-
Notifications
You must be signed in to change notification settings - Fork 0
/
pd3.py
192 lines (164 loc) · 8 KB
/
pd3.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
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import keras
from sklearn.model_selection import KFold
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import accuracy_score, confusion_matrix, f1_score, precision_score, recall_score
from sklearn.svm import SVC
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
import xgboost as xgb
from operator import itemgetter
def load_data(file_name):
df = pd.read_csv(file_name)
y = pd.Series(df.iloc[1:, -1].values, name=df.iloc[0, -1]).astype(np.int8)
patient_id = pd.Series(df.iloc[1:, 0].values,
name=df.iloc[0, 0]).astype(np.int16)
gender = pd.Series(df.iloc[1:, 1].values,
name=df.iloc[0, 1]).astype(np.int8)
baseline_feats = pd.DataFrame(
df.iloc[1:, 2:23].values, columns=df.iloc[0, 2:23]).astype(np.float64)
intensity_feats = pd.DataFrame(
df.iloc[1:, 23:26].values, columns=df.iloc[0, 23:26]).astype(np.float64)
format_feats = pd.DataFrame(
df.iloc[1:, 26:30].values, columns=df.iloc[0, 26:30]).astype(np.float64)
bandwidth_feats = pd.DataFrame(
df.iloc[1:, 30:34].values, columns=df.iloc[0, 30:34]).astype(np.float64)
vocal_feats = pd.DataFrame(
df.iloc[1:, 34:56].values, columns=df.iloc[0, 34:56]).astype(np.float64)
mfcc_feats = pd.DataFrame(
df.iloc[1:, 56:140].values, columns=df.iloc[0, 56:140]).astype(np.float64)
wavelet_feats = pd.DataFrame(
df.iloc[1:, 140:322].values, columns=df.iloc[0, 140:322]).astype(np.float64)
tqwt_feats = pd.DataFrame(
df.iloc[1:, 322:-1].values, columns=df.iloc[0, 322:-1]).astype(np.float64)
return {"patientId": patient_id,
"gender": gender,
"baselineFeats": baseline_feats,
"intensityFeats": intensity_feats,
"formantFeats": format_feats,
"bandwidthFeats": bandwidth_feats,
"vocalFeats": vocal_feats,
"mfccFeats": mfcc_feats,
"waveletFeats": wavelet_feats,
"tqwtFeats": tqwt_feats,
"label": y}
def convert_data(data, features):
if len(features) == 1:
return data[features[0]]
return pd.concat(itemgetter(*features)(data), axis=1).values
def build_block(input, size, activation, batch_norm, dropout):
layer = keras.layers.Dense(
size, activation=activation, input_dim=input.shape[1])(input)
if batch_norm:
layer = keras.layers.BatchNormalization()(layer)
if dropout:
layer = keras.layers.Dropout(0.1)(layer)
return layer
def build_autoencoder(input_shape, structure, activations, batch_norm=False, dropout=False, optimizer="sgd"):
input = keras.layers.Input(shape=input_shape)
layer = build_block(
input, structure[0], activations[0], batch_norm, dropout)
for i in range(1, len(structure)):
layer = build_block(
layer, structure[i], activations[i], batch_norm, dropout)
model = keras.models.Model(inputs=input, outputs=layer)
model.compile(loss="mean_squared_error", optimizer=optimizer)
model.summary()
return model
def run_model(X, y, autoencoder_params, encoder_params, classifier=None, classifier_params=None):
encoder_size = len(autoencoder_params['structure']) // 2
folds = KFold(n_splits=5, shuffle=True)
scores = {'accuracy': [], 'fscore': [], 'precision': [], 'recall': []}
conf_mats = []
for train_idx, test_idx in folds.split(X):
x_train, x_test = X[train_idx], X[test_idx]
y_train, y_test = y[train_idx], y[test_idx]
scaler = MinMaxScaler()
x_train = scaler.fit_transform(x_train)
x_test = scaler.transform(x_test)
autoencoder = build_autoencoder(**autoencoder_params)
history = autoencoder.fit(x_train, x_train, epochs=encoder_params.get(
