', 'ac+s', 'ac+sm',
'ch+cp',
'ch+db', 'ch+d', 'ch+ei', 'ch+gs',
'ch+j',
'ch+s', 'ch+sm', 'cp+db', 'cp+d',
'cp+ei', 'cp+gs', 'cp+j',
'cp+s', 'cp+sm', 'db+d', 'db+ei',
'db+gs', 'db+j', 'db+s', 'db+sm',
'd+ei', 'd+gs', 'd+j', 'd+s',
'd+sm', 'ei+gs', 'ei+j', 'ei+s',
'ei+sm', 'gs+j', 'gs+s', 'gs+sm',
'j+s', 'j+sm', 's+sm', 'xxx']
def evaluateSavedModel():
# Choose saved model to evaluate
model = load_model("saved/Piczak_CNN_pretrainsingle_trainmulti.h5")
# Setup preprocessor for loading extracted features
pp = preprocessor(parent_dir='../UrbanSound8K/audio')
# extracted features that should be loaded to calculate mean and std values
train_dirs = ["audio_overlap/folder1_overlap", "audio_overlap/folder3_overlap",
"audio_overlap/folder4_overlap", "audio_overlap/folder5_overlap", "audio_overlap/folder6_overlap",
"audio_overlap/folder7_overlap", "audio_overlap/folder8_overlap", "audio_overlap/folder9_overlap",
"audio_overlap/folder10_overlap"]
#pp.data_prep(train_dirs=[], test_fold="fold2", load_path="../UrbanSound8K/audio/extracted_short_60/")
# Load features
test_folder = "fold2"
pp.load_extracted_fts_lbs(train_dirs=train_dirs, test_fold=test_folder)
tb = TensorBoard(log_dir='./TensorBoard/piczak_CNN_singlelabel_pretrain_continue_multilabel')
# model.fit(pp.train_x, pp.train_y,validation_split=.1, epochs=25,
# batch_size=256, verbose=2, callbacks=[tb])
print("model evaluation")
scores = model.evaluate(pp.test_x, pp.test_y, verbose=2)
print("loss: {0}, test-acc: {1}".format(scores[0], scores[1]))
# Make predictions
preds = model.predict(pp.test_x)
# Evaluate predictions
evaluateModel(pp, preds, test_folder)
def evaluateModel(pp, preds, fold):
# ************ PROCESSING THE PREDICTIONS
preds[preds >= 0.5] = 1
preds[preds < 0.5] = 0
print("F1 SCORE:")
print(f1_score(pp.test_y, preds, average=None))
print("Hamming Loss:")
print(hamming_loss(pp.test_y, preds))
print("Zero-one loss:")
print(zero_one_loss(pp.test_y, preds))
# I reach here in plus_one_hot_encode, I want to transform it in one hot
y_test_preds = utils.from_plus_to_one_hot(np.array(pp.test_y))
preds_transf = utils.from_plus_to_one_hot(np.array(preds))
cm = ConfusionMatrix(np.array(y_test_preds).argmax(1), np.array(preds_transf).argmax(1))
print(cm)
ax = cm.plot()
ax.set_xticklabels(classes, rotation="vertical")
ax.set_yticklabels(classes)
plt.savefig("cmpre{0}.png".format(fold))
def piczac_cross_validation(epochs, load_path):
train_dirs = []
n_folders = 10
for i in range(1, n_folders + 1):
#train_dirs.append('fold{0}'.format(i))
train_dirs.append('folder{0}_overlap'.format(i))
print(train_dirs)
for fold in ((10, 1), (1, 2), (2, 3), (3, 4), (4, 5), (5, 6), (6, 7), (7, 8), (8, 9), (9, 10)):
val_fold = 'folder{0}_overlap'.format(fold[0])
test_fold = 'folder{0}_overlap'.format(fold[1])
#val_fold = 'fold{0}'.format(fold[0])
#test_fold = 'fold{0}'.format(fold[1])
train_dirs.remove(val_fold)
train_dirs.remove(test_fold)
pp = preprocessor(parent_dir='../../data/UrbanSound8K/audio')
pp.load_extracted_fts_lbs(train_dirs=train_dirs, val_fold=val_fold, test_fold=test_fold, load_path=load_path)
model = piczak_CNN_multi(input_dim=pp.train_x[0].shape, output_dim=pp.train_y.shape[1])
print("done")
print("OPTIMIZER")
#print(model.optimizer.lr)
#K.set_value(model.optimizer.lr, 0.002)
#model.optimizer.lr.set_value(0.0001)
#model.save('Models/model1_all_p2_bn{0}.h5'.format(str(fold)))
#model = load_model('Models/model1_all_p2{0}.h5'.format(str(fold)))
#model = load_model('Models/model1_all_p2_bnsec_overlap_{0}.h5'.format(str(fold)))
tb = TensorBoard(log_dir='./TensorBoard/' + 'overlap_run{0}'.format(fold[1]))
es = EarlyStopping(patience=10, verbose=1)
model.fit(pp.train_x, pp.train_y, validation_data=[pp.val_x, pp.val_y], epochs=epochs,
batch_size=1000, verbose=2, callbacks=[tb, es])
#model.save('Models/model1_all_p2_bnsec_overlap_9010_{0}.h5'.format(str(fold)))
preds = model.predict(pp.test_x)
evaluateModel(pp,preds, fold)
K.clear_session()
train_dirs.append(val_fold)
train_dirs.append(test_fold)
# val_fold = 'fold' + str(folds[0])
# test_fold= 'fold' + str(folds[1])
if __name__ == '__main__':
K.clear_session()
tf.reset_default_graph()
print("main")
utils.classes_number_mapper()
print(utils.dic_tot)
for i in sorted(utils.dic_tot.keys()):
print(i, utils.dic_tot[i])
# train_keras_cnn()
#evaluateSavedModel()
piczac_cross_validation(10, "../UrbanSound8K/audio/extracted_overlapping_50/audio_overlap")
#piczac_cross_validation(125, "../../feat_overlap_diff")