'epochs'), batch_size=encoder_params.get('batch_size'), validation_data=(x_test, x_test))
# plot_loss(history, encoder_params.get('epochs'), "autoencoder")
encoder = autoencoder.layers[:encoder_size]
for layer in encoder:
layer.trainable = False
if classifier is None:
model = keras.models.Sequential(
layers=[layer for layer in encoder])
layer = keras.layers.Dense(1, activation='sigmoid')
model.add(layer)
model.compile(loss=keras.losses.binary_crossentropy,
optimizer=classifier_params['optimizer'], metrics=["accuracy"])
history = model.fit(x_train, y_train, batch_size=classifier_params.get(
'batch_size'), epochs=classifier_params.get("epochs"), validation_data=(x_test, y_test))
# plot_loss(history, classifier_params.get('epochs'), "classifier1")
if classifier_params.get("rerun") == True:
for layer in model.layers[:encoder_size]:
layer.trainable = True
model.compile(loss=keras.losses.binary_crossentropy,
optimizer=classifier_params.get("optimizer"), metrics=["accuracy"])
history = model.fit(x_train, y_train, batch_size=classifier_params.get(
'batch_size'), epochs=classifier_params.get("epochs"), validation_data=(x_test, y_test))
# plot_loss(history, classifier_params.get(
# 'epochs'), "classifier2")
y_pred = np.round(model.predict(x_test))
else:
model = keras.models.Sequential(
layers=[layer for layer in encoder])
new_train = model.predict(x_train)
new_test = model.predict(x_test)
model = classifier(**classifier_params)
model.fit(new_train, y_train)
y_pred = np.round(model.predict(new_test))
scores['accuracy'].append(accuracy_score(y_test, y_pred))
scores['fscore'].append(f1_score(y_test, y_pred))
scores['precision'].append(precision_score(y_test, y_pred))
scores['recall'].append(recall_score(y_test, y_pred))
conf_mats.append(confusion_matrix(y_test, y_pred))
return scores, conf_mats
def plot_loss(model, epochs, context):
loss = model.history['loss']
val_loss = model.history['val_loss']
acc = model.history.get('accuracy')
val_acc = model.history.get('val_accuracy')
epochs = range(epochs)
plt.figure()
plt.plot(epochs, loss, 'b', label=f'Training loss-{context}')
plt.title('Training loss')
plt.plot(epochs, val_loss, 'r', label=f'Validation loss-{context}')
plt.legend()
plt.show()
if acc is not None:
plt.figure()
plt.plot(epochs, acc, 'b', label=f'Training accuracy-{context}')
plt.plot(epochs, val_acc, 'r', label=f'Validation accuracy-{context}')
plt.legend()
plt.show()
def __main__():
data = load_data('pd_speech_features.csv')
y = data['label']
features = ["gender", "baselineFeats", "intensityFeats", "formantFeats",
"bandwidthFeats", "vocalFeats", "mfccFeats", "waveletFeats", "tqwtFeats"]
X = convert_data(data, features)
params = {'input_shape': (X.shape[1],),
'structure': [600, 300, 100, 30, 100, 300, 600, X.shape[1]],
'activations': ['relu'] * 8,
'optimizer': keras.optimizers.Adam(lr=0.001),
'batch_norm': True}
print("################################")
encoder_params = {"epochs": 250, "batch_size": 256}
classifier_params = {'optimizer': keras.optimizers.Adam(
lr=0.0001), 'epochs': 200, 'rerun': False} # fine tunes if rerun=True
scores, confs = run_model(X, y.values, params, encoder_params=encoder_params,
classifier=None, classifier_params=classifier_params)
print("ACCURACY:", np.mean(scores['accuracy']))
print("F SCORE:", np.mean(scores['fscore']))
print("PRECISION:", np.mean(scores['precision']))
print("RECALL:", np.mean(scores['recall']))
print("CONFUSION MATRICES:")
for i in range(len(confs)):
print(confs[i])
__main__()